Cycles, Growth and Structural Change
Cycles, Growth and Structural Change
Cycles, Growth and Structural Change
With its unique combination of theoretical models and empirical analyses, this
collection of specialists’ essays contributes to an assessment of our understanding
of the relationship between cycles, growth and structural change. In this light it
also offers a critical evaluation of results in the recent literature on complex and
chaotic dynamics. The book is constructed upon the hypothesis that a re-unification
of the economic analyses of dynamic phenomena can be obtained through
comparison and cross fertilization of their up-to-now specialized approaches.
Recent developments in economic dynamics, one of the exciting frontiers in
current economic research, have been drawing from very diverse research areas
and different specialisms: economic theorists, applied economists and
econometricians, of course, but also mathematicians and researchers in information
theory, time series and dynamical system theories. The variety of approaches in
this book well represents the economists’ views of this research area. They range
from neoclassical and endogenous growth theories, to classical, evolutionary and
neo-Austrian dynamics; from real business cycle and information-based
econometrics to other non-linear techniques related to complex dynamics.
Likewise, modelling is done in the macro-, micro-, multi-agent and large dynamical
system styles. Moreover, the book offers materials for an evaluation of recent
literature which relies on complexity properties to search for alternative
formulations of economic policy issues.
Providing both historical evidence and theoretical formulations, written in a
critical vein by leading specialists in their own areas, Cycles, Growth and Structural
Change offers state-of-the art surveys of a fast expanding field and suggests a
research agenda for the future.
Macroeconomics
A survey of research strategies
Edited by Alessandro Vercelli and Nicola Dimitri
Oxford University Press, 1992
Edited by
Lionello F. Punzo
Contents
List of figures ix
List of tables xi
List of contributor xiii
Preface xvii
PART I
Facts and interpretations of growth and fluctuations 1
PART II
The macroeconomy and its dynamics 93
PART III
Dynamics by interaction 201
PART IV
Challenges for quantitative methodologies 281
Index 367
List of figures vii
Figures
Tables
Contributors
Preface
This volume presents contributions revolving around the general theme of the
relationship, empirical as well as theoretical, between economic oscillations,
growth and structural change. It is constructed upon a twofold hypothesis. A re-
unification of the analyses of the three phenomena may be obtained through a
comparison and cross fertilisation of the existing specialised approaches. Moreover,
the accomplishment of such a project is perhaps within reach, now. This preface
will discuss such a hypothesis, outlying the structure of the book and the string
connecting individual contributions.
Most chapters are revised versions of papers read at the International Summer
School of the same title, held in Siena in July 1998. Some chapters have been
added with the intent of broadening the view without however any attempt to
reach exhaustiveness. The field, in fact, represents one of the most exciting frontiers
in current economic research, calling for the individual as well as, more often,
joint efforts of specialists from very diverse areas: economic theorists, real life
applied economists, econometricians, of course, but also mathematicians and people
working in information-theory related areas. Cross fertilisation is at one of its
historical peaks, and it is taking place here more than elsewhere in the economic
discipline. The School was concluded by a lively workshop dedicated to the works
of the late Richard M. Goodwin, whose lifetime research has been dedicated to
the themes of this book.
Parts I and II
The interaction between economic fluctuation, growth and structural change is
the main thrust in Part I which is dedicated to a survey of facts and interpretations.
The strength of such an interaction is the fundamental motivation for Solomou’s
rejection of the ‘methodology of stylised facts’ searching for repetitive patterns
and ‘dynamical laws’ of one sort or another. In an essay rich in both historical
reconstruction and methodological reflections, the author highlights the
complications involved: the varying patterns of causality/interdependence relations
between the key factors of shorter term fluctuations, and the interlocking of
oscillations at various frequencies with structural change. He points out that,
contrary to common beliefs, the post-war experience of growth in the developed
countries is fundamentally a unique string in a long time series of events. Thus,
the author promotes a historical perspective as an alternative the methodology of
stylised facts which yields anyway a rather unstable and thus interpretation-
unreliable set.
Yoshikawa focuses upon the exemplary intermingling of growth and structural
change in the post-war history of Japan. He criticizes in particular various existing
Preface xxi
interpretations of the rapid growth era, to propose a two-phase model whereby an
early (1950–60) modernisation process of a Lewis-type dualistic economy is
followed by a dramatic break and re-orientation. The change in household structure
and urbanisation, driving labour force away from traditional agricultural
employment and creating new urban demand, generated the unprecedented
expansion of the earlier phase. This yielded, at the end of the era, to market
saturation creating the need for a new, externally driven expansion. Domestic
demand was substituted by foreign demand as an engine of growth. The chapter
contains also interesting remarks on the prospects of NICS of various generations
and on the experience of countries like Italy sharing with Japan a dualistic structure.
In the chapter by Böhm and Punzo a new framework is introduced where growth
is naturally associated with fluctuations and structural change is defined as a
qualitative change in the growth model, thus spanning a dynamic menu that is
richer than expected. The chapter reviews in this light empirical findings for a set
of European countries, the USA and Japan. It is shown that these went through
repeated structural changes, but also that growth models were different across
countries and were strung differently in each country’s history. Observed high
irregularity and cross-country variability reflect shock responses, of course, but
more deeply they reveal the workings of the countries’ own structures. Looking
for changes in the growth model is one way of capturing structural change, and it
seems to be the most natural one in the formalised setting in the dynamical systems
style: it is a regime shift. The notion of regime dynamics translates naturally into
that of Day’s multi-phase dynamics, whose application to very long run growth is
well illustrated in the chapter jointly written with Pavlov (opening Part II). Their
Generalised Evolutionary Model (GEM) focuses upon the evolution of economic
and social macrostructures through phases of growth following sometimes an
irregular, by no means determined sequence, and it accounts for such an evolution
through the working of an internal instability mechanim. The GEM is calibrated
to reproduce (in the simulation sense) what is known of the history of mankind,
an exercise which illustrates an approach to qualitative analysis aptly termed
qualitative econometrics.
In fact, the output of a dynamic model exhibits scenarios, artificial histories in
other words. In general, what matters in these simulation exercises cannot really
be the quantitative coherence or closeness of the artificial to the actual time series,
so that debate goes on about their uses. One proposal comes from the real business
cycle approach, illustrated in the chapter by Prescott, where some quantitative
criterion is retained to assess the model at hand against evidence. As an alternative,
following Day and Pavlov, artificial histories can be treated as qualitative
descriptions. Their worth is in their capability of reproducing the shifting across
phases (or regimes) following a given known pattern. Thus, the model may explain
to a certain extent features of actual macroeconomic behaviour; adding some
extrapolation exercises, we get to qualitative econometrics.
Clearly, an economy whose history exhibits repeated regime or phase switches
can hardly be described by an equilibrium technique. The analysis of scenarios of
out-of-equilibrium dynamics is the centre of the neo-Austrian approach proposed
by Amendola and Gaffard, highlighting the conditions for the emergence of co-
xxii Preface
ordination problems in the decision processes as well as within the production
processes. Thus, stability towards a given state or attractor (that is the core of the
conventional dynamical modelling) is no longer the key issue and, actually, it
turns out to be not even definable as a property, given the continually shifting
nature of the system’s theoretical equilibrium. Instead, it is the issue of the
sustainability of the dynamic process itself, i.e. the viability of the growth path,
that comes to the fore redefining the very contours of economic policy.
The two chapters concluding Part II review and contribute novel insights into
the out-of-equilibrium approach. Flaschel’s contribution offers a disequilibrium
framework where a monetary economy works according to rules and laws set out
by a composite model built upon the seminal contributions of Keynes, Marx and
Goodwin. This economy’s dynamic outcome shows a strong bias towards
fluctuation rather than steady growth, Flaschel building upon the alternative
tradition that looks at the market economy through the pessimistic eyes of the
classical thinker.
Iwai tackles the time-honoured issue of demonstrating the possibility of
reconciling the notion of long run with the notion of disequilibrium. The test
ground for his effort is a classical issue, the possibility of a long run state where
profits are above their normal levels. The effort proves successful with the re-
definition in statistical terms of the very notion of long run equilibrium, which
now allows for a cross-firm distribution of technologies, as it results from the
continuously on-going process of innovation. Baumol’s contribution, on the other
hand, looks at the long run growth aspects of that same process and offers an in-
depth analysis of some of its microeconomic features. He explains why the evidence
of unprecedented growth record of the market economy contradicts the poor
performance predicted by traditional growth theory. The key argument is centred
on the trade-off between flows of inventions and distribution of benefits, and
therefore on the novel observation that positive spillovers (linked with non-
appropriability) do not necessarily impede the innovation process. As, on the
contrary, a certain amount of spillovers does fuel performance, the author gives a
positive answer to the question previously raised, whether interaction can generate
growth and at the same time fluctuations. This vindicates Schumpeter’s (and
Goodwin’s) viewpoint.
Acknowledgement
I wish to thank for their comments (with the usual caveats) J.-L. Gaffard and
M. Puchet Anyul. Research for this preface and work in the final stages in the
production of this book were carried on as part of a project on ‘Recent developments
in qualitative analysis and object-oriented simulation in economic dynamics’,
financed by the Italian MURST as Progetto di ricerca scientifica di rilevante
interesse nazionale, 2000.
References
Goodwin R.M. (1982), Essays in Economic Dynamics, London: Macmillan
Goodwin R.M. (1983), Linear Economic Structures, London: Macmillan
Goodwin, R.M. (1990), Chaotic Economic Dynamics, Oxford: Oxford University Press
Goodwin R.M. and L.F. Punzo (1987), The Dynamics of a Capitalist Economy, Cambridge:
Polity Press, and Boulding, CO: Westview
Coricelli F., M. Di Matteo and F.H. Hahn (eds) (1998), New theories in growth and
development, Basingstoke; New York: Macmillan
Part I
1.1 Introduction
Current research on business cycles has focused on the post-war period. The aim
has been to derive the main ‘stylised facts’ of fluctuations. The approach here is
very different. The emphasis is on the need for an historical perspective to business
cycles. History inevitably provides us with a broader empirical basis that will allow
us to formulate more general theoretical and empirical questions. It is argued that to
focus on explaining the stylised facts of any one historical period to gain a general
insight on economic fluctuations will lead to serious errors of interpretation.
An historical perspective is important not because history repeats itself but
because history illustrates the evolutionary nature of business cycle behaviour and
gives us an understanding of the factors that generate change. In economic systems
that entail behavioural, institutional, structural and policy changes we can safely
predict that the nature of business cycles will not be stable over time. The case
for an historical perspective is founded on two important empirical features. First,
the process of economic growth of modern economies has undergone massive
structural change in the past two hundred years. One important aspect of struc-
tural change is the reallocation of production across sectors. For example, with
industrialisation we would expect a change in the role of agricultural cycles in
macroeconomic fluctuations. Given that these type of changes take place over a
very low frequency, we need to observe business cycles over the long-term to gain
insights on the resulting cyclical effects of structural change.
Second, policy regime changes have an observable effect on cyclical fluctua-
tions. Business cycles during the rules-driven policy framework of the pre-1913
gold standard epoch were very different to those of the inter-war period. Similarly,
the fluctuations of the Bretton Woods era were very different to those observed in
the post-1973 era. Major policy regime changes have been few in number, once
again, implying the need for a long-run perspective.
These observations raise serious doubts about the empirical relevance of the
so-called ‘stylised facts’ approach to business cycles. This approach assumes that
regularities exist over time and across countries (Lucas, 1981). However, many
of these empirical regularities are usually derived from studies of very short time
4 Solomos Solomou
periods (often the post-1960 era) and a small selection of countries (usually the
UK and the USA).
Even if we focus on a very limited set of empirical features we observe im-
portant changes. The average cyclical duration has changed over time. During
1870–1913 a number of variables (including aggregate investment, agricultural
production and construction sector output) fluctuated with a long swing duration
averaging about 20 years. During the inter-war period shorter fluctuations were
observed. During the post-war ‘golden age’ the average cyclical period fell to 5
years. During the post-1973 era the average duration has once again lengthened,
averaging approximately 10 years since the late 1970s. Cyclical amplitudes have
also varied significantly over time. Low macroeconomic volatility during 1870–
1913 gave way to high amplitude fluctuations during 1919–38; the stability of
the post-war ‘golden age’ has been followed by the relatively more volatile post-
1973 era. Patterns of co-movement of key variables have not been stable over time.
Much of post-war research on business cycles has noted that prices and output have
fluctuated contra-cyclically and has proceeded to explain this feature in terms of
‘real’ business cycle theory (Danthine and Donaldson, 1993). However, during
the classical gold standard the relationship was not stable and during the inter-
war period price and output fluctuations moved in a pro-cyclical manner (Cooley
and Ohanian, 1991). Assuming universal stylised facts is not a realistic way of
understanding business cycles.
This chapter surveys business cycle features across three historical epochs to
illustrate some of the themes discussed in this introduction. Section 1 considers the
pre-1913 epoch. Section 2 considers the changes observed in the inter-war period.
Section 3 considers the post-war epoch.
0.06
0.04
Deviations from Trend
0.02
0
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
19
19
19
19
19
19
19
-0.02
-0.04
-0.06
0.04
0.03
0.02
Deviations from Trend
0.01
0
1870
1872
1874
1876
1878
1880
1882
1884
1886
1888
1890
1892
1894
1896
1898
1900
1902
1904
1906
1908
1910
1912
-0.01
-0.02
-0.03
0.03
0.02
Deviations from Trend
0.01
0
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
19
19
19
19
19
19
19
-0.01
-0.02
-0.03
-0.04
0.04
0.03
0.02
Deviations from Trend
0.01
0
70
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
18
18
18
18
18
18
18
18
18
18
18
18
18
18
18
19
19
19
19
19
19
19
-0.01
-0.02
-0.03
-0.04
108
106
104
102
1913=100
100
98
96
94
92
90
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
France UK Germany USA
USA 0.01 –
(0.14)
Germany 0.03 –0.40 –
(0.12) (0.13)*
Japan 0.08 0.22 0.14
(0.17) (0.15) (0.17) –
The entries show the contemporaneous correlation of cyclical variations of output. Numbers in
parenthesis are standard errors: only the German–American correlation is significantly different from
zero. * denotes significant at the 5 per cent level.
Source: Backus and Kehoe, 1992, p. 876.
back research in this area. For example, Jevons (1884, p.235) argued for a sunspot
theory of the trade cycle. In general, modern economists dismiss such theories as
unconvincing and misleading, although they continue to find favour in some of the
literature (Zahorchak, 1983).
The impact of weather shocks will vary from sector to sector as the conditions
favouring one activity may be adverse to another. The extent of such dampening
will also depend on changes in the sectoral structure of the economy over time. The
largest weather-sensitive sectors during this period were agriculture, construction
and energy demand. In Britain, these three sectors accounted for approximately 25
per cent of GDP in the 1870s, falling to 17 per cent of GDP in the early twentieth
century. In Germany and France these ratios were approximately 50 per cent of
GDP in 1870, falling to 30–40 per cent by the early twentieth Century.
Solomou and Wu (2000) show that weather fluctuations had large effects on
all these sectors. Using data from Britain, the effect range of weather shocks on
the growth rate of construction and coal sectors was around ±0.05 during the
period 1870–1913. A similar range is observed for the agricultural sector during
1880–1913, although this was significantly wider during the 1870s. The aggregate
effect of these sector-specific effects depends on two features: first, the pattern of
covariance of effects across different sectors and second, the relative weight of the
different sectors in the macroeconomy and changes in these shares over time.
Consider the pattern of correlation of weather effects on different sectors (Fig-
ure 1.3 plots sector-specific weather effects). A significant positive correlation is
observed between agriculture and construction sector effects (r=0.51)2 . There is
no evidence that weather shocks were having a neutral effect on the macroecon-
omy because adverse shocks to one sector were consistently being compensated
by favourable effects to other sectors.
In order to evaluate the sum of all these sectoral effects the magnitude of
the sectoral shocks are aggregated using GDP shares as weights. Figure 1.4 plots
the shares of the three sectors in GDP. The agricultural sector is dominant in
determining the weather effect on the economy. This arises because of the relative
12 Solomos Solomou
0.100
0.050
0.000
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
18
18
18
18
18
18
18
18
18
18
18
18
18
18
19
19
19
19
19
19
19
-0.050
-0.100
-0.150
size of the agricultural sector over this period, a pattern of positive co-movement
between the weather effects affecting agriculture and construction and the wider
effect range of weather shocks to agriculture. Figure 1.5 illustrates this result with
a plot of the aggregate weighted weather effect and the weighted sectoral effects.
It is clear that the total weather effect on the economy is dominated by the effect
from agriculture and the co-movement with the weather effect on construction.
The effect from coal simply adds some randomness to the magnitude of the total
effect.
Aggregating sectoral weather effects using GDP shares as weights, suggests
that the impact of weather shocks remained large throughout this period. The effect
range of the sum of the sectoral shocks to GDP ranged from +0. 7 per cent of GDP
to −1. 5 per cent of GDP. These estimates represent the lower bounds given the
economic structure of other European countries compared to Britain. Moreover,
because weather shocks are autocorrelated, we can safely conclude that weather
added significant direct cyclicality to the economy. To dismiss the role of weather
(even as late as the post-1870 period) in business cycles is ahistorical.
Summarising, economic cycles during the pre-1913 period are best depicted
by multiple cycles. The causal structure behind such a process is likely to be
complex. A number of adjustment processes were made possible by the nature of
the world economy during this period. Free capital and labour movements across
national boundaries resulted in long cycles in both these variables. The economic
structure of the period also resulted in a coupling of weather cycles and sectoral
fluctuations. At the macro level the relationship was not a simple coupling of one
cycle with another. Instead, structural change acted as an important filter. Such
structural change meant that agricultural fluctuations had a declining effect on the
macroeconomy during the pre-1913 period, with the largest effects being observed
0.160
0.140
0.120
0.100
Per Cent
0.080
0.060
0.040
0.020
0.000
1872
1874
1876
1878
1880
1882
1884
1886
1888
1890
1892
1894
1896
1898
1900
1902
1904
1906
1908
1910
1912
construction coal agriculture
0.010
0.005
0.000
72
74
76
78
80
82
84
86
88
90
92
94
96
98
00
02
04
06
08
10
12
18
18
18
18
18
18
18
18
18
18
18
18
18
18
19
19
19
19
19
19
19
Effect on GDP
-0.005
-0.010
-0.015
-0.020
Macroeconomic trend-stationarity
During the pre-1913 gold standard era at least some of the major industrial countries
displayed trend-stationarity in their aggregate growth paths, with a tendency to
grow along a stable path in the long run (Mills, 1991; Solomou, 1998). This is the
case for Britain and America over the period 1870–1913 and for Germany during
1880–1913. During the inter-war era these underlying growth paths were displaced
in a persistently downward direction as a result of adverse business cycle shocks
during the early 1920s and early 1930s. The timing of these adverse shocks varied
by country: for example, Germany and Britain experienced persistent adverse
effects in the early 1920s while America saw persistence arising from the adverse
shocks of the early 1930s.
International cycles
Business cycles showed high levels of international co-movement in the inter-war
era relative to the pre-1913 era (Backus and Kehoe, 1992; Eichengreen, 1994).
This high level of international co-movement stands out even when comparisons
are made with the post-war era.
Economic cycles since 1870 15
These new descriptive features raise a number of important questions: first,
is the persistence of shocks influenced by the collapse of the rules-driven pol-
icy regime of the pre-1913 classical Gold standard, and its replacement by more
discretionary policy regimes in the 1920s and 1930s? Second, why did the in-
ternational adjustments of the long swing process (such as the inverse home and
overseas investment swings) come to an end during the inter-war period, and what
implications did this have for economic cycles? Finally, how do we account for
the internationalisation of the business cycle during the inter-war period? These
questions are addressed below.
movements were observed between the UK and the USA, Japan and the USA,
Canada and the USA, Germany and the UK and Canada and the UK. Insignificant
linkages are observed between Japan and the UK, Japan and Germany and Japan
and Canada. Such strong bilateral cyclical linkages are in marked contrast to the
inter-war era when the business cycle is far more of an international phenomenon
(Backus and Kehoe, 1992).
Using less formal statistical methods Zarnovitz (1981) compared the business
cycle timings of the US, Canada, the UK, Germany and Japan within the National
Bureau reference cycle framework during the period 1948–80. He also finds strong
co-movements between different sets of countries, such as the US and Canada and
the UK and Germany, rather than an international business cycle that affects all
the major countries simultaneously.
Using quarterly output data for a larger cross-section of countries for the pe-
riod 1959–89 Danthine and Donaldson (1993) document both bilateral and inter-
national business cycle linkages. At the bilateral level there are some very strong
cyclical co-variations. For example, Germany shows significant co-movements
with Austria, France, Italy, Switzerland and the UK; the UK has strong bilateral
linkages with France and Germany. Aggregating the EC countries into one bloc
we can also compare how each individual EC member varies with the EC bloc.
The co-movements are very much weaker, suggesting that it is strong bilateral
ties that are important in the intra-European cyclical linkages, not bloc behaviour.
Economic cycles since 1870 21
Comparing the individual EC members with the USA, their linkages have been far
stronger than with the EC bloc. The data also allow us to compare inter-bloc cycli-
cal influences: comparing the cyclical variations of the EC, the USA and Japan
we see strong positive co-movements across these three major economic zones.
At this level of aggregation there is some evidence of international business cycle
linkages.
Notes
1 Given the stochastic trends of nominal effective exchange rates (EER) we investigated
the existence of a common stochastic trend using the Johansen co-integration frame-
work. The nominal EER movements of the core countries are found to be co-integrated
over this period.
2 The correlations between construction and coal and agriculture and coal are −0. 14
and 0.001 respectively
3 One aspect of inter-war international capital flows, which may have destabilised the
world economy, is the link between capital flows and the ‘recycling’ of reparation
payments (Solomou, 1998).
4 Commodity production is a composite index of industrial and agricultural production.
5 Most countries were also influenced by a short cycle 5 to 10 years and a shorter
inventory cycle.
References
Abramovitz, M. (1968) ‘The Passing of the Kuznets Cycle’, Economica, XXXV, 349–67.
Aldcroft, D.H. and Fearon, P. (eds) (1972) British Economic Fluctuations, 1790–1939,
London.
Backus, D.K. and Kehoe, P.J. (1992) ‘International Evidence on the Historical Properties
of Business Cycles’, American Economic Review, 82, (4), 864–88.
Balke, N.S. and Gordon, R.J. (1989) ‘The Estimation of Prewar Gross National Product:
Methodology and New Evidence’, Journal of Political Economy, 97, 1, 38–92.
Bayoumi, T. and Eichengreen, B. (1996) ‘The Gold Standard and the International Monetary
System’, in T. Bayoumi, B. Eichengreen and M.P. Taylor Modern Perspectives on the
Gold Standard, Cambridge: Cambridge University Press.
Bernanke, B.S. (1983): ‘Nonmonetary Effects of the Financial Crisis In the Propagation of
the Great Depression’, American Economic Review, 73, 257–76.
24 Solomos Solomou
Bernake, B.S. and James, H. (1991) ‘The Gold Standard, Deflation and Financial Crisis in
the Great Depression: An International Comparison’, In R.G. Hubbard (ed.), Financial
Markets and Financial Crises, Chicago: University of Chicago Press.
Blanchard, O. (1993): ‘Consumption and the Recession of 1990–1991’, American Economic
Review, 83, 2, 270–4.
Bordo, M. D. and H. Rockoff (1996) ‘The Gold Standard as a Good Housekeeping Seal of
Approval, Journal of Economic History, 56, (2), 389–426.
Burns, A.F. (1960) ‘Progress Towards Economic Stability’, American Economic Review,
50, March, 1–19.
Cairncross, A.K. (1953) Home and Foreign Investment, 1870–1913: Studies In Capital
Accumulation, Cambridge: Cambridge University Press .
Capie, F. and Mills, T.C. (1991) ‘Money and Business Cycles in the US and U.K., 1870 To
1913’, Manchester School, 53, (Supplement), 38–56.
Capie, F. and Mills, T.C. (1992) ‘Money and Business Cycles in the United States 1870–
1913: A Re-examination of Friedman and Schwartz’ Explorations in Economic History
29, 3, 251–73.
Cogley, T. and Nason, J.M. (1995) ‘Effects of The Hodrick and Prescott Filter on Trend and
Difference Stationary Time Series: Implications for Business Cycle Research’, Journal
of Economic Dynamics and Control, 19, 253–78.
Catao, L.A.V. and Solomou, S.N. (1993) ‘Business Cycles During the Gold Standard, 1870–
1913’, Department of Applied Economics Working Paper, 9304, Cambridge University.
Calomiris, C.W. (1993) ‘Financial Factors In the Great Depression’, Journal of Economic
Perspectives, 7, (2) 61–85.
Cooley, T.F. and Ohanian, L.E. (1991) ‘The Cyclical Behaviour of Prices’, Journal of
Monetary Economics, 28, 25–60.
Crafts, N.F.R., Leybourne, S.J. and Mills, T.C. (1989) ‘The Climacteric In Late Victorian
Britain and France:A Reappraisal of Evidence’, Journal of Applied Econometrics 4, (2),
103–18.
Crafts, N.F.R. and Mills T.C. (1992) ‘Economic Fluctuations, 1851–1913: A Perspective
Based On Growth Theory’, In S.N. Broadberry and N.F.R. Crafts (eds) Britain In The
World Economy 1870–1939, Cambridge.
Danthine, J.P. and J.B. Donaldson (1993) ‘Methodological and Empirical Issues In Real
Business Cycle Theory’, European Economic Review, 37, (1), 1–36.
Eichengreen, B.J. (1989) ‘The Political Economy of The Smoot Hawley Tariff’, Research
In Economic History, 11, 1–44.
Eichengreen, B.J. (1992) Golden Fetters: the Gold Standard and the Great Depression,
1919–1939, Oxford: Oxford University Press.
Eichengreen, B.J. (1994) ‘History of the International Monetary System: Implications For
Research In International Macroeconomics and Finance’, In F. Van Der Ploeg, F. The
Handbook of International Macroeconomics, Oxford.
Eichengreen, B.J. and Sachs, J. (1985) ‘Exchange Rates and Economic Recovery In the
1930s’, Journal of Economic History, 45, 925–46.
Ford, A.G. (1962) The Gold Standard 1880–1914: Britain and Argentina. Oxford: Claren-
don Press.
Hall, R.E. (1993) ‘Macro Theory and the Recession of 1990–1991’, American Economic
Review, 83, (2), 275–9.
Harvey, A.C. and Jaeger, A. (1993) ‘Detrending, Stylized Facts and the Business Cycle,
Journal of Applied Econometrics, 8, (3), 231–47.
Economic cycles since 1870 25
Hicks, J.R. (1982) ‘Are there Economic Cycles?’, In J.R. Hicks, Money, Interest and Wages:
Collected Essays On Economic Theory, Vol. II, Oxford.
Hillinger, C.(ed.) (1992) Cyclical Growth In Market and Planned Economies, Clarendon
Press, Oxford.
Jevons, W.S. (1884) Investigations in Currency and Finance, London: Macmillan.
Kindleberger, C.P. (1983) The World In Depression,1929–1939, Harmondsworth: Penguin.
Khatri, Y.J., Solomou, S.N. and Wu, W. (1998) ‘The Impact of Weather on UK Agricultural
Production, 1867–1913’,Research in Economic History, 18, 83–102.
Kondratieff, N.D. (1935) ‘The Long Waves in Economic Life’, Review of Economic Statis-
tics, (1935), XVII, (6), 105–115.
Kydland, F.E. and Prescott, E.C. (1982) ‘Time To Build andAggregate Fluctuations’, Econo-
metrica, 50, (6), 1345–70.
Lamb, H.H. (1977) Climate: Present, Past and Future, London: Methuen.
Lamb, H.H. (1982) Climate, History and the Modern World, London: Methuen.
Lewis, W.A. (1978) Growth and Fluctuations 1870–1913, London: Allen and Unwin.
Lucas, R.E. (1977) ‘Understanding Business Cycles’, In K. Brunner and A.H. Meltzer
(eds), Stabilisation of the Domestic and International Economy. Carnegie-Rochester
Conference Series On Public Policy 5, Amsterdam and New York: North Holland.
Lucas, R.E. (1981) Studies In Business Cycle Theory, Oxford: Blackwell.
Mills, T. (1991) ‘Are Fluctuations In UK Output Transitory Or Permanent?’, The Manchester
School, LIX, (1), 1–11.
Nugent, J.B. (1973) ‘Exchange Rate Movements and Economic Development in the Late
Nineteenth Century’, Journal of Political Economy, 81, (5), 1110–35.
Phelps Brown E.H. and Hopkins, S.V. (1950) ‘The Course of Wage-Rates in Five Countries,
1860–1939, Oxford Economic Papers, 2, (2), 226–96.
Romer, C. (1986) ‘New Estimates of Prewar GNP and Unemployment’, Journal of Eco-
nomic History, 2.
Rostow, W.W. and Kennedy, M. (1979) ‘A Simple Model of the Kondratieff Cycle’, in Paul
Uselding (ed.),
em Research in Economic History, 4.
Rowthorn, R.E. and Solomou, S.N. (1991) ‘The Macroeconomic Effects of Overseas In-
vestment On The UK Balance of Trade, 1870–1913’, Economic History Review, XLIV,
(4), 654–64.
Schumpeter, J.A. (1939) Business Cycles: A Theoretical, Historical and Statistical Analysis
of the Capitalist Process, New York: McGraw Hill.
Sheffrin, S.M. (1988) ‘Have Economic Fluctuations Been Dampened? A Look At Evidence
Outside The United States’, Journal of Monetary Economics, 21, 73–83.
Solomou, S.N. (1987) Phases of Economic Growth, 1850–1973: Kondratieff Waves and
Kuznets Swings, Cambridge: Cambridge University Press .
Solomou, S.N. (1994) ‘Economic Fluctuations, 1870–1913’, In R. Floud, and D.N. Mc-
Closkey (eds), The Economic History of Britain Since 1700, Vol. 2, Second Edition,
Cambridge.
Solomou, S.N. (1996) Themes In Macroeconomic History: The UK Economy 1919–1939,
Cambridge University Press.
Solomou, S.N. (1998) Economic Cycles, Manchester University Press.
Solomou, S.N. and Catao, L.A.V. (1998) ‘Effective Exchange Rates, 1879–1913’, Depart-
ment of Appied Economics Working Paper 9814, Cambridge University.
26 Solomos Solomou
Solomou, S.N. and Wu W. (1999) ‘Weather Effects on European Agricultural Output, 1850–
1913’, European Review of Economic History, December, 351–73.
Solomou, S.N. and Wu W. (2000) ‘Macroeconomic Effects of Weather Shocks 1870–1913’,
Department of Appied Economics Working Paper, Cambridge University.
Temin, P. (1976) Did Monetary Factors Cause The Great Depression?, New York: W.W.
Norton.
Temin, P. (1989) Lessons From The Great Depression, Cambridge, Mass: MIT Press.
Thomas, B. (1973) Migration and Economic Growth, Revised Edition, Cambridge: Cam-
bridge University Press.
Tobin, J. (1980) Asset Accumulation and Economic Activity: Reflections On Contemporary
Macroeconomic Theory, Chicago University Press, Chicago.
Wigley, T. M. L. and Atkinson, T. C. (1977) ‘Dry Years in South-East England since 1698’,
Nature, 265, (5593), pp. 431–34
Zahorchak, M. (1983) Climate: The Key to Understanding Business Cycles, Linden, N.J.
Zarnowitz, V. (1981) ‘Business Cycles and Growth’, Reprinted As Chapter 7 of V. Zarnowitz
(1992) Business Cycles: Theory, History, Indicators and Forecasting, University of
Chicago Press, Chicago and London.
Zarnowitz, V. (1992) Business Cycles: Theory, History, Indicators and Forecasting, Uni-
versity of Chicago Press, Chicago and London.
2 Growth and fluctuations
The post-war Japanese
experience
Hiroshi Yoshikawa
2.1 Introduction
The century-long experience of the Japanese economy provides economists with
a great opportunity to study growth, business cycles and structural change. Japan,
so poorly endowed with natural resources, has grown to become one of the major
industrial nations. Figure 2.1 shows annual growth rates of real GDP for the pre-
war period from 1885 to 1940 and the post-war period from 1955 to 1995. The
pre-war Japan kept 3 per cent growth on average for more than half a century, and
the growth rate accelerated to almost 6 per cent in four decades after World War II.
The standard Ramsey model cannot satisfactorily explain this postwar acceleration
of growth.
Her growth has not been necessarily smooth. A glance at the figure reveals that
the pre-war Japanese economy was on the whole more unstable than the post-war
economy. The standard deviation of the growth rate is, in fact, 3.5 per cent for the
whole post-war period whereas it is 3.7 per cent for the pre-war period (Table 2.1).
The standard deviation for the post-war period is, however, inflated by the fall of
growth rate in the early 1970s. If we divide the sample period taking into account
this break, we find that the standard deviation declines to 2 per cent as against 3.7
per cent for the pre-war period. Why was the post-war Japanese economy more
stable than the pre-war economy? It is an interesting question. But in this chapter,
I simply draw attention to this important fact, and focus on the post-war period. A
good general reference for the post-war Japanese economy is Nakamura (1981).
The Japanese economy started from ashes after the Second World War. Real
GDP fell to a half of the pre- war peak level, and the economy suffered from
hyperinflation. Most Japanese economists believe that the economy completed the
12
10
6
Percent
–2
–4
–6
1885
1889
1893
1897
1901
1905
1909
1013
1917
1921
1925
1929
1933
1937
1941
1945
1949
1953
1957
1961
1965
1969
1973
1977
1981
1985
1989
1993
Figure 2.1 Growth rate of real GDP
600
Inflow
400
200
0
Outflow
–100
Figure 2.2 Population flow 1955–80 into and out of Tokyo, Osaka and Nagoya metropoli-
tan areas.
Source: Annual Report on Internal Migration Statistical Bureau, Management and Co-
ordination Agency.
4.0
3.5
Growth rate of households
2.5
Percent
2.0
1.5
1.0
0.5
0.4
1956 1960 1965 1970 1975 1980 1985 1990
Note: The figure for 1970 is an outlier; Okinawa prefecture (Ryuku Island) was returned to
Japan by the United states in 1970.
32 Hiroshi Yoshikawa
needed only one of each consumer durable such as a refrigerator, television set,
washing machine and car. In fact, young people giving up agriculture left rural
villages for urban industrial areas where they formed new households. Household
formation necessarily generated additional demand for houses, consumer durables,
and electricity. In this way, population flow sustained high domestic demand in
the period of high economic growth, 1955– 70.
Along with the creation of a large number of households, the high growth
period was also the diffusion process of newly available consumer durables. The
diffusion of consumer durables was facilitated by a steady decline in prices of
those products over time on the one hand, and an increase in income on the other.
Electric washing machines, for example, first appeared in the Japanese market in
1949. At the time, a machine cost 54 thousand yen while the average annual labour
income was about 50 thousand yen. Understandably, only twenty machines were
sold per month! By 1955, only six years later, however, the price of a washing
machine had been reduced to 20 thousand yen while the average annual income
had risen to above 200 thousand yen. About a third of households could afford to
own a washing machine in 1955. The same story holds for other consumer durables
as well. Since it was urban cities that led this diffusion process, urbanization not
only created new households but also in itself sustained high demand for those
consumer durables. By the end of the 1960s, however, most of the then available
consumer durables saw saturation in the domestic market.
This whole process of domestic demand-led high economic growth (1955–70)
is schematically summarized in Figure 2.4. Channels 1, 2 in the diagram have been
well recognized: capital accumulation in the industrial sector raising labour de-
mand brings about population flow from rural agricultural areas to urban cities. In
addition to these well-recognized channels, I emphasize the oft-neglected and yet
very important fact that such population flow in turn, creating new households and
raising demand for consumer durables and electricity, ultimately sustained prof-
itability of investment in manufacturing industry (channels 3, 4, 5). I underline that
the role of newly available consumer durables was not confined to demand for those
products themselves. Through an input-output interrelationship, they augmented
demand for intermediate goods such as steel and electricity, and accordingly high
investment in those sectors.
In this virtuous circle for high economic growth, low real wages were not so
instrumental as Lewis (1954) emphasized. Rather it was growth of domestic de-
mand that sustained profitability of investment. And for growth of consumption
demand, a steady rise in real wages, rather than low repressed wages, is a con-
tributing factor. In fact, in the pre-war period, except for 1920–21 and 1929–30,
real wages saw little increase, while in the post-war period they enjoyed steady
growth. A steady rise in real wages sustained effective demand for the post-war
Japanese economy because the key product was consumer durables which not yet
being international competitive, had to find a domestic market.
It is important to note that the pre-war Japanese economy was, like today’s
NIES, a typical export-led economy (see Shinohara (1961)). The key industry was
cotton, and low real wages were in fact instrumental for international competi-
tiveness. The export/GNP ratio in pre-war Japan was, for example, 20 per cent
Growth and fluctuations 33
(Embodied technology)
Figure 2.4 Domestic-demand-led high economic growth of the Japanese economy, 1955–
70.
(1931–40) as against 10 per cent in the 1955–70 period. Chenery, Shishido and
Watanabe (1962), using the input–output analysis, also drew a similar conclusion
on the difference between the pre-war and post-war Japanese economies.
The domestic demand-led virtuous circle for economic growth based on the
Lewisian dual structure is not unique to the post-war Japanese economy. Kindle-
berger (1967, 1989), for example, discusses the post-war growth of the European
economy in a Lewisian model; See also Boltho (1982). The relation between house-
hold formation and economic growth is also well-known. The so-called building
cycle has a long tradition (See, for example, Hickman (1974)). However, in the
case of the pre-war US economy, particularly prior to the restrictive immigration
law of 1924, it was immigration rather than domestic population flow that endoge-
nously responded to the growth of income and fueled household formation. And
in the building cycle, it is residential construction that brings about a close relation
between household formation and economic growth.
In the post-war Japanese economy, it was firm’s fixed investment rather than
residential construction that played a crucial role in growth and fluctuations. In the
1960s, residential construction was less than half of total private construction. Its
share rose to above 70 per cent in the mid-1970s when the growth rate sharply fell.
Household formation by way of diffusing consumer durables rather than residential
construction, ultimately sustained high investment in manufacturing industry. In
this sense, my explanation for the Japanese economic growth during the period
of 1955–70 is more similar to Gordon (1951) than to the building cycle theory.
34 Hiroshi Yoshikawa
Table 2.3 The relative contribution of demand components to the business cycle (%)
Gordon argued that high investment, both residential and nonresidential, in the US
in the 1920s was fueled by urbanization and diffusion of the automobile.
In the Japanese economy during 1955–70, population flows and the consequent
household formation by way of diffusing newly available consumer durables con-
tinuously stimulated economy-wide investment demand. Fluctuations in invest-
ment demand were in turn the most important generating force of business cycles
from the late 1950s through the 1960s. To demonstrate this point, I show the extent
to which different demand components have accounted for different shares of the
change in GNP (Table 2.3). Since the Japanese economy has been growing rapidly,
almost all variables increase in absolute terms even in recessions I therefore first
calculated the change in each variable measured from trough to peak in case of a
recovery, and from peak to trough in case of a growth recession. I then subtracted
the latter from the former to obtain the difference. Table 2.3 reports the relative
contribution of each demand component to this cyclical difference in the change
in real GNP. It is the average for the eight cycles during the period beginning
February 1957 through March 1990. For the sake of comparison, I also present
results for the United States.
In Japan throughout the whole period, the relative contribution of fixed invest-
ment has been the greatest of all the demand components: 60 per cent of GNP on
average. In contrast, in the United States fixed investment accounts for only 25 per
cent on average of the change in real GNP. The relative contribution of inventory
and housing investments is greater in the United States than in Japan. Changes
in housing investment in Japan are not really systematic over the business cycle.
On the other hand, until the mid-1960s, inventory investment had a large impact
on the Japanese business cycle: a 60–70 per cent contribution. A substantial por-
tion of the inventory investment was, however, raw materials – which were also
imports. Therefore, the contribution of inventory investment and imports almost
canceled each other out. As a result, fixed investment retained its importance. As
a long-term trend, the role of inventory investment in the business cycle seems to
have diminished in both Japan and the United States.
Net exports have been counter-cyclical in Japan’s business cycle except for the
years 1977–85, in which economic growth was export-led as we will see shortly.
In particular, imports have been very counter- cyclical: the fraction of output was
–52 per cent on average, compared to –17 per cent in the United States. Until very
recently, the bulk of Japanese imports consisted of raw materials and therefore
moved very mechanically in parallel with the level of aggregate economic activity.
The contribution of consumption to GNP seems to be in large part similar in
the two countries. As for government expenditures, we find them counter-cyclical
Growth and fluctuations 35
for Japan (–12 per cent of GNP on average) but neutral (0.4 per cent) for the United
States. In sum, the major differences between Japan and the United States lie in
the facts that fixed investment plays a much larger role in the business cycle in
Japan than in the United States.
Coming back to the high growth era, we find that the situation changed dramat-
ically around 1970. By then the pool of the so-called disguisedly unemployed in
the agricultural sector had been largely exhausted. Therefore the population flow
from the rural sector and the associated urban household formation both sharply
decelerated. At the same time, the then available consumer durables saw saturation
in the domestic market. In this way, the domestic demand-led virtuous circle for
high economic growth was lost. Judging from Figures 2.2 and 2.3, we note that
this structural change occurred around 1970, a few years in advance of the first
oil embargo in 1973 (see also Horie et al, 1987).
The structural change is clearly seen for individual industry as well as the
macroeconomy. Table 2.4 shows the time series of capacity and investment in the
petrochemical industry. By the end of the 1960s, the industry had faced the major
turning point. The situation of the industry was typical, not exceptional. The first
oil shock hit the Japanese economy which had already seen the structural change
The most important single factor in the Japanese slowdown is the sharp decline
in the rate of technical change. I have now succeeded in linking that decline
Growth and fluctuations 37
directly to energy prices through the energy using bias of technical change in
Japan.
Jorgenson’s analysis rests entirely on the traditional growth accounting. His result
actually attributes nearly a half of a fall in the rate of economic growth in the 1970s
to capital input. And yet, noting that ‘after the energy crisis as well as before, the
growth rate of capital input was higher than that of output’, Jorgenson argues that
‘rather than causing the slowdown, the growth of capital after the energy crisis
contributed to the continued growth of output at unsustainable levels’. Thus he
concludes that ‘the decline in the growth rate of capital is not the cause of the
slowdown in Japanese economic growth’.
Capital accumulation is certainly not exogenous, however. And what brought
about a fall in the rate of economic growth and thereby caused the decline in the
growth rate of capital in the first place? Jorgenson’s answer is, of course, a fall
in the rate of technical change caused by the oil shock. In the growth accounting,
technical progress is identified as the part of the blossom of economic growth
which cannot be explained by contributions of production factors, the so-called
‘residual’. I argue, however, that whether the residual measured by the growth
accounting method really captures technical progress or not is an open issue.
Two approaches, Bruno and Sachs (1985) and Jorgenson (1988), both attribute
a fall in the rate of economic growth in the 1970s ultimately to the oil shock in
1973. These approaches cannot explain, however, why the second oil shock which
occurred in 1979 did not bring about a similar fall in the rate of economic growth;
the average growth rates of real GNP for 1973–80 and 1981–90 are 4.1 per cent
and 4.2 per cent , respectively. During the first oil crisis, the oil price quadrupled
in 1973–74 while in the second oil crisis it only doubled in 1979–80. It might be
argued that the first oil crisis hit the Japanese economy harder than the second oil
crisis. But transfer payments to OPEC necessitated by an increase in the oil price,
when seen as relative to GNP, were actually comparable during the two oil crises,
at 3.8 per cent and 4.1 per cent , respectively. The supply side analyses, such as
Bruno and Sachs (1985) and Jorgenson (1988), which attribute a fall in the rate of
economic growth to the first oil crisis are, therefore, inconsistent with the fact that
the second oil crisis did not entail a similar fall in the growth rate. Nor can they
explain why the growth rate of the oil-importing Korean economy fell so sharply
at the second oil crisis while the effect of the first oil crisis was relatively small,
which is converse to the Japanese case. I maintain that demand is an indispensable
part of any reasonable explanation of the 1970s.
However, I do not mean to argue that the oil crisis did not affect the supply
side of the Japanese economy, but I do argue that a permanent fall in the rate of
economic growth beginning in the early 1970s was caused by a domestic structural
change as explained above rather than the first oil crisis. In this respect, I concur
with Maddison (1987). By his careful growth accounting, Maddison finds that the
growth rate of real GDP in Japan would have been 3.8 per cent during 1973–84
as against its actual value 3.6 per cent , ‘if it had been possible to maintain the
relation between energy growth and GDP growth in the previous period’ (his Table
15b). In his view, the effect of the oil shocks on growth rate is plainly minor.
38 Hiroshi Yoshikawa
2.3 The post high growth period: 1970–90
In any case, the Japanese economy saw a major structural change around 1970.
Table 2.1 shows that the rate of growth of real GNP was halved from 10 per cent to
4 per cent , but at the same time the Japanese economy became much more stable
than in the high growth period. I address this issue of stability first.
to 10.5. At the beginning of the 1970s, investment seems to have been governed
by micro-specific rather than economy- wide factors. By way of enlarging the
diversity of timing of investment among industries, it enhanced the stability of
investment in the economy as a whole. The calculation above suggests that this
effect alone explains more than half of a decline in the variance of the growth rate
of investment in the second period.
Beyond that, there are two other factors to have enhanced stability of invest-
ment. One is an increase in the share of the non-manufacturing sector. Investment
in the non-manufacturing sector, compared to that in the manufacturing sector is
stable: standard deviations are 14.1 and 8.8 (1968 Q2 – 90 Q4), respectively. An
increase in the share of the non-manufacturing sector from 54.1 per cent (1968–72)
to 63.8 per cent (1973– 90), therefore, necessarily contributed to the stability of
total investment beginning in the 1970s.
The motives for investment also saw a major change. The Development Bank
of Japan has surveyed the investment of large corporations by motives. According
to this survey, 67.4 per cent of investment was done for the purpose of augmenting
capacity in 1969, but its share declined to 28.7 per cent in 1980. Now measured
by standard deviation of the growth rate, the volatility of investment motivated
by capacity augmentation is much greater than that of other kinds of investment
such as labour/energy saving and R and D (research and development): 24.8 per
cent and 14.0 per cent , respectively (1977–90). The change in the motives for
investment, therefore, also contributed to the stability of total investment.
I emphasize that the three factors mentioned above are all real rather than
monetary. Some economists argue or used to argue that stability of monetary
growth is responsible for the stability of real GNP. Suzuki (1985), for example,
argues that
Since 1975 the money growth rate in Japan has become more stable and the
uncertainty associated with it has also decreased due to the announcements
of targets. And since 1976 there has been an accompanying decrease in the
variability of the inflation and real growth rates.
However, the major influence which brought about the stability of real GNP be-
ginning in the 1970s was investment, and the stability of investment in turn was
40 Hiroshi Yoshikawa
60 Equipment
50
Percent
40
Construction
30
caused by three real factors.1 The stability of monetary growth is basically nothing
but a mirror image of the stable real economy.
Acknowledgement
Section 2.2 draws upon Yoshikawa (1995).
Notes
1 However, the real business cycle theory cannot adequately explain business cycles in
Japan. See chapter 2 of Yoshikawa (1995) for details.
2 Beginning in 1997, the credit crunch exerted substantially negative effects on the real
economy, however. See Motonishi and Yoshikawa (1999).
Growth and fluctuations 45
References
Aoki, M. and H. Yoshikawa, (2001) ‘Demand Saturation: Creation and Economic Growth’,
Journal of Economic Behavior and Organization (forthcoming).
Balassa, B., (1964) ‘The Purchasing Power Parity Doctrine: A Reappraisal’, Journal of
Political Economy December.
Boltho, A., (1982) ‘Growth’, in A. Botho (ed.) The European Economy: Growth and Crisis,
Oxford: Oxford University Press.
Bruno, M. and J. Sachs (1985) Economics of Worldwide Stagflation, Cambridge, Mass.:
Harvard University Press.
Chenery, H. B., S. Shishido and T. Watanabe (1962) ‘The Pattern of Japanese Growth,
1914–54’,
em Econometrica, January.
De Long, J. B. and L. Summers (1991) ‘Equipment Investment and Economic Growth’,
Quarterly Journal of Economics, May.
Gordon, R. A. (1951) ‘Cyclical Experience in the Interwar Period: The Investment Boom
of the Twenties’, in Conference on Business Cycles, New York: National Bureau of
Economic Research.
Hickman, B (1974) ‘What Became of the Building Cycle?’, in P. David and M. Reder (eds),
Nations and Households in Economic Growth, New York: Academic Press.
Horie, Y., S. Naniwa and S. Ishihara (1987) ‘The Changes of Japanese Business Cycles’,
Bank of Japan Monetary and Economic Studies, 5, (3), December.
Inada, K., S. Sekiguchi and Y. Shoda (1992) The Mechanism of Economic Development,
Oxford: Oxford University Press.
Jorgenson, D. (1988) ‘Productivity and Economic Growth in Japan and the United States’,
American Economic Review, May.
Kindleberger, C. (1967) Europe’s Postwar Growth, Cambridge, Mass.: Harvard University
Press.
Kindleberger, C. (1989)‘The Iron Law of Wages’, in his Economic Laws and Economic
History, Cambridge: Cambridge University Press.
Kuroda, M. and K. Yoshida and D. Jorgenson (1984) ‘Relative Price Changes and Biasis of
Technical Change in Japan’, Economic Studies Quarterly, August.
Lewis, W. Arthur (1954) ‘Economic Development with Unlimited Supplies of Labour’,
Manchester School of Economic and Social Studies, 5, May.
Lincoln, E. (1988) Japan: Facing Economic Maturity, Washington DC: Brookings Institu-
tion.
Maddison, A. (1987) ‘Growth and Slowdown in Advanced Capitalist Economies: Tech-
niques of Quantitative Assessment’, Journal of Economic Literature, June.
Minami, R. (1968) ‘The Turning Point in the Japanese Economy’, Quarterly Journal of
Economics, August.
Minami, R. (1970) The Turning Point of the Japanese Economy (in Japanese), Tokyo:
Sobunsha.
Motonishi, T. and Yoshikawa, H. (1999) ‘Casuses of the Long Stagnation of Japan During
the 1990s: Financial or Real?’ Journal of Japanese and International Economics, June.
Nakamura, T. (1981) The Postwar Japanese Economy: Its Development and Structure,
Tokyo: University of Tokyo Press.
Ohkawa, K. and H. Rosovsky (1973) Japanese Economic Growth, Stanford: Stanford Uni-
versity Press.
Shinohara, M. (1961) Growth and Cycles of the Japanese Economy (in Japanese), Tokyo:
Sobunsha.
46 Hiroshi Yoshikawa
Suzuki, Y. (1985) ‘The Macroeconomic Performance of the Japanese Economy and Mone-
tary Policy’, (in Japanese), Bank of Japan, Institute for Monetary and Economic Studies,
Kinyu Kenkyu, August.
Taylor, J. B. (1989) ‘Differences in Economic Fluctuations in Japan and the US: The Role
of Nominal Rigidities’, Journal of Japanese and International Economies, June.
West, K. (1992) ‘Source of Cycles in Japan, 1975–1987’, Journal of Japanese and Inter-
national Economies, March.
Yoshikawa, H. (1990) ‘On EquilibriumYen-Dollar Rate’, American Economic Review, June.
Yoshikawa, H. (1993) ‘Monetary Policy and the Real Economy in Japan’, in Kenneth Sin-
gleton (ed.) Japanese Monetary Policy, Chicago: University of Chicago Press.
Yoshikawa, H. (1995) Macroeconomics and the Japanese Economy, Oxford: Oxford Uni-
versity Press.
Yoshikawa, H. and T. Okazaki (1993) ‘Postwar Hyper-Inflation and the Dodge Plan, 1945–
50: An Overview’, in J. Teranishi and Y. Kosai eds., The Japanese Experience of Eco-
nomic Reforms, London: Macmillan.
3 Productivity–investment
fluctuations and structural
change
Bernhard Böhm and Lionello F. Punzo
3.1 Introduction
In this chapter, we review the growth experience in the last two decades of three
European countries and compare it with that of Japan and the US. The stylized
facts of these economies (Sections 3.5–3.6) exhibit dynamic discontinuities in
whose interpretation we employ concepts imported from dynamical systems and
bifurcation theories. Relying upon a framework of cross-regime dynamics (intro-
duced in Sections 3.2–3.6) where regime switches represent structural changes,
our approach departs from traditional and modern growth theories also in two
other respects. On one side, we use a sectorally disaggregated description of the
economies. On the other, we focus on observed sectoral paths which are typically
fluctuating, instead of looking for long run, steady states homogeneous within the
dynamic structure of a given economy. In fact, dynamic heterogeneity turns out to
be so pervasive that it has surfaced in recent growth literature.1
From a methodological point of view, the present exercise goes in the direction
of developing qualitative econometrics as suggested by R. H. Day (e.g. in (1992),
(1993), (1994)). This approach to modelling stresses emulation of qualitative prop-
erties of given historic data sets rather than quantitative estimation and prediction.
We make an attempt at marrying it with an inductive framework which focuses
upon the interdependence between structure and dynamics. Sections 3.10–3.14
experiment with recent techniques for the econometric modelling of structural
change, while the last section reviews some of the open issues in the research
agenda on multi-regime dynamics for large scale model economies.
while the growth rate of gross, physical capital formation per person employed in
sector j is
gi,j = d(Ij /Ej )/ dt∗(Ej /Ij ) (3.2)
An economy is a set of, say, k distinct sectoral sub-systems whose time behaviour
is represented by paired time series of growth rates. There will be therefore 2 by
k pairs at each date in the history of the given economy (from 2 by k time series).
A pair of values for (3.1)–(3.2) gives the co-ordinates of the dynamic path of the
assigned sector in a plane. The co-ordinate space will be called the Framework
Space (FS) after a certain logical operation has been performed.
The plane is endowed with the Innovation and Accumulation axes, taken in
the conventional order. The former is associated with the growth rate of value
added per person employed, the index of productivity growth chosen here. The
other axis records the pace of investment plotting values of the variable (3.2),
indexed again by both the sector index and the corresponding date. The abscissa
axis is the Innovation axis as in the neo-Schumpeterian interpretation produc-
tivity dynamics is functionally independent of capital accumulation, at least in
the extremistic representation proposed before. Conversely, the Accumulation axis
monitoring the process of change in investment intensity, would be the focus of
conventional aggregate theories of growth and technological progress. One novelty
in our framework is that the two axes are plotted one against the other.
A sectoral path can now be traced as a sequence of dated states, or pairs of
co-ordinate values in FS. States are dated according to a ‘clock’, a device defin-
ing conventionally the relevant time horizon. One can consider the time intervals
used to construct the original data set as the ‘natural’ periodisation. This does not
necessarily provide a clock for all investigative purposes.12 A sort of temporal ag-
gregation, imposing an ‘artificial clock’, can often be useful as a simple smoothing
device.
The criteria to carry this out may vary with the phenomena under investigation
and one’s own viewpoint. One may, thus, orientate the determination of time in-
tervals on for example the macroeconomic business cycle, measured by the time
span between peak or trough values of the GDP growth rate. Alternatively, external
or supplementary information can be used to determine a time breakdown based
upon the rhythms of some exogenous and/or domestic shocks.13 These two can
be interpreted as macroeconomic clocks and be employed to generate a uniform
periodisation across sectoral dynamics. However, a general cycle period may differ
sensibly from sectoral cycles and, for certain investigations, the properties of sec-
toral fluctuations must be retained. A clock and the implied periodisation should,
Productivity–investment fluctuations and structural change 51
therefore, be considered as experimental devices whose worth depends upon what
they allow us to see.
Once the clock has been chosen, a state represents the growth path of a sector
at a given date. Thus, in comparison to more conventional approaches, the second
novelty introduced here is that a path is observed through two state variables,
instead of one. Moreover, overall evolution is reconstructed as a set of growth
paths akin to a segmented trend14 , exhibiting in general a variety of oscillatory
patterns. Of course, segmentation collapses into a single (smooth) trend if we
choose the long run periodisation, i.e. a ‘date’ as long as the total span of the time
series available.15
The geometric representation in FS borrows concepts from the mathematical
theory of dynamical systems. Any given sequence of dated states or trajectory
is, thus, a single instance or realization of the phase portrait in the sector’s own
phase diagram. As any of the empirically given pairs of growth rates typically
jumps dramatically around its state space, sectors are similar to generic dynamical
systems in the two-dimensional (gv , gi ) plane with some complicated dynamics.
That the actual history of a sector gives us a single time series of states (single
trajectory) implies that it is difficult (if not impossible) to recover the mathematical
model, or dynamic equation(s) with appropriate restrictions, which generate such
trajectory. However, one such system does exist conceptually and it may take a
formulation in continuous or discrete time.
We pick up the latter as it is standard in the econometric and time series
approaches. Some of the econometric issues involved in practical model selection
are discussed in Sections 3.10–3.14, where also the issue of linear versus non-linear
specifications is reviewed. Here, for the sake of the argument, we may simply posit
a two-dimensional system of first order, generally non-linear difference equations
3.4 Regimes
It was pointed out how the one-sector model (3.3) reveals a twofold difficulty. One
relates to mathematical formulation for, if we choose discrete time dynamics as
might seem more appropriate, we are already in a realm of chaotic dynamics, given
that in principle the system (3.3) is non-linear.19 The other difficulty is related to
the insufficient statistics to identify the explicit form of the system, i.e. the pair of
Φ1 , Φ2 functions. Both problems are obviously greatly amplified once we allow
for a feature of reality, that is the dynamic interdependence among sectors as when
positing system (3.4). In fact we are dealing with k parallel but coupled two-
dimensional dynamical systems (the k sectors). Even if we assume each of them
to be relatively simple, the overall or global dynamics can be anything one may
imagine. An economy has a complex dynamical scheme.
One may hope that, in real economies, interdependences take up some simpler
form, with asymmetries introducing hierarchical orders, hence decomposing (3.4)
into smaller, in principle computationally simpler, block systems of non-linear
equations. At any rate, modelling (3.4) explicitly requires a strategy to choose
the Φs.20 On one hand, one can introduce a set of hypotheses into an estimated
version of (3.4), to the effect of generating dynamic behaviours compatible with
Productivity–investment fluctuations and structural change 53
actual, observed dynamics. The latter approach is based on an understanding of
the dynamical system (3.4) as a reduced form of some, yet unspecified, model in
structural form. Alternatively, in a more standard approach we start with assump-
tions defining one such structural form, then the corresponding reduced form is
derived, and finally simulated dynamics is analyzed in comparison with the actual
one. Along this line, it would be natural to begin with production functions, etc. as
is typical of production-oriented modelling, and to choose them with the desired
properties. The two strategies are, of course, compatible with one another, the dif-
ference being essentially methodological. (Section 3.11 illustrates an application
of the former and reviews some of the estimation problems involved.21 )
Here we proceed heuristically as follows. If (3.4) above renders the global
model of a sector’s behaviour, one can segment it into a set of local models giving
dynamics under particular conditions, i.e. for certain values of the two co-ordinates.
The intuition is that, locally, one can almost always represent dynamic behaviour
with a description that in principle is simpler (often, linear) than the overall dynamic
model. Such a local model is meant to explain the dynamics from one path to a
‘nearby’ path of the same family or dynamical class. A regime is a family of growth
paths that are all generated by one and the same standard model, a canonical model,
for restricted sets of values of its parameters. Thus, in the FS we are going to
distinguish six regimes plus one special regime, the Harrodian generalized set. To
start from the latter, all paths exhibiting the typically steady state property of time-
constant ratios of investment to value added belong to the Harrodian generalized
set. The latter, of course, includes as special cases those steady state paths of
constant levels of capital and output discussed in conventional growth theories (as
shown in Section 3.7). The set of paths with investment and value added growing at
the same rate (i.e. the 45◦ line) represents the natural extension. When it comes to
empirical analyses, it will be convenient to speak of a Harrodian corridor around
the line, where the two rates are almost equal to each other, to allow for small
deviations, errors and the like, in statistical data. As the problem is general,22 it is
better to keep calling it a set, and recall that it has the typical property of Harrodian
paths, being a knife-edge.23
The ratios between growth rates in the FS yield one of the parameters of the
canonical model. They can be either larger or smaller than one, and a ratio of
exactly one can be treated as a bifurcation value. It corresponds to all paths be-
longing to the Harrodian set, and it can be used to characterize that set compared
to all others. On the other hand, the four semi-axes can be used to yield the second
parameter. All pairs of values of growth rates in the first and third quadrants pre-
serve the same signs (positive value is associated with positive, and vice versa),
while for paths in the 2 and 4 quadrants signs are interchanged. This reflects the
fact that the underlying relationship between levels of variables, i.e. ν and i, is
increasing or decreasing.24 This can be represented, simplistically, by a second
parameter ranging on the real line: for positive (or negative values) we get thus
either relationship.25 The Harrodian set together with the other 4 semi-axes in
a FS 26 can now be used to induce a particular partition into dynamical regimes.
Each regime corresponds to a family of realizations of the canonical model for
54 Bernhard Böhm and Lionello F. Punzo
values of the two parameters in partitions of the parameter space induced by their
‘bifurcation values’, 1 and 0, respectively.
In the Innovation regime (regime I) corresponding to the area of the first quad-
rant below the Harrodian set, all paths show positive productivity growth rates
exceeding positive investment growth rates. The name was justified in Section
3.2, as in this area the functional association between productivity growth and
gross capital formation is nil or weak. Likewise, the area above the set, where
productivity falls behind investment growth, is the regime that can be associated,
though not uniquely, with conventional growth theories which conceive growth as
a capital-driven path (it is regime VI). With the quadrants numbered clockwise,
beginning with the innovation regime – and observing that the positive and the
negative quadrants are further subdivided by the Harrodian set – a classification is
obtained: with number II being associated with ‘restructuring’ and showing nega-
tive investment growth but positive productivity growth, while the remaining three
are mirror images of those just described.
It is only when the (gv , gi ) co-ordinate plane is endowed with this theory-
induced partition, that it makes sense to call it framework space. Traditionally,
theories see only regimes I and VI (and their polar cases, III and IV respectively)27 .
The introduction of regimes II and V presents us with the possibility of analyzing
oscillations that fall outside standard economic dynamics. Oscillations are now
fundamentally across growth paths and these cannot be treated as (sometimes,
purely virtual) long run equilibria or steady states.
Dynamics that takes across regimes can be associated with structural change,
for it is the ‘model of growth’ that is changing, then, not just its quantitative
properties.
• Structural change phasing was different across countries, a facet which can
be partially attributed to political business cycles and other related phenom-
ena (e.g. implementation of specific economic and industrial policies).
• Finally, Harrodian paths are rarely observed and generally they are short
lived.
The latter observation is quite important for the ensuing discussion: the Solow-path
belongs to the same set. But, if we compare our representations of the long-run
(the last picture in each movie), they tend to look similar, which seems to support
the hypothesis of a steady state prevailing in the long-run. Against our structural
dynamics that is working in the medium run, this observation suggests also that long
run (equilibrium) theories basically disregard important segments of the histories
of actual economies. They rely on the optical deformation created by comparing
pictures that are taken at some time distance and leave out the history to which
they belong.
It is this history that reveals the interesting feature, it exhibits the fluctuating
character of economic dynamics and the changing structures that support it. It can
be directly compared with some of the findings of the growth empirics literature.
(See, for instance, van Ark and Toniolo (1996).)
60 Bernhard Böhm and Lionello F. Punzo
3.7 Growth theories and the fluctuating dynamics in the FS
This was a brief review of what can be seen by applying our analytical framework
to data that are normally analyzed by a growth approach. The fact that growth rates
are here allowed to fluctuate, shows that the FS dynamics tries to embed growth
into the more general phenomenon of economic oscillations. With the possibility
of moving across regimes (travelling along special ‘traverses’35 ) such structural
dynamics incorporates business oscillations and growth as special cases.36
For a comparison with growth analyses, we shall briefly consider the neoclas-
sical aggregate theory of growth and the disaggregated one associated with the
multisectoral model. Both versions employ, explicitly or implicitly, some notion
of production function. We have argued that our approach does not (necessarily)
depend on production functions. This is in a sense trivially true as, technically
speaking, we have no ‘formal model’. Or, at least, not a unique model: for formal
models we depend on others, and this is what we want to show in this and partly
the next sections.
Aggregating over our sectors, we may obtain the growth rates of aggregate
productivity and investment intensity. If then, we average them over the whole
time period (collapsing the sets of dates into a single one), we get something
resembling long run values. Finally, projection onto the horizontal axis of the
history of a given economy thus condensed by a single long-run path, gets us the
key ingredient in the aggregate growth description. Output is GDP, and hereafter
will be denoted as usual by Y . It can readily be calculated as the sum of VAs. Theory
starts with level values of aggregative variables, Y , K for total capital stock, E and
N for total employment and labour force, and so on so forth. From this first level
are derived ratios like y ≡ Y /E, k ≡ K/E, and the like. Finally, we obtain growth
rates to describe dynamic properties of growth paths. In the FS also, we follow a
similar three-layer procedure, keeping the hierarchy between the lowest layer of
raw data and the subsequent layers which are obtained via simple manipulation.
In other words, just like in the standard growth approach, from the lower we also
derive a unique ‘higher level’ description. In the opposite direction the procedure,
of course, does not lead to unique identification, at most it leads to a whole class.
That is, a path in the FS is associated with a whole set of pairs of time series in
the level values of the variables. (This is an implication of the parameterization
introduced above to define regimes.)
Now it can be argued that a notion of regime(s) can be traced in growth theory
and is implied by much of recent growth empirics, though there it is interpreted
as an equilibrium behaviour. In fact, neo-classical growth theory is said to imply
a ‘prediction’ on the long-run dynamic behaviour of an economy. This says that,
under certain conditions (basically reflecting properties of the production function),
(i) there will be a unique steady state value of y, the output per capita, equal under
full employment, labour productivity; and that (ii) the long run growth rate of output
per capita obeys the equation: g − n = λ, where n ≡ dN /dt N −1 , the natural rate
of growth, and λ is the rate of growth of technological progress. The long-run rate
of growth of per capita output would then be controlled entirely by exogenous
Productivity–investment fluctuations and structural change 61
technological progress. In the absence of the latter, it would be equal to zero. One
can express this result by saying that the endogenous growth rate of per capita
output gy = (g − n), and of labour productivity (if e = n), are both equal to zero.37
One can write this growth rate as a co-ordinate value on a real line and call
the latter the growth line of per capita output or productivity. Each gy value is
associated with a growth path. Hence, in the absence of technological progress of
exogenous type, the neo-classical theory says that the long-run steady state value
of gy lies at the origin of this line. If it is not there, this is due to λ not being zero.38
Thus, the productivity growth line decomposes into the union of an equilibrium set,
made up of a unique point, and the set of all other paths.39 While the equilibrium
path is sustainable in the long run, all other growth rates can only be associated
with short run dynamics. At any rate, stability in the large implies that, eventually,
the observed path is the one associated with the origin (or with the exogenously
fixed value gy = λ). Any other path describes transient dynamics.
Notice that, due to this twofold property, uniqueness of equilibrium and sys-
tem global stability, this is a non-linear dynamical systems with properties in a
sense typical of a linear one.40 On the other hand, notice also that many dynamic
paths that are associated with different time series in levels (e.g. in the values of
Y, K, E and N) are zoomed into a single equilibrium path on the gy -axis. This
shows that the latter is like a one-dimensional version of our Framework Space.
Hence, in principle, we can introduce a two-regime classification here too: the
systematic equilibrium behaviour, and the transient dynamics. An analogous idea
can be found in, for example, the endogenous growth literature which, re-phrased
in our jargon, distinguishes between a Solow-type regime (with diminishing re-
turns to capital) and a ‘non-Solow’ regime. Any growth rate greater than the Solow
rate is sustainable in the long run if constant returns to capital are at work, hence
parameterization of the implied model is made with reference to the parameters
of the underlying production functions. The introduction of two distinct regimes
on the gy -line rationalizes the possibility of settling into different steady states.41
The related partition can be used to explain as an equilibrium outcome, the per-
sistent difference of growth rates (increasing divergence of growth paths) across
countries, a feature of recent growth empirics that has found large support in the
evidence.
We can re-set the above argument to show some other of its implications and
ramifications. The Neo-classical theory ‘exhausts’ itself on the gy -axis, for the
steady state goes along with full employment and the long run rate is exogenously
fixed by demographic forces as well as by technological change. There is no need
for supplementary endogenous explanatory variables, except for the adjustment
or transitional dynamics. It is really this that, together with the full employment
hypothesis, implies uniqueness of the equilibrium solution.42 This is also the con-
ceptual basis for cross country studies. With each economy one can associate the
observed growth path as a pair of values for y and the productivity growth rate.
All such paths however, monitored through growth rates, are expected to eventu-
ally converge to one and the same value on the growth line. It is the simplicity
of the neo-classical model, that the properties of one dynamics mirrors to a large
62 Bernhard Böhm and Lionello F. Punzo
extent the properties of the other. Still, we have a dual dynamics between levels
and growth rates that is put together via the various notions of convergence (but,
in particular, with β-convergence).43
The key idea that has emerged in the literature, is that there may be more than
one long-run growth rate, so that countries need not converge to a unique long-
run value of y (nor to a small interval around it). It is to account for the apparent
diversity of the growth phenomenon across countries, if not yet to ‘explain it’,
that the single axis of productivity growth is not enough. Kaldorian and (all) other
endogenous growth theories do exactly this: they introduce extra axes to explain
the long run. They are at variance for their choices of such axes.
Thus, some of the debates raging in the recent, theoretical and empirical, growth
literature can be re-interpreted as statements about long-run dynamics in a one-
dimensional or a two- (or larger) dimensional version of our FS. In the latter,
one can enrich the picture on the gy -axis in different ways, e.g. admitting the
possibility of multiple steady states. This idea extends naturally to the analysis of
many countries, belonging to different clubs and then settling in the neighborhood
of a distinct long run attractor. Or else one can do country by sectors studies, like
those in our ‘movies’ of Appendix 3.2. In all cases, stability is an open issue and
a sensitivity to initial conditions exhibits properties typical of some non-linear
chaotic systems.
Therefore, in our two-dimensional FS there is room for all long run theories:
at its origin, if they are of the neo-classical sort; at its right, if they are projections
of paths in two (or more) variables, one variable being used to explain the other.
Hence, exogenous and endogenous explanations of growth can be taken to rep-
resent two classes of ‘models of long run dynamics’ that need not be excluding
each other. They may replace one other at different times as well as in different
economies or sectors. This depends on the empirical evidence.
However, to capture the stylized facts of the experience of growth, we need to
do more and account for the whole FS. The common long run view of the growth
theories confines itself to the first quadrant and to the equilibrium phenomena that
may appear there. To charter the rest of the plane, we need to look at dynamics
in the ‘shorter run’, shorter than the time span preferred by the growth views, and
allow for out-of-equilibrium dynamics. This is the dynamics taking across paths.
Treatment of investment as the second variable brings in the instability due to
its typically volatile time profile. Growth theory can speak of stable monotonic
dynamics as long as investment is kept out of the picture (e.g. by confining it to
explain the short run adjustment at business cycle frequencies). Once a shorter time
horizon is chosen, investment behaviour re-appears. No surprise, then, in the FS
the typical mode becomes oscillatory dynamics in growth rates. In those instances
where it is fairly regular, it resembles a growth cycle. The empirical evidence shows
that in general it is not of such a simple dynamic variety.44 In the appropriate time
horizon, in fact, the interplay between the chosen variables may take up different
forms: multiple schemes, time varying schemes. To accommodate this feature, the
model implicit in our FS has to be a non-linear system of simultaneous equations
with (at least) two state variables for each sector.
Productivity–investment fluctuations and structural change 63
In this light, the Framework Space is a heuristic tool to classify certain em-
pirically observed ‘growth phenomena’, when they can be actually observed, and
inductively produce a usable theory of actual facts.45 And it is proposed as a first
step for an approach that recognizes the possibility of different theories being
consistent to each other.
t
D(zτ |Zτ −1 ; Θ) (3.6)
τ =1
D(zt |Zt−1 ; Θ) = F(Wt |Zt ; a)G(Yt |Yt−1 , Xt ; b)H (Xt |Yt−1 , Xt−1 ; g). (3.7)
One can recall some standard definitions, in the present context. Thus, strong exo-
geneity would require (Xt |Xt−1 ; g), lagged endogenous variables may not influence
the exogenous ones. Super exogeneity would in addition require parameter vectors
b and g to be independent. Then a change in b would not influence g, which makes
the specification immune to the Lucas-critique.
The partial log-likelihood function of the model can be written
log L(Θ) = L(Θ; yt |xt , yt−1 ) (3.8)
t
and forms the basis for estimation. As the chances of arriving at a correct model
specification according to the true model and the restrictions found by conditioning
and marginalizing are very small, the aim is to find an ‘adequate’ model, i.e.
statistically acceptable and not outperformed by rival models.
Assuming that this process has been successful (and in particular, the condition-
ing step has identified exogenous and endogenous variables), a general dynamic
specification of the form (3.5) allows a derivation of long run properties of the rela-
tionship between the two variables on the basis of the short run information on the
data over the period of observation. Now, to recall our variables, let us interpret y
for the level of labour productivity, and re-denominate x with i = I /E, the intensity
of gross capital accumulation (or gross capital accumulation per employed), and
to simplify, let us use national aggregate data. Moreover, for analytical reasons,
growth rates in the FS will be approximated by logarithmic first differences.
68 Bernhard Böhm and Lionello F. Punzo
Taking only first order lags in the general equation (3.5), one has
which has a static steady state solution for growth rates: D log y = D log i = 0 given
by
log y = a/(1 − d) + [(b + c)/(1 − d)] log i − c/(1 − d)gi − d/(1 − d)gy (3.11)
which is seen to depend on the levels of the variables involved and shows, therefore,
time dependence.
Taking differences of (3.11), assuming the restriction b + c + d = 1 to hold
and the long-run growth paths constant (but different from zero), yields D log y =
D log i, a result compatible with points on the 45◦ line of the diagram. This is
a generalized Harrodian set as it is the linear (one-dimensional) subspace of FS
spanned by all paths with growth rates for the two variables, equal to each other.
A narrow band around it was previously called the ‘Harrodian corridor’ in view
of measurement inaccuracy or, as in the present case, the stochastic nature of the
approach. If, on the other hand, the restriction b + c + d = 1 does not hold, we
have a long run steady state of proportional growth off the Harrodian subspace48 .
Productivity–investment fluctuations and structural change 69
The value of [(b + c)/(1 − d)] will indicate the nature of such a long run state
and may classify the system under investigation to which regime the path would
belong. (This justifies the treatment of the model as parameterized by that value, as
carried out in Section 3.4). If, finally, actual growth paths were variable, shifts of
the linear relationships of growth rates would be implied, that depend on second
order dynamic movements, the changes of long run growth rates or (long run)
growth acceleration. In fact it is easily seen that taking differences of (3.11) is
equivalent to assuming a dynamic (first order) relationship in growth rates apart
from a constant, i.e.
where the errors of the two regimes (uit ) are identically normal distributed with
mean zero and constant variance and zt is a transition variable. D(zt ) is a heavyside
function taking the value of one if zt ≥ c and zero if zt < c, a constant. Since esti-
mation of this discrete model is complicated Goldfeld and Quandt have suggested
a continuous approximation to the transition function using the cumulative normal
(c, σ 2 ) distribution function with the errors of the regimes assumed to have equal
constant variances σ 2 . Using this transition function in the regression will define
a smooth transition regression model. This idea has also been followed up in the
time series literature (cf. Tong (1990)).
Markov switching times series models may be suitable for time series with
jumps or oscillations (i.e. sequences of short jumps). One can proceed along the
following steps. First test hypotheses on the number of states present in a particular
time series. This determines the number of regimes. Then define the appropriate
switching model and its likelihood function. Thereafter estimate the parameters
and calculate the ML estimates of transition probabilities.
A simple example of the application of a discrete time series switching model
to a two-dimensional problem like productivity and investment intensity growth is
the following. Assume that each of the variables may be observed in two regimes,
one defined by positive growth rates, the other one by negative ones. A combination
72 Bernhard Böhm and Lionello F. Punzo
of the two variables involved will therefore produce four regimes in which each
pair of growth rates can be found. The sequence of regimes obtained can then be
read off the simplified time series graph which contains just the averages of values
of the whole period in which one regime obtains. Representing this in the phase
space gives just four points (i.e. average growth rates), each in one regime, each
of which is characterized by one of the four quadrants.
Figure 3.1 Scatter plot of growth rates of investment and value added per employment
Thus, changes should occur during the second and third and around the fifth and
the tenth period. We can graph these generated series y1 and y2 in diagram (Figure
3.4) and draw a scatter plot (Figure 3.5):
One can observe that there are structural changes in both series around period
10. Whereas series y1 exhibits the break only during one period the regime shift
in series y2 lasts longer (between periods 2 and 3, and 5, and 10). Turning to their
growth rates and the framework space as given by the following graphs in Figures
3.6 and 3.7, one can readily recognize the regime shifts despite the added noise to
the two equations in levels.
Productivity–investment fluctuations and structural change 75
Table 3.1 Estimation of NL least squares
y1, y2
50
0
0 10 20 30
Time
y1(t)
y2(t)
100
y1
50
0 10 20 30 40
y2
(y2, y1)
The framework space makes these shifts even more obvious. We find four pairs
of growth rates to be located outside the area around the zero point, corresponding
to regime shifts according to the construction of the transition function.
The cluster around zero follows from the relationship of both endogenous
variables with the exogenous variable, a stochastic trend, which enters linearly
whenever the effect of the transition function is zero. Transformed into growth
rates both variables have to fluctuate around zero due to the noise component.
We may conclude this excursion into some of the more promising econometric
approaches which might shed some light on the possibility to identify relevant
economic processes with a quotation by T. Haavelmo in his Nobel lecture (1989):
2
Growth rates
2
0 5 10 15 20 25 30
Time
dy1
dy2
2
dy1
2 1 0 1 2
dy2
(dy2,dy1)
45°-Line
We have seen that our framework space can accommodate several theories. How-
ever, it still remains a formidable task for theories to provide a convincing expla-
nation of economic change and for the econometrician to account for the changing
relevance of those theoretical explanations.
Acknowledgement
The author acknowledges the financial support of the Italian CNR, contract n°
98.03803.CT10, and of Siena University through its PAR 1999. Part of the work
reported in this chapter is also the result of a close collaboration with other re-
searchers in the IDEE Project, financed by the EU within the TSER action of the
5th S & T Framework Programme.
Notes
1 See for instance Bernard and Jones (1996), and the debate on convergence reviewed
e.g. in Durlauf and Quah (1998).
2 Thus, we leave aside models explaining dynamics on the basis of demand behaviour
(e.g. Keynesian models) or which conjugate demand and supply conditions in a theory
of structural dynamics (e.g. see the work of Morishima , Pasinetti, for instance). We
also leave out the neo-Austrian approach, see Amendola and Gaffard (1988), (1998).
Our framework is compared with the latter in Amendola, Gaffard and Punzo (1999).
3 For a discussion, see Scott (1989).
4 We mean alternative to the neo-classical one.
5 Again, that there is no production function does not mean that one cannot insert one
compatible with the Kaldorian model.
6 There is, of course, a distinct formulation of the model of growth as reflecting embodied
technological progress, the one proposed by Arrow with capital stocks of different
vintages, but this version is not the one that interests us here.
Productivity–investment fluctuations and structural change 81
7 Goodwin (1967); see also Goodwin and Punzo (1987); Goodwin (1990). Punzo (1995)
discusses the relation between the present empirical framework and those works.
8 This construction was introduced in Böhm and Punzo (1994) and discussed in Punzo
(1995).
9 In fact, one can build an analogue framework with different sets of variables and their
growth rates (if this is suggested by the problem to investigate). Notice that gross
capital formation could have been re-defined so as to measure capital in a broader
sense. Here, we take physical capital in the Kaldorian and Solow type of tradition.
10 See Appendix 2 on statistical data bases.
11 In the econometric sections below, it will be useful to resort to a log approximation of
(3.1) and (3.2).
12 Clearly, there is nothing natural about this choice that is dictated by statistical difficul-
ties rather than the timing of the phenomenon under observation.
13 The clock chosen in Böhm and Punzo (1994) reflects the hitting of exogenous shocks,
and therefore reports external information. It was used to check, to a certain extent at
least, the relevance of aggregate shocks in explaining the structural dynamics of the
economies involved.
14 There is some relation with the econometric literature on broken trends here. The idea
of a variety of patterns of growth, on the other hand, has recently become popular even
in the tradition of endogenous and exogenous aggregative theories. There, however, it
is attributed to cross country, rather than to cross sectoral dynamics.
15 This is the choice of growth theory as is done in the conventional mode. Thus, the
latter too has a place in the FS, justifying the latter’s name.
16 The system (3.3) is taken to have no forcing term, an analogue of a VAR formulation.
The forcing term may be used to introduce consideration of the influence on system
dynamics of an exogenous force, that can have stochastic or deterministic nature or can
sum up the two. The forcing term will therefore determine, together with endogenous
properties, the system’s long run dynamics. For such formulation see again Section
3.11.
17 A notion introduced by R. M. Goodwin in his (1947) contribution, and later investigated
in Goodwin and Punzo (1987).
18 Again, see Böhm and Punzo (1994, 1995). The literature on Growth Empirics is full
of exercises of the first type.
19 The alternative is obviously to assume that the model be linear, as done in much
econometric practice, again see Sections 3.11–3.12, but this has a cost that may be
quite high.
20 The dynamic specification of (3.4) as a first order autoregressive process is only used
for simplicity. The appropriate order of these processes needs to be derived from
econometric testing.
21 Of course, estimation problems enter the latter modelling strategy as well, though at a
different stage.
22 It will be the case whenever we will deal with borderlines between regimes in the
partition of the state space below.
23 It is a linear one-dimensional subspace, whence the knife edge property descends.
24 Compare with the discussion in Section 3.11 below, in particular equation(s) 3.10,
which contains two parameters.
25 And of course, its zero would be again a bifurcation value and correspond to the origin.
In this light, the origin corresponds to a bifurcation value of a family of maps with two
parameters.
26 We are using orthogonal co-ordinates.
27 Actually the former, i.e. the first quadrant , can be called the growth and the latter, the
third quadrant, the contraction quadrants.
28 As in Böhm and Punzo (1995).
82 Bernhard Böhm and Lionello F. Punzo
29 Movie refers also to the possibility of animating the graphs, a technique amply demon-
strated in the work of R. Abraham and C.D. Shaw (1989).
30 It is to be understood that the decision of defining the FS in terms of two coordinate
variables (the growth rates) already makes the model more complex than a standard
growth model. The latter can be argued to imply a one-variable version of our FS.
Two however is surely less than the desired dimension for describing an economy;
however, it already shows all the complexity that a higher dimensional model would
have to deal with. The choice is made, among other reasons, to keep a multisectoral
model manageable.
31 This and the following sections are based upon Punzo (1997).
32 See Industrial Policies in OECD countries, Annual Review 1990, OECD Paris, 1990.
33 See for example Komiya, Okuno and Suzumura (1998), Moriguchi (1991), (1995),
Yoshikawa (1995).
34 For more details, the interested reader is invited to look at the publications from the
ongoing IDEE project, as well as previous publications by the authors and other as-
sociates to the project. See also Amendola and Gaffard (1988), (1998); Amendola,
Gaffard and Punzo (1999); Böhm (1996).
35 ‘Traverse’ in the sense of Hicks (1973).
36 This belongs to the tradition of Schumpeter and Goodwin.
37 Let d Log E/dt ≡ e , E being aggregate employment. It is typical of neo-classical
and endogenous growth models (including the Kaldorian and Cambridge tradition) to
introduce as an assumption a full employment equilibrium path. This is not accepted
into our framework.
38 Given a known value of λ, we can always measure the economy’s growth rate in
deviation. Thence, the notion of ‘endogenous growth rate’ is in this case the residual.
39 The latter is the union of two open intervals, the closure being the equilibrium set (a
point or one-dimensional manifold).
40 In particular, the long-run state, an attractor, is a monotonic equilibrium path, the
simplest of all dynamic morphologies.
41 An idea ingenuously exploited by, for example, Durlauf and Johnson (1995), though
in a different analytical setting.
42 If employment levels were let free, in other words, we would have an interval of long-
run values. Typically, in growth theories this possibility is associated with short-run
dynamics, as they all descend one way or another from Harrodian dynamics.
43 β-convergence is an exercise whereby a cross country regression is run of growth rate
on initial level of output per capita. Hence, a one dimensional FS space with a single
gy -axis is completed with a y-axis. We talk of β-convergence when the regression
coefficient is negative, implying that the poor countries run faster towards the long run
equilibrium path. See the vast literature, e.g. Baumol, Nelson and Wolff (1994), and
Durlauf and Quah (1998). The resulting framework is not a two-dimensional FS in our
sense, for one variable is interpreted as the explanation of the other. Econometrically
it is a one equation model, rather than the two equation system implicit in our FS and
described by (3.3) or (3.4) above.
44 See for instance Goodwin (1991), or two collections, Benhabib (1992) and Day and
Ping Chen (1993) for a sample of a burgeoning theoretical literature.
45 Changing the variables on the y-axis then can lead to different frameworks where the
classification between exogenous and endogenous changes too, as they are relative to
the explanatory variables on such axis. In our framework where the system is generally
simultaneous, the distinction in principle does not make sense.
46 See for instance some of the models used later to discuss the econometrics of structural
change.
47 See Goodwin (1990).
48 See the treatment of the model as parameterized by b = (b + c)/(1 − d), in Section
3.4.
Productivity–investment fluctuations and structural change 83
49 This result obtained from using a first order lag-polynomial is easily generalized
to higher orders. The relevant restriction for long run unit elasticities will then be
B(L)[A(L)]−1 = 1 in the notation of eqn. (3.5).
50 Thus, the emergence of a long run steady state in growth rates is a very special case
of a growth cycle collapsing to a single state as a global attractor.
51 We are referring to non-stochastic versions of Growth and BC theories with which
our approach can be directly compared. In fact, the latter can be best understood
if formulated with a deterministic, i.e. non stochastic, model, where irregularity is
endogenously generated rather than shock induced. This does not imply that stochastic
phenomena cannot be added to it. Most of modern economic dynamics is formulated
in a stochastic environment, to derive some statistically average long-run properties.
52 Actually there is a growing literature on this, (see for instance some of the papers
in the collection edited by Benhabib (1992)), though such literature focuses upon the
theoretical possibility of observing highly irregular dynamics associated with relatively
simple multisectoral models.
53 See Brida, Puchet and Punzo (2000).
54 See Lavezzi (2000).
References
Abraham, R. and C. D. Shaw (1989), Dynamics. The Geometry of Behavior, Aerial Press,
Santa Cruz, CA.
Amendola, M. and J.-L. Gaffard (1988), The Innovative Choice, Oxford University Press,
Oxford.
Amendola, M. and J.-L. Gaffard (1998), Out of Equilibrium, Clarendon Press, Oxford.
Amendola, M., J.-L. Gaffard, and L. F. Punzo (1999), ‘Neo-Austrian Processes’, Indian
Journal of Applied Economics. 8(2) 277-93.
van Ark, B. and N. Crafts (1996), Quantitative Aspects of Post-war European Economic
Growth, Cambridge University Press, Cambridge.
Baumol, W. J., R. R. Nelson and E. N. Wolff (eds) (1994), Convergence of Productivity,
Oxford University Press, New York.
Benhabib J. (ed.), (1992), Cycles and Chaos in Economic Equilibrium, Princeton University
Press, Princeton.
Bernard, A. B. and Jones, C. I. (1996) ‘Comparing Apples to Oranges: Productivity Conver-
gence and Measurement Across Industries and Countries’, American Economic Review,
86 (5), 1216–38.
Böhm B. (1996), ‘Dynamic Econometric Specification and the Analysis of Structural
Change’, Dynamis, Quaderno 2/96, IDSE, Milano.
Böhm, B. and L. F. Punzo (1994): ‘Dynamics of Industrial Sectors and Structural Change
in the Austrian and Italian Economies, 1970–1989’, in Böhm B. and L. F. Punzo (eds),
Economic Performance. A Look at Austria and Italy, Physica Verlag, Heidelberg.
Böhm, B. and L. F. Punzo (1995), ‘Structural Change in the Context of Uneven Regional
Development. The Path of Italian Dualistic Economy Revisited with a New Dynam-
ical Approach’, paper presented at the XI International Conference on Input Output
Techniques, New Delhi, 1995.
Brida, G., M. Puchet and L. F. Punzo (2000), ‘Coding economic dynamics to represent
regime dynamics’, WP no. 307, Dept. of Political Economy, Siena.
Day, R. H. (1992) ‘Complex Economic Dynamics: Obvious in History, Generic in Theory,
Elusive in Data’, Journal of Applied Economics, 7, S9-S23.
84 Bernhard Böhm and Lionello F. Punzo
Day, R. H. (1993), ‘Non-linear Dynamics and Evolutionary Economics’, in Day, R.H.,
and P. Chen (eds) (1993), Non-linear Dynamics and Evolutionary Economics, Oxford
University Press, New York.
Day, R. H. (1994), (2001), Complex Economic Dynamics, vols. 1 and 2, The MIT Press,
Cambridge, Mass.
Day, R. H. and Ping Chen (eds) (1993), Non-linear Dynamics and Evolutionary Economics,
Oxford University Press, Oxford.
Doornik J. A. and D. F. Hendry (1994), PcFiml 8.0, Int. Thomson Publishing.
Durlauf, S. N. and D. Quah (1998) ‘The New Empirics of Economic Growth’, in Taylor, J.
and M. Woodford (eds.), Handbook of Macroeconomics, Elsevier Science, Amsterdam.
Durlauf, S. N. and P. A. Johnson (1995), ‘Multiple Regimes and Cross-country Growth
Behaviour’, Journal of Applied Econometrics, 10(4) 365-84.
Goldfeld, S. M. and R. E. Quandt (1972), Non-linear Methods in Econometrics, North
Holland, Amsterdam.
Goodwin, R. M. (1947), ‘Dynamical Coupling with Special Reference to Markets Having
Production Lags’, Econometrica; reprinted in Goodwin, (1982).
Goodwin, R. M. (1967), ‘A Growth Cycle’, in C.H. Feinstein (ed.), Socialism, Capitalism
and Economic Growth; reprinted in Goodwin (1982).
Goodwin, R. M. (1982), Essays in Non-linear Economic Dynamics, Macmillan, London.
Goodwin, R. M. and L. F. Punzo (1987), The Dynamics of a Capitalist Economy. A Multi-
Sectoral Approach, Polity Press and Westview Press, Cambridge and Boulder, CO.
Goodwin, R.M. (1990), Chaotic Economic Dynamics, Clarendon Press, Oxford.
Haavelmo, T. (1989) ‘Nobel Memorial Lecture’, in K. G. Mäler (ed.), Nobel Lectures in
Economic Sciences, World Scientific Publishing Company.
Hansen B. E. (1992), ‘Testing for parameter instability in linear models’, Journal of Policy
Modeling, 14, 517–533.
Hendry D. F. (1987), ‘Econometric Methodology: A personal perspective’, in Bewley T.F.
(ed.) Advances in Econometrics, ch.10. Cambridge University Press, Cambridge.
Hendry D. F.,A. J. Neale and F. Srba (1988), ‘EconometricAnalysis of Small Linear Systems
using PC-FIML’, Journal of Econometrics, 38, 203–226.
Hendry D. F., A. R. Pagan and J. D. Sargan (1983), ‘Dynamic Specification’, in Griliches,
Z. and M. D. Intriligator (eds) Handbook of Econometrics, North Holland, Amsterdam.
Hicks, J. (1973) Capital and Time, Oxford University Press, Oxford.
Johansen, S. (1988), ‘Statistical analysis of cointegration vectors’, Journal of Economic
Dynamics and Control, 12, 231–254.
Kaldor, N. (1957) ‘A Model of Economic Growth’, The Economic Journal, LXVII, Decem-
ber, 691-724.
Kaldor, N. and J. Mirrlees (1961), ‘A New Model of Economic Growth’, Review of Economic
Studies, 29, 174–90.
Komiya, R., M. Okuno and K. Suzumura (eds) (1988), Industrial Policy of Japan, The
University of Tokyo Press.
Lavezzi, A. (2000), ‘Structural instability and unemployment: Evidence from the Italian
regions’, WP no.292, Dept. of Political Economy, Siena.
MacKinnon, J. G. (1991), ‘Critical Values for Cointegration Tests’, ch.13 in Engle, R. and
C. W. J. Granger (eds), Long-Run Economic Relationships, Oxford University Press.
Maddala, D. S. (1977), Econometrics, McGraw Hill, New York.
Moriguchi, C. (1991) ‘The Japanese Economy and Economic Structural Adjustments’ Eco-
nomic Studies Quarterly 42(1) (March), 1–11.
Productivity–investment fluctuations and structural change 85
Moriguchi, C. (1995) ‘Japan’s Macro Economic Policy and Japan–US Economic Relations
During the 1980s’. In Dutta, M. (ed.) Economics, Econometrics and the Link, Elsevier
Science, Amsterdam.
Punzo, L. F. (1995), ‘Some Complex Dynamics for a Multi-sectoral Economy’, Revue
Economique, 46, 1541–60.
Punzo, L. F. (1997), ‘Structural cycles and convergence during processes of economic
integration’, published in Spanish as ‘Cyclos Estructurales y Convergencia durante los
procesos de integracion economica’, Revista de Economia, Banco Central del Uruguay,
4(2) 3–5.
Quah, D. (1993), ‘Empirical cross section dynamics in economic growth’, European Eco-
nomic Review, 37 (April), 426–34.
Ramsey, J. B. (1974), ‘Classical model selection through specification error tests’, in P.
Zarembka (ed.), Frontiers in Econometrics, Academic Press, New York.
Scott, M. F. (1989), A New View of Economic Growth, Clarendon Press, Oxford.
Teräsvirta, T. (1996), ‘Modelling Economic Relationships with Smooth Transition Regres-
sions’, Working Paper No.131, Stockholm School of Economics, November 1996.
Tong, H. (1990), Non-linear Time Series. A Dynamical Systems Approach, Oxford Univer-
sity Press, Oxford.
Yoshikawa, H. (1995), Macroeconomics and the Japanese Economy, Clarendon Press, Ox-
ford.
Appendix 1
The data base
Data for Germany, France, and the USA are taken from the OECD publication,
National Accounts, Vol II (various issues).
Value added by kind of activity (table 12), gross fixed capital formation by kind
of activity of owner (table 3), employment by kind of activity (table 15).
For Germany all value added and investment data at constant prices have been
converted into 1991 prices, for USA they have been converted into 1985 prices,
and for France into 1980 prices.
Data for Italy have been constructed from two sources: for the period 1970
to 1980 they are from the ISTAT publication Annuario di Contabilità Nazionale,
Tomo 2, tav1.7 etc. and converted into constant prices 1980. For data from 1980
until 1992 they are from the ISTAT data bank as published. Tests for the period
1980 to 1983 for both data sets have shown that the calculated growth rates are not
affected by the systemic change in Italian national accounts.
The data for Japan come from regular publications by EPA, Economic Planning
Agency.
Table 3.A1 Coding of sectors
Italy 1970-73
Italy 1973-79
25 25
13
20 20
15 42 15
10 60 10
58
01 28
5 53 36 5
01
gi
gi
15 24 53 5036
0 060 28 47 24
47 60 15
74 06 17 58 42
-20 -15 -10 -5 -5 0 86 5 1050 15
69 20 25 -20 -15 -10 -5 -5 0 86 5
74 10 15 20 25
69
-10 17
-10
-15 -15 13
-20 -20
gv gv
25 25
20 20
15 15
10 47
10 17 28
47 58 50 24
15 28 1
5 36 586 69 156036
6 6042 13
124 17 6 42
58 13 0
0 53 53
69 86
-20 -15 -10 -5 0 550 10 15 20 25 -20 -15 -10 -5 0 5 10 15 20 25
-5 74
-5
74
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 20
15 13 15
6
10 10
28 28
5 60 5 6013
58 36 6 15 47
gi
4236
gi
53 15 58
1 24
0 69 17 50 0 53 50
74 24 60
86 86 5
-20 -15 -10 -5 0 5 10 15 20 25 -20 -15 -10 -5 74 0 17
10 15 20 25
-5 47 42 1 -5
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 1 20
37 36
33
15 15
7
10 9 10
32 4
6
5 5 2
5 38 34
gi
gi
8 5
2 31 4 7 1 31
0 0 35 39
34 638
39
-20 -15 -10 -5 0 5 10
36 15 20 25 -20 -15 -10 -5 0 9 335 10 15 20 25
-5 35 -5 8
32
37
-10 -10
-15 -15
-20 -20
gv gv
25 25
34
20 20
15 15 36
2 32 38
10 10 9
35 33
32
6
5 9 5 31 5 7 37
3339 1
gi
gi
37
31 1 390 8
0 6 3836
34 35 4
-20 -15 -10 -5 0 8 55 10 15 20 25 -20 -15 -10 -5 -5 0 54 10 15 20 25
-5 7
2
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 20
15 15
10 10
36
5 5 39 5 34 7 1
9 38
gi
7
gi
35 5 3233
1
0 9 0 63139
638 31 3735 4
8 36 4 -20 -15 -10 -5 0 8
2
5 10 15 20 25
-20 -15 -10 -5 0 5 10 15 20 25
-5 33 32 -5
34 37
-10 -10
-15 2
-15
-20 -20
gv gv
25 25
20 20
15 15
2 9
4
9
10 10
2 39
33 1
34
5 8 1 5 5
38
6 36 32
gi
gi
7 39 36
0 0 333735
315 317 4
37 6
-20 -15 -10 -5 0 5 10 15 20 25 -20 -15 -10 -5 0 8 5 10 15 20 25
-5 32 -5
38
-10 -10
34 35
-15 -15
-20 -20
gv gv
25 25
20 20
34
15 15
10 10 39
6
33
35 383632
5 25 37
2 4 32 7
gi
31 5
gi
1
37 9 7
0 39 838 0
36 31 9 1
-20 -15 -10 -5 6 35 345
0 10 15 20 25 -20 -15 -10 -5 0 4
-5 33 -5 85 10 15 20 25
5
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 20
15 15
5
10 33 10
31 36
6 37 1 2 7
5 8 39 5 39 1
3336 338
4
gi
gi
9
35 38 6 4
32 37
531732
0 4 7
2
0 35
34 8
-20 -15 -10 -5 0 5 10 15 20 25 -20 -15 -10 -5 0 5 10 15 20 25
-5 -5
-10 -10
-15 -15
-20 -20
gv gv
20 20
15 15
3731
10 734 10
21 3836 37
3339
8 4 36 32 34
5 69 2 5 31335
9
38 33
7
gi
gi
5 136
0 0 5
6 32
-20 -15 -10 -5 0 5 10 15 20 25 -20 -15 -10 -5 0 8 94 5 10 15 20 25
-5 -5
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 20
15 15
37
10 6 10 39
38 34 4
5 9 5 9
gi
31
gi
32 36 35
8 234 31 5
0 0 7 32 38
435
7 39 33
36 8
-20 -15 -10 -5 -5 0 37 5 101 15 20 25 -20 -15 -10 -5 -5 0 6 15 10 15 20 25
5 33 2
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 20
35
15 15
31
10 10
38 37 3731
5 36 1 5 34 35
38
34 39
gi
6 39
gi
2 36 32
0 6 08 94 6 1
2 4 32 5
33
-20 -15 -10 -5 -5 0 57 10 15 20 25 -20 -15 -10 -5 -5 0 5 10 15 20 25
9 33
5 8
-10 -10
-15 -15
-20 -20
gv gv
25 25
20 20
15 15
23 4 ELT R
18
T EX
10 17 10
23 11
22 3 PR EC
2 21 19
AUT
5 6 5 918 AUT 7
16IND 2 19 6
gi
gi
17 MACH
0 8 PR EC 9 3
0 IND T EX
11 20 22 4 10
-20 -15 -10 -5 0 5 10 15 20 7 25 -20 -15 -10 -5 0 20 5 10 15 20 25
10
-5 -5 21
8
16
-10 MACH -10
ELT R
-15 -15
-20 -20
gv gv
25 25
20 20 20
PR EC T EX
22 16
15 ELT R 15 17
AUT 4 19
21
23 MACH
10 22 7
10
6 6 IND
16 11 23
T EX 9
5 9 8 8 5 2 7
21 11 10
gi
gi
3IND 20 P REC
18 3 AUT
4 10 18 EL T R
0 0
17 2 MACH
-20 -15 -10 -5 0 19 5 10 15 20 25 -20 -15 -10 -5 0 5 10 15 20 25
-5 -5
-10 -10
-15 -15
-20 -20
gv gv
20 20
10
15 15
AUT
9 11
10 17AUT 7 10 23
P REC
MACH 3 11
20 22 18 17 22 9 ELT R
4 T EX
5 16 5 2
23 IND 81820 7
gi
4 19
gi
2
21 19 EL T R 10 3
21
0 P REC 0 IND 16 6
MACH
-20 -15 -10 -5 T EX0 5 10 15 20 25 -20 -15 -10 -5 0 5 10 15 20 25
-5 6 -5
-10 -10
-15 -15
-20 -20
gv gv
4.2.2 Economies
To capture the character of economic development in the very long run, a number
of factors, some of which have not been incorporated into macroeconomic growth
models so far, have to be taken into account. We take these up now.
The organizational structure of a modern society includes households, pro-
ducers, marketing firms, financial enterprises, and public institutions of various
kinds. It provides the coherent framework of rules and procedures within which
work can occur. It must be supported by human effort. The humans devoted to
this effort form the infrastructure for a given socioeconomic system that mediates
the human energy devoted to coordinating production and exchange, to providing
social cohesion for effective cooperation, for training and enculturating the work
force, and for producing the public goods such as waste disposal and public safety
required for the well being of the work force. The knowledge that makes this hu-
man infrastructure effective is the administrative technology, a term due to Ester
Boserup (1996). It must augment the production technology. As there are many
institutions that are involved in the various infrastructural functions, the broader
term social technology might be preferred.4
The adult, x, population is divided between the labor force engaged directly
in production, L, and the administrative or social workforce that manages the
infrastructure, M . In a decentralized economy much of this will be part of the
private sector. A large part will also be part of the public sector. Both are necessary
for a productive labor force. Given this, the number of adults in the infrastructure
outside the family, M , and the number of adults in the labor force, L = x − M .
Diseconomies are, of course, implied by resource scarcity. They also accrue
because of the increasing complexity of planning, communicating, and coordi-
nating production as the economy grows. The ability to overcome them depends
on the administrative technology and on the social space which this technology
‘produces’. The social space defines the maximum number of individuals, say N ,
compatible with an effective socioeconomic order and with the feasible operation
Qualitative dynamics and macroeconomic evolution 99
of the society’s production process. Social slack is the difference between the so-
cial space and the current number of people, S = N − x. If there is positive social
slack, then more people can be accommodated within the economy. As social space
is ‘used up’, cooperation becomes increasingly difficult, social conflict increases,
and productivity declines. When S ≤ 0, society cannot function. Only when the
social slack is positive can the society function. These internal diseconomies can
yield absolutely diminishing returns to population within an economy.
Assume that the technological production function satisfies the usual assump-
tions:
These assumptions imply that both labor and social slack are necessary for positive
production and that both labor and social slack contribute positive but declining
marginal productivities. The parameter B is, as usual, the total factor productivity
level. Substituting for L and S, the production function can be written
0 , x ∈ \(M , N )
H (x) := (4.4)
G(x − M , N − x) , x ∈ (M , N ).
In words, output depends on the technology level, labor effect and slack effect.5
A given economy is characterized by (i) its (representative) family function
(4.1), and (ii) its aggregate production function (4.4).6
4.2.3 Cultures
The key characteristic of the above formulation is that an economy based on a given
social technology is bounded by its social space. At some point in the expansion
of human numbers within an economy, the population may reach a level at which
a new economy with the same system but with a newly constituted infrastructure
can be split off in such a way as to increase welfare, in this way overcoming the
internal diseconomies of population size. In effect, the social space is increased by
increasing the number of similar economies. Contrastingly, if productivity were to
fall sharply enough as the potential limit is approached, separate economies could
merge to form a smaller number of economies with the same type of system, in
effect economizing on infrastructure.
Here we assume these possibilities can be represented by fission and fusion, the
former being the splitting of a given economy into two; the latter being the fusion
of two or more economies to form a single one using the same basic system as
before.7 In addition to the internal diseconomies implied by resource scarcity and
100 Richard H. Day and Oleg V. Pavlov
social space, external diseconomies should be recognized that derive from the total
population of all the economies together. These diseconomies are, for example,
caused by the environment’s diminishing waste absorbing capacity as population
expands and by the increase of the cost of extracting and refining resources as stocks
decline. We assume that the environmental capacity depends on the production and
administrative technology of a given system and can be expressed in terms of a
maximum population density. Diminishing absolute returns to the work force can
eventually come to pass as the total population becomes large.
For a given culture, the internal diseconomies of population can be overcome by
replication; the external ones cannot. The aggregate effects of these diseconomies
on production can be represented by a continuous environmental damage function
d = D(x̄ − x) (4.5)
where D(x̄) = 1, D(0) = 0 and D (x̄ − x) ≤ 0. The damage function reduces pro-
ductivity as environmental capacity becomes progressively exhausted. We refer
to x̄ as the environmental capacity and the term x̄ − x as the environmental slack.
Once the world is full in the sense that external diseconomies become important,
the replication of economies with the same basic structures must eventually come
to an end.8
Out of all the conceivable numbers of economies that could exist by means of
the fission/fusion process for a given system, we choose the one that is locally effi-
cient for the population of a given size. We call a production function that optimizes
the number of economies in the system the cultural production function. Given
the environmental damage function, the locally efficient number of economies is
(i, j) = IJ (x, B) := arg max max 2l Bp Hp (x/2l )Dp (x̄p − x). (4.8)
p∈T l∈N
++
where B = (B1 , . . . , Bτ ). Then the locally efficient culture is given by the culture,
i, with 2j similar economies. The production function that is locally efficient with
102 Richard H. Day and Oleg V. Pavlov
respect to the selection of a culture and the number of economies using it, is given
by
The set X ij is the set of population sizes for which Kij (x, B) is the cultural production
function with the number of economies yielding the highest total (and average)
output.
is defined by the τ equations (4.10). The ith component of the vector Bt+1 is
generated by equation (4.10a) and the remaining coefficients by equation (4.10b).
The process is asymptotically monotonically stable if 0 < ρi < 1. We can think of
B̃i as the technology potential for a given system.
The value of the technology parameter when a given regime is entered for the
first time will be called the innovating technology level, denoted by the parameter
B0 . Assuming that 0 < B0 < B̃, productivity will grow. The larger the potential and
the smaller the innovating level, the larger is the technology gap; and the larger
this gap, the more rapid the initial rate of productivity enhancement. As the stock
of practical knowledge accumulates, the rate of accumulation eventually declines,
and the technology level approaches its potential asymptotically. This process does
not expand the ultimate limits on population allowed by the associated social and
environmental spaces. That can be done in this model only by switching to a more
advanced system.
Qualitative dynamics and macroeconomic evolution 103
4.3 GEM: A ‘general’ evolutionary model
4.3.1 The complete model
Putting all the above together, a ‘general’ evolutionary model, which we call
‘GEM’, emerges in which a society is portrayed as evolving through a sequence
of alternative numbers of economies of a given type and switching among alter-
native socioeconomic systems in response to the standard of living of the current
generation and to the potential standard of living that can be ‘selected’ through
‘self–reorganization’.
Let us summarize the model as a whole. The state variables of the system consist
of the population, xt , and the vector of productivity levels, Bt = (Bt1 , . . . , Btτ ) with
one element for each member in the cultural menu. The locally efficient culture
and number of economies is given by
Yt = K(xt , Bt ). (4.13)
Average per capita standard of living is given by
⎧
⎪
⎨ 0 , xt ∈ D0(i,j)t
xt+1 = (α/q) 2it Btit (xt − M it )βit (N it − xt )1−βit − ηxt , xt ∈ Ds(i,j)t
⎪
⎩
(1 + n)xt , xt ∈ Dn(i,j)t .
(4.18)
Whenever xt enters a different phase zone, the equations governing (xt , Bt ) change.
This happens if the culture does not change (it+1 = it ) but the number of economies
increases or decreases due to fission or fusion does change, i.e. jt+1 = jt . Or, it
happens if a different culture is adopted so that it+1 = it . In this case, the number
of economies making up the new system will also change through integration or
disintegration, as the case may be.
When xt enters a new phase zone, we denote the entry time ‘s’. Then s0 = 0 and
sk is the entry time for the k th regime switch. With this definition we can define
an episode as the consecutive length of time periods governed by a given regime.
With this definition (i, j)t is the same for t ∈ {sk , sk + 1, . . . , sk+1 }. A sequence,
(i, j)sk , k = 0, 1, 2, 3, . . . describes a trajectory in terms of the sequence of phases
and phase zones through which it passes. Such a sequence is called a scenario.
Given this interpretation, the following additional endogenous variables are
determined:
aggregate production Yt
average welfare per capita yt
the time of entry into each episode sk
the duration of each episode sk+1 − sk
the governing regime in each episode (i, j)t = IJ (xt , Bt )
the governing number of economies
using the dominant system 2jt
the size of the aggregate infrastructure 2 M it
jt
To describe the major developments throughout the entire span of Homo Sapi-
ens Sapiens and to take advantage of the known archaeological information, a
reasonable minimal specification would be:
In reality, various geographical areas traversed these stages at very different times
and the advance through them did not increase uniformly from lower to higher
index. Rather, progress from one to another, especially in earlier times, was in-
terrupted by reversions to lower level stages. Moreover, fluctuations in income,
population and capital have been typical. The overall picture is one of growth at
fluctuating rates with sometimes smooth, sometimes turbulent transitions when
jumps and reversions occurred until a ‘higher’ stage became firmly established. A
summary of the archaeological and historical evidence concerning the transition
through these stages and the various regime switching events is presented in Day
(2000, Chapter 23).
A rough time line for the permanent transitions to the several stages is given
in Table 4.1.
1.00e+009
5.00e+008
1 1 1 1
0
1.00 1025.50 2054.00 3080.50 4107.00
5.00
1.00
1.00 1025.50 2054.00 3080.50 4107.00
2500000.00
1250000.00
a)
0
1.00 930.75 1860.50 2790.25 3720.00
5.50e+008
276e+008
b)
2000000.00
3721.00 3818.00 3915.00 4012.00 4109.00
Figure 4.3 Details of the population dynamics (number of families). Note the changing
time scale from (a) to (b)
110 Richard H. Day and Oleg V. Pavlov
6.00
3.50
1.00
1.00 1027.00 2053.00 3079.00 4105.00
5.00
3.00
1.00
3600.00 3727.25 3854.50 3981.75 4109.00
1 2
4.91e+008
1 2 1 2
20.75 1 2
4070.00 4090.00 4110.00 4130.00 4150.00
Figure 4.6 Two extended simulations compared. 1 and 2 indicate number of families
4.7 Conclusion
Growth theory in the hands of Tinbergen, Solow, and Swan was designed to explain
growth in the industrial countries during the first half of the twentieth century. Their
models performed remarkably well and revealed the important roles of capital
accumulation and productivity improvement. The elaboration of the theory by
Lucas, Romer, and their followers incorporates the allocation of capital and human
capital to productivity enhancement, in this way endogenizing the explanation of
productivity advance in the long run, that is, over a century or more.
The approach described here emphasizes another fundamental aspect of eco-
nomic growth that arises when the process is viewed over the very long run, that
is, over millennia. That aspect is the discrete change in the production technology
and social organization that has occurred along with the proliferation of economies
with a given fundamental system, the fluctuation in their numbers, their unification
in the transition to a system with a more elaborate social infrastructure, and their
occasional disintegration into a larger number of economies with less elaborate
112 Richard H. Day and Oleg V. Pavlov
infrastructures. The present theory characterizes this process and shows that pa-
rameter values can be chosen that fit the historical record in a qualitative sense,
that is, that lead to model generated regime changes and fluctuations among the
number of economies that are known to have occurred, according to a time line
roughly in accord with the evidence.
This would seem to be the most that can be expected of such an exercise.
Nonetheless, several important insights are suggested. First, population growth
within a given fixed social system seems to have been limited, even given the
presence of improving productivity. To overcome the limits, something more than
doing the same thing better seems to be indicated. Overcoming the diseconomies
that eventually emerge would seem to require basic changes in the way things
are done, both in production and social organization. Endless growth, therefore,
cannot be taken for granted, and collapses worse than a great depression can befall
any culture that allows population growth to go unchecked while failing to re-
organize itself to provide the means for maintaining coherence and symbiosis in
its increasing numbers.
Our representation of cultural selection and phase switching does not explain
the process by which such transitions are actually brought about. It only ex-
plains conditions sufficient to force such changes to occur. Both historian Quigley
(1979) and archaeologist Flannery (1999) have discussed these processes in socio–
political terms. Models like ours are very limited in what they can contribute to
an understanding of the grand process of macroeconomic evolution. But they can
provide a rigorous explanation of the demoeconomic forces that are involved.
Acknowledgement
This chapter was presented as a paper at the Summer School on ‘Economic Fluc-
tuations and Structural Change’ held in Siena, June 27–July 7, 1998. It draws on
an earlier lecture presented at the Columbia University Conference, ‘Managing
Plant Earth’, April 1997.
Notes
1 Growth economists have, until recently, thought of the ‘long run’ in terms of centuries.
Here we mean by ‘very long run’ growth over millennia.
2 See Day and Cigno (1978) and Day (1994). In the meantime, many alternative ap-
proaches have been and are being explored for studying microeconomic evolution.
3 The basis in household preferences for this Malthusian form is derived and the empirical
basis for it reviewed in Day, Kim and Macunovich (1989). A more general version that
incorporates a declining birth rate at high income levels is also introduced there. This
feature, however, is not incorporated in the present model.
4 It must be emphasized that infrastructural functions are carried out in both public
and private domains. The importance of the latter is sometimes overlooked. Large
scale corporations allocate roughly half their expenditures on educational, research,
managerial and administrative functions and roughly half on the production of goods
and services. Although some economists would include such things in the category
of intermediate goods used in the production process, it is worth distinguishing them
Qualitative dynamics and macroeconomic evolution 113
because their individual productivity cannot be measured in the usual ways (output
per hour expended). Their productivity, like that of the elements of public infrastruc-
ture, is only reflected in the productivity of the entire organization. The contribution
to the organization’s success of individual scientists, teachers, managers, accountants
is impossible to measure except by profit comparisons among similar organizations.
Likewise, a productive public infrastructure will be reflected in some measure of ag-
gregate accomplishment such as political, military or economic dominance, and/or a
high level of culture and wide distribution of welfare.
Infrastructure has recently been receiving increasing attention. See North (1981) for
very broad aspects and the World Bank (1994) for numerous details. For a suggestive
attempt to quantify infrastructural effects on productivity, see the working paper by
Charles I. Jenes and Robert Hall, ‘Measuring the Effects of Infrastructure on Economic
Growth’, Stanford University.
5 The conditions expressed in (4.3b) imply that lim H (x) = ∞, that lim H (x) = −∞,
x→M x→N
and that for all x ∈ (M , N ), H (x) < 0, so H (·) is strictly concave on [M , N ].
6 Day and Min (1996) show that such an economy can display all the simple and complex
possibilities: convergent growth, cycles, erratic fluctuations, and collapse. If contin-
uous (exponential) productivity improvement is incorporated, then growth or fluctu-
ations around a rising trend are possible or, as before, growth – possibly expanding
fluctuations around a rising trend – followed by a collapse.
7 An alternative which allows for emigration and immigration is the process of ‘shedding
and assimilation’ introduced in Day (2000).
8 For a discussion of the existence of an upperbound on population, see Cohen (1995a,b).
9 Necessarily, a different scale is used in each graph.
References
Boserup, Ester (1996) ‘Development Theory: An Analytical Framework and Selected Ap-
plication’, Population and Development Review, 22, 505–15.
Cohen, Joel (1995a) ‘Population Growth and Earth’s Human Carrying Capacity’, Science,
269, 34–6.
Cohen, Joel (1995b) How Many People Can the Earth Support?, Norton and Co., London,
UK.
David, Paul (1994) ‘Do Economies Diverge? Comment on Day’, in G. Silverberg and
L. Loete (eds), The Economics of Growth and Technological Change: Technologies,
Nations, Agents, Edward Elgar Publishers, Hants, UK, pp. 69–71.
Day, Richard H. and Alessandro Cigno (eds) (1978) Modeling Economic Change, Elsevier
North–Holland, Amsterdam.
Day, Richard H., Kyoo–Hong Kim and Diane Macunovich (1989) ‘Complex Demoeco-
nomic Dynamics’, Journal of Population Economics, 2, 139–59.
Day, Richard H. and Jean–Luc Walter (1989) ‘Economic Growth in the Very Long Run: On
the Multiple–Phase Interaction of Population, Technology, and Social Infrastructure’,
Chapter 11 in W. Barnett, J. Geweke, K. Shell (eds), Economic Complexity: Chaos,
Sunspots, Bubbles and Nonlinearity, Cambridge University Press, Cambridge.
Day, Richard H. and Min Zhang (1996) ‘Classical Economic Growth Theory: A Global
Bifurcation Analysis’, in T. Puu (ed.), Chaos, Soilitons, and Fractals, 7, 12, 1969–88.
Day, Richard H. (2000) Complex Economic Dynamics, Volume II, An Introduction to Dy-
namic Macroeconomics, The MIT Press, Cambridge, MA.
Easterlin, Richard A. (1978) ‘The Economics and Sociology of Fertility: A Synthesis’,
Chapter 2 in Charles Tilly (ed.), Historical Studies of Changing Fertility, Princeton
University Press, Princeton, NJ.
114 Richard H. Day and Oleg V. Pavlov
Flannery, Kent V. (1999) ‘Process and Agency in Early State Formation’, Cambridge Ar-
chaeological Journal, 9, 3–21.
Jenes, Charles I. and Robert Hall (1999) ‘Measuring the Effects of Infrastructure on Eco-
nomic Growth’, Working Paper, Stanford University.
Jones, Charles I. (1999) ‘Was an Industrial Revolution Inevitable? Economic Growth Over
the Very Long Run’, Mimeo, Department of Economics, Stanford University.
Kremer, Michael (1993) ‘Population Growth and Technological Change: One Million B.C.
to 1990’, Quarterly Journal of Economics, 108,4, 681–716.
Lee, Ronald Demos (1988) ‘Induced Population Growth and Induced Technological
Progress: Their Interactions in the Accelerating Stage’, Mathematical Population Stud-
ies, 1, 265–88.
North, Douglass (1981) Structure and Change in Economic History, W.W. Norton & Co.,
New York.
Quigley, Carroll (1979) The Evolution of Civilization, Liverty Press, Indianapolis, IN.
World Bank (1994) World Development Report 1994, Infrastructure for Development, Ox-
ford University Press, Oxford.
5 Out-of-equilibrium
dynamics
Mario Amendola and Jean-Luc Gaffard
But once we recognize that the time over which change takes place is a con-
tinuing and irreversible process which shapes the change itself, as we have
to do when we consider a qualitative change, ‘it is impossible to assume the
constancy of anything over time . . . The only truly exogenous factor is what-
ever exists at a given moment of time, as a heritage of the past’. (Kaldor 1985,
p.61) In the analysis of an out-of-equilibrium process . . . we thus have to
consider as a parameter, and hence as exogenous, not some given element
chosen beforehand by reason of its nature or characteristics, but whatever, at a
given moment of time, is inherited from the past. What appears as a parameter
at a given moment of time is therefore itself the result of processes which
have taken place within the economy: processes during which everything –
including resources and the environment, as well as technology – undergoes
a transformation and hence is made endogenous to the change undergone by
the economy. Thus, while the standard approach focuses on the right place to
draw the line between what should be taken as exogenous and what should be
considered instead as endogenous in economic modeling – a line that moves
according to what we want to be explained by the model – out of equilibrium
. . . the question is no longer that of drawing a line here or there but rather
one of the time perspective adopted. Everything can be considered as given at
a certain moment of time, while everything becomes endogenous over time.
(Amendola and Gaffard 1998, pp.32–3)
Finally, it must be stressed that the ‘fundamentals’, which determine the equilib-
rium values of the relevant magnitudes of the economy, no longer play the same
118 Mario Amendola and Jean-Luc Gaffard
role out of equilibrium, when the focus is on a process rather then on a given
configuration of the economy. Different evolution paths can be associated in fact
to given fundamentals, according to how the out-of-equilibrium process actually
evolves, and the fundamentals themselves undergo a change during this process,
given the very definition of qualitative change. The ‘fundamentals’, in other words,
are no longer fundamental.
5
F (w/p-w0/p0)
4
a)
3
0
–0.30 –0.25 –0.20 –0.15 –0.10 –0.05 0 0.05
w/p-w0/p0
7
5
F (P-P0)
4
b)
3
0
–1.00 –0.80 –0.60 –0.40 –0.20 0 0.20
P-P0
8
5
F (U-U0)
c) 4
0
0 0.05 0.10 0.15 0.20 0.25
U-U0
F (w/p-w0/p0) 6
a) 4
0
–0.25 –0.20 –0.15 –0.10 –0.05 0
w/p-w0/p0
8
5
F (P-P0)
b) 4
0
–0.9 –0.8 –0.7 –0.6 –0.5 –0.4 –0.3 –0.2 –0.1 0 0.1
P-P0
35
30
25
F (U-U0)
20
c)
15
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
U-U0
14
12
F (w/p-w0/p0)
10
a) 8
0
–0.12 –0.10 –0.08 –0.06 –0.04 –0.02 0 0.02 0.04
w/p-w0/p0
70
60
50
F (P-P0)
40
b)
30
20
10
0
–0.40 –0.30 –0.20 –0.10 0 0.10 0.20
P-P0
100
90
80
70
60
F (U-U0)
c) 50
40
30
20
10
0
0 0.01 0.02 0.03 0.04 0.05 0.06
U-U0
10
8
F (w/p-w0/p0)
a) 6
0
–0.4 –0.3 –0.2 –0.1 0 0.1 0.2
w/p-w0/p0
14
12
10
F (P-P0)
8
b)
6
0
–1.4 –1.2 –1.0 –0.8 –0.6 –0.4 –0.2 0 0.2
P-P0
18
16
14
12
F (U-U0)
10
c)
8
0
0 0.05 0.10 0.15 0.20 0.25 0.30
U-U0
Figure 5.4 Scenario 1 with increase in saving rate: a) distribution of real wages, b) distri-
bution of productivity, c) distribution of final employment.
Out-of-equilibrium dynamics 125
On the contrary, it brings about intertemporal complementarity problems that may
result in cumulative imbalances and erratic fluctuations in the growth rate of the
economy which hamper its viability.
5.6 Appendix
The following model which has been used for the simulation analysis whose re-
sults have been summed up in the figures in the text, is based on a neo-Austrian
representation of the production process, where
a = [a1 , . . . , anc , anc +1 , . . . , anc +nu ]
is the vector of labour input coefficients, and
b = [bnc +1, . . . , bnc +nu ]
is the vector of final output coefficients, in the different periods of the phase of
construction (1, . . . , nc ) and of the phase of utilization (nc + 1, . . . , nc + nu ).
Productive capacity at time (t) can be written
xc (t) = [x1 (t), x2 (t), . . . , xnc (t)]
xu (t) = [xnc +1 (t), xnc +2 (t), . . . , xnc +nu (t)]
where xc (t) and are xu (t) the vectors of the production processes in the phase of
construction and the phase of utilisation, respectively.
In a steady state:
xj (t) = xj−1 (t − 1) = xj+1 (t)G
xj (t) = xN (t)G N −j ; N = nc + nu ; j = 1, . . . , N
where G=1+g is the growth factor.
The labour supply is equal to:
LS (t) = LS (0) [1 + g]t
where g is the natural growth rate.
The resources required to carry out production and to sustain consumption are
financial resources. There are ‘external’and ‘internal’financial resources.
m(t) = min [p(t)s(t), p(t)d(t)]are the money proceeds from sales (internal fi-
nancial resources), where s and d are the supply and demand of final output and p
its price. f (t) is money supply (external financial resources)
In a neo-Austrian model, the wage fund represents the resources which sustain
the process of capital accumulation. It is determined by the minimum between the
available financial resources F(t) and the wage fund constrained by the available
human resources:
ω(t) = min [F(t), w(t)Ls (t)]
with
F(t) = m(t − 1) + hf (t − 1) + f (t) − c(t)
where c(t) is the take out (the resources withheld from financing production pro-
cesses) and hf (t) the monetary idle balances of finance which pile up when the
human resource constraint is more stringent than the financial constraint:
hf (t) = F(t) − w(t)Ls (t)
The decisions are taken as follows:
126 Mario Amendola and Jean-Luc Gaffard
Production decisions The current production is the difference between the cur-
rent supply and the stocks actually put back on the market. It cannot be greater
than the existing output capacity inherited from the past.
q t) = s(t) − o(t − 1) ≤ b
x(t − 1)
where o(t) = s(t) − d(t). It determines the vector of production processes in the
phase of utilisation xu (t)
The money value of current supply is determined on the basis of expected
money proceeds
p(t)s(t) = Em(t)
Price and wage decisions Price and wage change from one period to the next in
reaction to market disequilibria
gp (t) = κΦ(t − 1)
gw (t) = νΨ(t − 1)
where gp (t) is the rate of variations of price, Φ(t − 1) the rate of excess demand
for the final ouput, gw (t) the rate of variation of wage,Ψ(t − 1) the rate of excess
demand for labour.
References
Amendola, M. and J-L. Gaffard (1988) The Innovative Choice Oxford: Basil Blackwell.
Amendola, M. and J-L. Gaffard (1998) Out of Equilibrium, Oxford: Clarendon Press.
Benhabib, J. and K. Nishimura (1985) ‘Competitive Equilibrium Cycles’, Journal of Eco-
nomic Theory, 35, 284–306.
Grandmont, J.M. (1985) ‘On Endogenous Competitive Business Cycles’, Econometrica,
53, 995–1046.
Hicks, J.R. (1973) Capital and Time, Oxford, Clarendon Press.
Kaldor, N. (1985) Economics without Equilibrium, Cardiff University College Press.
Kydland, F.E. and E.C. Prescott (1982) ‘Time to Build and Aggregate Fluctuations’, Econo-
metrica, 50, 1345–70.
Kydland, F.E. and E.C. Prescott (1996) ‘The Computational Experiment: an Econometric
Tool’, Journal of Economic Perspectives, 10, 69–85.
Lucas, R.E. (1980) ‘Methods and Problems in Business Cycles Theory’, Journal of Money,
Credit and Banking, 12, 696–715.
6 Disequilibrium growth in
monetary economies
Basic components and the KMG
working model
Peter Flaschel
6.1 Introduction
This chapter surveys, also for the general reader interested in non-market
clearing models of growth and fluctuations, the foundations – and the core
K(eynes)M(etzler)G(oodwin) model built on them – of the general framework un-
derlying joint past and present work with Carl Chiarella and others on integrated
disequilibrium models of monetary growth. The core KMG model of disequilib-
rium growth and its analysis is founded in this chapter on specifically reformulated
and extended partial dynamic models of the literature, the PC-AC approach (of
Goodwin and Rose) and the IS–LM–PC approach (of the textbook literature). We
also briefly indicate at the end how the fundamental KMG model thus obtained
(with its six basic laws of motion) can be extended into the direction of fairly
detailed, high-dimensional macrotheoretic disequilibrium growth models of mon-
etary economies, with strong relationships to a variety of models currently used
for structural macroeconometric model-buildings and their applications.
In order to indicate the scope and perspective of the macrodynamics to be
considered in the following let us briefly discuss here the following graphical
representation of the essential components of an integrated Keynesian disequilib-
rium growth theory. Figure 6.1 shows in the middle what might be considered
the backbone of Keynes’s (1936) General Theory, the basic causal nexus, that
makes goods markets behavior (via the investment decisions of firms) dependent
on what is achieved on financial markets, and labor markets in turn on the outcome
on the goods markets with their determination of output through expected sales
(and intended inventory changes), which in turn depend on actual aggregate goods
demand (and actual inventory changes).
In the center of interest in the macroeconomic debate of the last two decades
has been, however, quite a different module of the macroeconomy, representing
so-called supply side features or, as we prefer to characterize it, the dynamic wage–
price block of a fully-specified approach to macrodynamics. This block is shown
as wage–price spiral bottom left in Figure 6.1 and it is surely no exaggeration
to state that it has dominated the development of macrostatic and macrodynamic
mainstream thinking in the last decades, even to the extent that it was claimed that
Disequilibrium growth in monetary economies 129
Money supply rule Dombusch exchange
rate dynamics
Taylor interest rate rule
Asset
Keynes effect Blanchard equity and
markets
r, rI ,… bond dynamics
Expected medium-run
inflation (Mundell effect)
Wage Labour
Wage price spiral
inflation markets Fiscal policy rules
the Keynesian IS–LM model when integrated with supply side effects and rational
expectations degenerates to a supply side dynamics without effective demand con-
straint, see Sargent (1987, Ch.5), and for a critique of such results Flaschel (1993)
and Flaschel, Franke and Semmler (1997).
The consequences of such a concentration on supply side issues has been
that neither the relevance of the above considered causal nexus, nor the feedback
structure (or repercussions) we have added to it in Figure 6.1 (the arrows leading
back into the middle area) have received much interest. It may indeed be claimed,
that interest in the many stabilizing or destabilizing feedback chains known from
the literature on Keynesian macrostatics or macrodynamics (of the 1960s and the
1970s in particular), was nearly absent in more recent mainstream macrodynamic
model building with their wage–price interactions, up to the use of stabilizing
Keynes- and Pigou-effects, which often simply served the purpose of providing
for the deterministic part of the models a shock-absorber scenario on the basis of
which impulse-propagation mechanisms could be erected and investigated. Even
if the Keynesian demand constraints were given some attention, there was thus
generally the scenario of rapid convergence to full employment steady states unless
shocks continued to hit the economy.
Yet, there are (locally) destabilizing effects of rising inflation and expected
inflation (Mundell-effects), there are destabilizing Metzlerian inventory acceler-
130 Peter Flaschel
ator mechanisms, there is the implication (when aggregate demand depends on
income distribution and the real wage) that either price of wage flexibility must be
destabilizing (normal or adverse Rose-effects), there are destabilizing Fisher debt
deflation effects with respect to investment or consumption behavior (in particular
if there is high debt of firms or certain types of households). There are cumulative
processes in exchange rate and asset market dynamics (where expected and actual
rates of appreciation or depreciation do exercise a positive feedback on each other).
There are therefore a variety of reasons on the basis of which one might form the
expectation that steady states of integrated Keynesian or disequilibrium growth
models are more likely surrounded by centrifugal forces than by centripetal ones,
implying the necessity to introduce additional nonlinearities should the dynamics
depart too much from the steady state due to these forces.
The study of growth with endogenously generated fluctuations, based on the
disequilibrium adjustment processes and resulting feedback chains of the just char-
acterized type and coupled with additional nonlinearities they may give rise to far
off the steady state, thus should remain on the agenda of macrodynamic theoriz-
ing, if only as a (not yet) well-established and well-known alternative scenario
with which achievements of other integrated approaches to monetary growth can
be compared. There is of course always the still fashionable possibility, to model
economic dynamics such that the jump variable technique can be applied, which
by definition removes from sight all instabilities initially present in the dynamics,
so that there remains not much to be compared. Models employing the jump vari-
able technique have however stressed the importance of a treatment of anticipated
future events, a topic that should also be considered and solved by the integrated
disequilibrium growth dynamics we have in mind.
In Figure 6.1 we finally present (top-left and bottom-right) the addition of
policy feedback rules to the considered interaction of a structure of market domi-
nance with the wage–price spiral and the shown feedback structures of the private
sector, yet not necessarily with the understanding that policy can just manipulate
this scenario from the outside, but that there may be interactions with the behav-
ior assumed for the private sector and the type of policy considered to a more or
less significant degree. Policy issues are however not yet well developed in the
approaches we shall review in this chapter and thus only supplement here our
summary of traditional Keynesian model building, or better what it should have
been, but has not been yet, at least from the dynamically fully integrated point of
view.
Our perspective in the following is thus integrated Keynesian disequilibrium
growth analysis as it is obtained, on the one hand, from prototypic models of
fluctuating growth and, on the other hand, of inflation and stagflation. This implies
in a natural way the inclusion of four of the feedback structures just discussed,
with particular stress on the dynamics of wages and prices, but also of quantities
and thus on goods and labor market reaction patterns. Due to space limitations we
arrive here however only at a fundamental or working model type of integrated
disequilibrium growth theory, while further extensions (open economies, financial
markets, policy rules and more) remain for future research.1
Disequilibrium growth in monetary economies 131
In Section 6.2 we will consider supply side dynamics in isolation, yet not one
of the AS variety, but one that extends the basic approaches of Goodwin (1967)
and Rose (1967), here called AC-PC analysis, towards more refined treatments of
wage–price dynamics, and also towards an inclusion of adverse adjustments of
real wages and real debt (in the case of deflation in particular). Section 6.3 will
then consider the textbook IS–LM model augmented by a certain type of wage–
price dynamics (based on Friedman’s views on full employment and the role of
inflationary expectations). We will show there that the resulting dynamical system
is far from being well understood, and that it will give rise more likely to persistent
fluctuations in employment and inflation rates rather than to the shock absorber
behavior that is generally believed to be the outcome here. We also show that the
labor market NAIRU may be a goods market NAIRU in fact.
Section 6.4 considers modern discussions of wage or price Phillips curves and
tries to offer a unifying framework for such discussions (to be integrated into our
systematic evolution of integrated models of disequilibrium growth later on). This
section therefore demonstrates that much remains to done when AC-PC analysis is
extended to include Keynesian goods market dynamics. We also show that a special
case drawn from this general framework of wage–price dynamics may give rise
to an endogenous explanation of the NAIRU which is quite different from the one
that rules the roost in the literature on this rate of (un)employment.
In Section 6.5 we then present and investigate our basic modeling of integrated
disequilibrium growth theory, the core or working KMG model of disequilibrium
growth obtained as endpoint of a systematic discussion of models of monetary
growth of orthodox type in Chiarella and Flaschel (2000b), namely the Tobin type
models, the Keynes–Wicksell model types, the IS–LM growth model (without and
with smooth factor substitution) and IS–LM growth models based on the (incom-
plete) dynamic multiplier story. This section briefly presents theoretical results
on the resulting six-dimensional dynamics (non-existing so far in the literature
on this level of generality). We here study Rose-, Mundell- and Metzler-effects
in their interaction and thus obtain, on the one hand, an already fairly advanced
feedback structure between goods and labor markets (including the conventional
Keynes-effect as short-cut to the interaction with asset markets). In comparison
to Figure 6.1, on the other hand, the working KMG model still provides only a
limited picture of the working of actual economies (on the macrolevel). The reader
is referred to Chiarella and Flaschel (2000h) for the consideration of extensions of
this fundamental integrated disequilibrium growth model towards the other top-
ics included in figure 6.1 and also towards a theoretical penetration of modern
structural macroeconometric model building of the type shown in the figure.
NAIRU V NAGRW ω
V ω
ω V
V ω
V ω
V V
V V
ω ω
ω
• the PC curve exhibits negative values below this rate, and thus implies falling real
wages in this domain,
• the PC shows positive values to the right of the NAIRU, implying rising real wages
on this side.
We thus do not need in the following that the PC, in terms of employment, is
monotonically increasing as is generally assumed in the literature.4
Corresponding to this real wage PC we assume as second building block of
our model an Accumulation Curve (AC) which postulates that the rate of change
134 Peter Flaschel
V̂ of the rate of employment is a function of the level of real wages, again with
a benchmark value, a Non-Accelerating-Growth Rate of Wages or NAGRW ω̄
of the real wage, which separates rising from falling rates of employment. As
before the conclusions of the Goodwin growth cycle analysis do not demand the
monotonicity of the curve shown top right in Figure 6.2.
On the basis of the AC and PC curves shown in Figure 6.2 one gets the ad-
justments of the rate of employment and of real wages as shown in the middle of
this figure. In order to get from that the dynamic consequences for the interaction
of real wages with the rate of employment (bottom right in Figure 6.2) one has
to mirror the implications of the PC part of the model along the 45◦ degree line
(bottom left). The phase space bottom right then simply integrates the neighboring
situations as shown by the arrows in this space. The further implication of this
model, that all curves (in the positive part of the phase space shown) must be
closed orbits, can of course not be proven in this way. In order to get this result
in an intuitively understandable way, one has to consider the following type of
function:
V ω
L(ω, V ) = PC(Ṽ )/Ṽ d Ṽ − AC(ω̃)/ω̃ d ω̃.
V̄ ω̄
The graph of this (Liapunov) function has the form of a global sink (under the
assumptions made) with its minimum at ω̄, V̄ and with all level curves (where the
function assumes a given value) closed. Projected into the ω, V phase space these
closed curves are just the orbits of the considered dynamics, since it is easily shown
that L is constant along the trajectories of the investigated dynamical system (Figure
6.2, bottom right). We add that this proof applies, on the one hand, to very general
situations as far as functional shapes of the PC and AC curves are concerned, but
that it, on the other hand, has to be checked carefully for (and will often not be
applicable to) systems which do not rely on the simple cross-dual nature of the AC-
PC interaction shown in Figure 6.2.5 We thus have that all trajectories generated
by the interaction of PC and AC dynamics represent periodic motions of the real
wage and the rate of employment as the one shown in Figure 6.2, bottom right. We
do not describe this overshooting dynamic in its details here, since this has been
done many times already, including the original article of Goodwin (1967).
NAIRU V ≤ Vp NAGRW ω
V ω
ω Vp
V ω
V ω
V V,V p ?
V V
?
ω ω
ω
Figure 6.3 The Goodwin growth cycle model in perspective (with influences from goods
and asset markets)
Ḋ = pI − ρpK.
This is here to be combined with a money wage PC of the type ŵ = βw (V − V̄ ),
with w the money wage, and an investment equation of the type I /K = α(ρ − ρmin )
with α > 1 and the definition of the pure rate of profits ρ (net of interest payments
rD on loans) given by: pY p − rD − wly Y p /pK. The price level p and the rate of
interest r are considered as given for the time being, as well as the benchmark
rate of profit ρmin . The parameter ly denotes the labor coefficient of the assumed
fixed proportions technology (the other coefficient being given by the potential
output–capital ratio yp = Y p /K). Since this is still a supply side dynamics, actual
output will always equal potential output due to Say’s Law assumed to prevail in
this type of model.
These structural equations of the model give rise to:
where u = w/p · ly denotes the share of wages, where n is the rate of natural growth,
d the debt to capital ratio D/(pK) of firms, and ρ = yp (1 − u) − rd the pure rate
of profit in this supply driven approach to economic growth. The dynamically
endogenous variables are u, V as in the original Goodwin model and a new one,
the ratio d.
Proposition7
Vo = V̄ + n/βw (6.4)
uo = (yp − ρo − rdo )/yp , ρo = ρmin + n/α (6.5)
do = 1 − ρo /n = (α − 1)/α − ρmin /n (6.6)
2. This steady state is not globally asymptotically stable with respect to shocks
of the debt capital ratio d which, when sufficiently large, can lead to an
explosive development of the debt ratio d.
The details of this proposition and its proof are provided in Chiarella, Flaschel
and Semmler (2000) and will not be repeated here. We simply conclude here that
debt financed investment makes the Goodwin growth cycle convergent for small
138 Peter Flaschel
r and for small shocks in its state variable d, but that sufficiently high debt per
unit of capital can make these dynamics divergent ones, implying a situation of
corridor stability in the place of the closed orbit structure of the original Goodwin
(1967) approach.
These dynamics are based on the following static (and linearized) relationship
representing Keynesian goods market equilibrium, here directly expressed in terms
of the rate of capacity utilization Uc of firms:
w w
Uc = Ū + d1 ( − ( )o ) + d2 (d − do ), d1 , d2 ≤ 0
p p
This equation is used here as a shortcut for the delayed feedback chain on the
market for goods to be introduced in Section 6.5 on the KMG model type (and its
richer concept of aggregate demand). We have assumed in this equation that output
and capacity utilization depend negatively on the real wage, based on the particular
view that the negative real wage effect on investment dominates the positive one
on consumption (the orthodox point of view), and have also assumed that output
and capacity utilization depend negatively on the debt ratio d, again because in-
vestment depends negatively on it. The above goods market representation allows
for Rose (1967) type real wage effects of traditional type (where price flexibility
will be destabilizing) and for Fisher debt effects (where price flexibility will also
be destabilizing), but it still excludes Mundell-effects, for example based on the
Disequilibrium growth in monetary economies 139
inflationary expectations mechanism considered in Section 6.3. Finally we have
ρ = y − wp ly y − rd = yp Uc (1 − wp ly ) − rd for the rate of pure profits ρ.
The first two laws of motion for wages w and prices p can be easily derived,
under one additional assumption stated below, from the following wage–price
adjustment equations:
where ly , yp are again the labor coefficient and the potential output–capital ratio
of the fixed proportions technology and where l = L/K denotes the factor endow-
ment ratio of the economy. Note also that the third equation (6.9) now concerns the
evolution of the potential rate of employment V p = ly Y p /L as discussed in connec-
tion with Figure 6.3. The dynamical system (6.7) – (6.10) therefore needs further
140 Peter Flaschel
reformulation in order to make it an autonomous system of dimension 4, since V
now depends on Uc , l and since (6.9) can no longer be used to describe the actual
evolution of the rate of employment as in the earlier treatments of the Goodwin
approach to cyclical growth (where there was always full capacity growth).
A simple reformulation of the dynamics (6.7) – (6.10) is in this regard provided
by making use of the following relationship between the ratios V , V p , Uc :
V = ly Y /L = ly Y p /L · Y /Y p = V p · Uc
by which the rate of employment can be removed from the above 4D dynamics
which are then based on the state variables w, p, V p , d, since Uc has been assumed
to be a function of w, p, d.
The interior steady state of these dynamics in the state variables w, p, V p , d is
characterized by:9
do = 1 − r/n (6.13)
Uco = Ūc (6.14)
Vop = Uco /V̄ (6.15)
ρo = yp Uc (1 − ωo ly ) − rdo = r, ω = w/p (6.16)
1 − (ρo + rdo )/(Uc yp )
ωo = (6.17)
ly
po = determined by initial conditions (6.18)
wo = po ωo (6.19)
Proposition
1. Assume 0 < r < n and that βp , κp , d2 are all sufficiently small. Assume fur-
thermore that the investment parameter α is such that αr − n > 0 holds
true. Then: The steady state (6.13) – (6.19) of the dynamics (6.7) – (6.10) is
locally asymptotically stable.
Disequilibrium growth in monetary economies 141
2. The steady state (6.13) – (6.19) of the dynamics (6.7) – (6.10) is not locally
asymptotically stable for all price adjustment speeds βp chosen sufficiently
large.
3. Assume that nominal wages are completely fixed (βw , κw = 0). Then: The
dynamics (6.7) – (6.10) is monotonically explosive, implying higher and
higher real wages and debt to capital ratios, for all initial debt capital ratios
sufficiently high and all real wage levels above their steady state value.
This model is based on a dynamic theory of effective demand whereby the time
rate of change Ẏ of IS–LM equilibrium output Y is postulated to depend positively
on the rate of change of real balances M /p:
M /p = M̂ − p̂ = µ̄ − π
Ẏ = a1 (µ̄ − π) + a0 f¯
π̇ = βw a1 (µ̄ − π) + βπe βw (Y − Ȳ ) + βw a0 f¯
V = a0 + a1 m + a2 π e , a1 , a2 > 0 (6.21)
m = µ̄ − βw (V − V̄ ) − π e (6.22)
π̇ e
= βπe (π − π ) = βπe βw (V − V̄ )
e
(6.23)
There are always two steady states of the dynamics (6.22) – (6.23), one that is
interior to the right half of the phase plane and thus economically meaningful
and one that lies on its boundary:
The dynamics around the border steady state are always of saddlepath type
(det J < 0), while the dynamics around the interior steady state will be rep-
resented by a stable node, a stable focus, an unstable focus and an unstable
node as the parameter βπe , the adjustment speed of inflationary expectations,
is increased from close to zero to close to infinity.
The proof of these results is simple and is provided in Flaschel and Groh (1998).
These results in particular state that the dynamics are never globally asymptotically
stable and are also not locally asymptotically stable if inflationary expectations
are adjusted with sufficient speed. The monetarist belief in the overall asymptotic
stability of the private sector is therefore not justified in an IS–LM–PC framework,
which at best allows for corridor stability (when the Keynes-effect is sufficiently
strong relative to the Mundell-effect), but not for more, see also Groth (1993) on
Disequilibrium growth in monetary economies 145
this matter. If the steady state is locally explosive it will be globally explosive. The
dynamics therefore are not a viable one in this case and the question must be posed
as to what can make them bounded in such an explosive situation.
Keynes (1936) in fact did provide the basic answer to this question when
stating:
Thus it is fortunate that workers, though unconsciously, are instinctively more
reasonable economists than the classical school, inasmuch as they resist re-
ductions of money-wages, which are seldom or never of an all-round character
. . . (p.14)
The chief result of this policy (of flexible wages, P.F.) would be to cause a
great instability of prices, so violent perhaps as to make business calculations
futile . . . (p.269)
We use the following stylized modification of the PC used so far in order to provide
a mathematical expression for the institutional fact just quoted:
= max{βw (V − V̄ ) + π e , 0}.
w
This Phillips curve says that money wages behave as in the preceding subsec-
tion if their growth rate is positive, but stay constant if they would be falling in the
previous situations. There is thus no wage deflation possible now. This assumed
kink in the money wage PC could be smoothed or some wage deflation could be
allowed for, but this will not alter the conclusions significantly. We consider this
kinked Phillips curve as a much better description of reality than the one that is
linear throughout.14
The immediate consequence of this new form of the Phillips curve is that
system (6.22) – (6.23) now only applies when βw (V − V̄ ) + π e ≥ 0 holds while it
must be replaced by
= µ̄
m (6.24)
π̇ e = −βπe π e (6.25)
in the case βw (V − V̄ ) + π e < 0.15 We thus get a system of differential equations
which is only continuous now, but which can be made a smooth system in an
obvious way. We call this system the patched system while we refer to the earlier
dynamics as the unpatched one.
There are a variety of propositions that can be formulated in the context of such
a patched dynamics, see Flaschel and Groh (1998), but due to space limitations we
will consider here only one of them which describes the outcome of the explosive
case of the Dornbusch and Fischer (1996) model in the patched situation when
there is steady state inflation (µ̄ > 0). This proposition refers to Figure 6.4 which
represents the considered dynamics in the m, π e state space.
S0
.
πe = 0
D0
S
E'
. D1
w=0 W
.
m=0
S1 .
πe = 0 E m
Figure 6.4 Implications of the kinked Phillips curve in the case of steady state inflation
2. Assume that the interior steady state W is locally repelling (for values of
βπe chosen sufficiently large). Then: every trajectory in Do converges to a
persistent cycle around W (and in D1 ).
As this figure shows the domain below the separatrix S of the saddlepoint
S0 in the nonnegative orthant is now an invariant domain Do , i.e. no trajectory
which starts in it can leave it. Note also that the domain below the ẇ = 0 isocline
is governed by the above revised dynamics in place of the one of the preceding
section, which however only alters the direction of the dynamics on the horizontal
axis. This axis is now also an isocline (π̇ e = 0) of the patched dynamics (up to point
E). Note furthermore that the trajectory which starts in E, followed up to point E
and then vertically continued up to the m-axis, also defines an invariant domain
D1 of the patched dynamics which moreover is attracting for all trajectories in the
interior of Do . We thus have that all orbits in Do (with the exception of the ones
on the vertical axis) are either inside of D1 or are entering the domain D1 (from its
left) at some point in time.
We do not go into the proof of this proposition or a deeper explanation of the
phase diagram shown, but refer the reader to the quoted work in this regard. We
simply close this subsection here by stating that the situation in Figure 6.4 is far
away from anything that can be found on IS–LM–PC dynamics in the literature,
and this is simply due to the fact that there is a Mundell effect (in IS–LM), a
simple growth law (of wages) and a fundamental institutional asymmetry in the
PC analysis to be used, which makes the overall dynamics viable up to shocks in
m or π e that go beyond the shown separatrix S.
Disequilibrium growth in monetary economies 147
6.3.3 Reinterpreting the IS–LM–PC dynamics
We now proceed to a reformulation and reinterpretation of the structural equations
underlying the dynamical system of the preceding subsection which do not alter its
mathematical formulation and stability features, but which give the goods market
in the place of the labor market the decisive role in the explanation of the stability
and instability scenarios just discussed. In order to introduce this reformulation of
IS–LM–PC analysis, we start from a stylized representation of empirical results
(testing the conventional NAIRU model) provided by Fair (1997a), and extended
in Fair (1997b). Fair’s (1997a) reconsideration of the structural price and wage
equation leads him to the (here simplified) result that it is in fact the price Phillips
curve which determines the shape of the integrated Phillips curve of the literature,
while wage inflation is following price inflation more or less passively. This implies
that the integrated Phillips curve now refers to demand pressure on the goods
market and not as is customarily assumed on the labor market. In terms of the two
wage and price Phillips curves considered in the preceding section this leads to
their following special reformulation
ŵ = κw p̂ + (1 − κw )π e (6.26)
p̂ = βp (Uc − Ūc ) + κp ŵ + (1 − κp )π e (6.27)
or, if solved as in the preceding section, but now with expected medium-run infla-
tion shown explicitly (no longer equal to zero):
κp
p̂ = βp (Uc − Ūc ) + π e . (6.28)
1 − κw κp
This is the same type of Phillips curve as used in the preceding subsection with
the only (economically seen very important) difference that the rate of capacity
utilization is now used in place of the rate of employment as the measure of demand
pressure that drives price inflation. Formally seen, this Phillips curve can even be
represented exactly as in the preceding subsection (in the situation where labor
supply and capital stock growth are still excluded due to the medium run nature
of the performed analysis), if one makes use of the following implications of our
assumption of a fixed proportions technology:
Y 1 Ld 1 L Ld 1 L
Uc = p
= p = p = p V
Y ly y K ly y K L ly y K
without implying that the demand pressure driving inflation is a labor market
phenomenon and also not that the NAIRU rate c0 /c1 implied by it is related to
labor market issues. In fact the above shows that it is the NAIRU rate of firms’
capacity utilization while there is in fact no labor market NAIRU at work in the
148 Peter Flaschel
present model. This shows that there is the possibility that the literature on the
conventional type of integrated PCs has completely misinterpreted the NAIRU
phenomenon.
On the basis of this reformulation of the (across markets) integrated Phillips
curve the analysis of the preceding subsection can be repeated word by word, with
the interpretational differences just stressed, namely that Ūc determines the steady
state value of the rate of employment and that the destabilizing Mundell-effect is
now basically due to the behavior of firms. This implies that price flexibility is bad
for economic stability in a third way (adding to the Rose adverse real wage effect
and the Fisher debt effect considered in the preceding section), namely through
the so-called Mundell effect, which says that there can be a positive (destabilizing)
feedback mechanism leading from rising inflation to rising expected inflation and
then, via the real rate of interest, to rising aggregate demand and thus to further
inflationary impulses. This holds if the negative Keynes effect on aggregate demand
based on a positive correlation of the nominal price level and the nominal interest
rate is not strong enough to overthrow this cumulative tendency in the interaction
of expected and actual rates of price inflation. Price flexibility must therefore be
regarded with suspicion from at least three different angles, in particular when
it occurs under deflationary pressure, since floors to economic activity are not
so easily established as ceilings (the latter are built into the system via supply
bottlenecks or via monetary and fiscal policy, which may stop accelerating growth,
but which cannot so easily revive declining economic activity).16
A recent discussion concerning the core of practical macroeconomics, in the
papers and proceedings issue of the American Economic Review 1997, provides
numerous statements for and against the scope and relevance for traditional Key-
nesian dynamics, in particular in the applied area, with those against generally
referring to a lack of microfoundations of Keynesian analyses, contrasted to the
progress of the macroeconomic theory of the last two decades. Nevertheless, there
continue to exist observations of the kind:
Right or wrong, the IS–LM model, and its intellectual cousins, the Mundell-
Fleming model and the various incarnations of aggregate supply – aggregate
demand models, have proved incredibly useful at analyzing fluctuations and
the effects of policy.
(Blanchard, 1997, p.245)
Yet, in view of the analysis presented in this subsection and in the preceding one,
it can be claimed that traditional Keynesian analysis is in fact still poorly understood
(or at least represented) even on the textbook level, but also in more advanced types
of analysis, if dynamic issues are addressed. There is no thorough discussion of
the many scenarios even the simple IS–LM–PC model can give rise to, there is no
investigation of the additional instabilities arising from Rose real wage and Fisher
debt effects considered in Section 6.2, nor is there any far-reaching analysis of the
full picture of traditional Keynesian dynamics as sketched in the introduction to
this chapter and as filled with more details in the sections that will follow. There is
Disequilibrium growth in monetary economies 149
no well-documented general analysis of the wage–price spiral, based on demand
pressure and cost-push terms, as we shall present it in the following section, no
detailed study of an integrated theoretical model with sluggish price as well as
quantity adjustments and varying rates of capacity utilization for both labor and
capital as we shall present in Section 6.5 and no such model where endogenous
growth, financial markets, policy feedback rules, and more, are systematically
investigated as to their contribution to the overall behavior of the macroeconomy.
There are however integrated models which address many of these issues,
but not all of them, from an applied perspective. Yet, these empirically motivated
structural macroeconometric models have until recently17 never been analyzed in
detail from the theoretical perspective, since these models are filled with a lot of
empirical details, often not complete with respect to long run aspects, and not
represented in the form of a theoretical reference model in the literature which
would have allowed the analysis of their steady states, their stability and of the
mechanisms that would ensure global boundedness should the steady state be
surrounded by centrifugal forces. Recent structural macroeconometric models tend
to include long run considerations now, but they still continue to believe that
the deterministic part of the dynamics they consider is behaving like a shock
absorber, a dynamic scenario with a long tradition in dynamic economic theory,
but nevertheless only one possibility of many others, in particular in the high order
dynamical systems any integrated macrodynamics will necessarily lead us to.
It can therefore not really be claimed that we do not have models at our disposal
which enrich the early Keynesian quantity dynamics by detailed price dynamics,
asset market behavior, questions of open economies and more, but it is surely true
that these models or better their common theoretical core (removing lags, special
features and pure replication by disaggregation) are not presented and investigated
to any satisfying degree. In view of the above quotation from Blanchard (1997) it
must therefore be added that the usefulness of dynamic IS–LM analysis is to be
regarded as very limited and fragmented in its present state where there are only
more or less isolated examples for such an analysis available, see Turnovsky (1977,
1995) for some left over ruins of this type. Furthermore, there is still the confusion,
see again the above quotation from Blanchard (1997), that AS–AD dynamics, as
for example presented in Sargent (1987, Ch.5), is the model of Keynesian medium
and long run analysis. This, however, cannot be true simply due to the fact that
capital is always fully employed in these approaches, while Keynesian dynamics
should in principle study the reasons for the possible under-employment (or over-
employment) of all factors of production.
We thus conclude that even those authors who show some sympathy for the
traditional Keynesian way of analyzing the macroeconomy do not really describe
what such analyses have been capable of solving and where they are still in their
state of infancy (as far as systematic explorations of the fluctuating growth pat-
terns they can give rise to are concerned). A step closer to the true alternative in
macrodynamic analysis is Barro (1994) when he states:
We have available, at this time, two types of internally-consistent models that
allow for cyclical interactions between monetary and real variables. The con-
150 Peter Flaschel
ventional IS/LM model achieves this interaction by assuming that the price
level and the nominal wage rate are typically too high and adjust only gradually
toward their market-clearing values. The market-clearing models with incom-
plete information get this interaction by assuming that people have imperfect
knowledge about the general price level.
(Barro, 1994, p.4)
It is clearly stated here that both wage and price rigidity and imbalance in the
labor and the goods market are the basic building blocks of the disequilibrium
approach to macrodynamics, which cannot therefore be of the AS–AD variety. Yet
it is not made clear that the analysis of the ‘cyclical interactions between monetary
and real variables’ is not yet very far developed if the scenario presented in the
introduction of this chapter (and more) is really taken seriously. However, it is
admitted by Barro that disequilibrium approaches can be internally consistent.
We conclude this section with the observation that much remains to be done
even on the level of traditional Keynesian IS–LM growth dynamics in order to
obtain a well-understood reference situation against which the achievements of
more recent studies of the dynamic implications of market imperfections, supply
side bottlenecks (and also of perfect market clearing approaches) can be evaluated
and put into perspective. To demonstrate this in more detail will be the topic of the
remaining sections of this chapter, see also Chiarella, Flaschel, Groh and Semmler
(2000).
In Section 6.4 we shall start this discussion with a critical evaluation of the
generality of applied Phillips curve analyses where, as we shall see, numerous
alternatives have been proposed and claimed to be the crucial ones, but where no
unifying approach so far exists which conceives all these studies as special cases
of a general formulation of wage price dynamics based on non-market clearing
in both labor and goods markets. We shall supply in the section such a general
approach as an extension of what we formulated in Section 6.2 and as a detailed
representation of the supply side features of traditional Keynesian theory as sum-
marized in Figure 6.1 of Section 6.1. On the basis of this general framework for
wage price interactions we shall then add to the discussion of Section 6.3 a model
(and theoretical interpretation) of the presented IS–LM–PC analysis of this section
which further contributes to the insight gained, which are, that there is in fact not
a uniquely determined understanding of medium-run IS–LM analysis. The topics
to be solved by the IS–LM–PC approach are thus far from being settled and un-
derstood, in contrast to what is generally declared to be the case in the literature,
where these issues are generally considered as sufficiently treated and on the basis
of this understanding as outdated.
ŵ = βw (·) + κw p̂ + (1 − κw )π e
p̂ = βp (·) + κp ŵ + (1 − κp )π e
and thus represent (when appropriately reordered) two linear equations in the
unknowns ŵ − π e , p̂ − π e that can be uniquely solved for ŵ − π e , p̂ − π e when
κw , κp ∈ [0, 1] fulfill κw κp < 1, giving rise then to:
1
ŵ − π e = [βw (·) + κw βp (·)]
1 − κw κ p
1
p̂ − π e = [βp (·) + κp βw (·)].
1 − κw κ p
Integrating across markets for example the two PCs approach (6.32), (6.33) in
this way, would thus imply that four qualitatively different measures for demand
pressure in the markets for labor as well as for goods have to be used both for money
wage and price level inflation for describing their deviation from expected inflation
in the usual way by an expectations augmented PC, see Laxton et al. (1998) for
a typical example, where, as is customary, only one measure of demand pressure
(on the labor market) is considered. Making furthermore use of Phillips’ (1954)
three types of control, the obtained integrated PCs will be further differentiated,
leading to twelve types of expressions for demand pressure that may appear in the
integrated (across markets) price level PC that rules the roost in the mainstream
literature. Furthermore, as in Section 6.2, two different types of NAIRUs will
then be present in the integrated (wage and) price PC which in general cannot be
identified with each other. Finally, as already mentioned, further differentiation
may concern the cost pressure terms of the PCs shown above, but will not be
considered here in its details.
The stage of wage and price Phillips curves considerations now reached thus
exhibits in each case six different measures of demand pressure in the correspond-
ing PC, which, when transformed into integrated PCs, spanning across markets,
in the way just shown, leads us to the following fairly complex expressions for
expectations augmented PCs:18
1
ŵ = π e + [βw1 (V − V̄ ) + βw2 V̂ + βw3 (V − V̄ )dt
1 − κw κp 0
+ βw4 (V w − 1) + βw5 V̂ w + βw6 (V w − 1)dt
0
154 Peter Flaschel
+ κw (βp1 (Uc − Ūc ) + βp2 Ûc + βp3 (Uc − Ūc )dt
0
As should be obvious now, the second of these equations represents ‘the’ inte-
grated price Phillips curve of this extended approach to wage and price inflation
and its various measures of demand pressure (where the actual wage and price in-
flation cost-push cross reference has been removed by mathematical substitution).
Obviously, this equation is much more complicated than the simple expectations
augmented price Phillips curve of the theoretical literature (or its Walrasian rein-
terpretation as a Lucas supply curve).
Let us briefly consider various applied approaches to PC measurements on the
basis of the equations (6.34), (6.35). Fair (1997a,b), as already shown, provides
one of the rare studies (disregarding structural macroeconometric model building
for the moment) which start from two PCs, though he makes use of βp1 = 0 solely
as far as demand pressure variables are concerned. In his view the price Phillips
curve is therefore the important one.
Concerning, modern macroeconometric model building, we find in Powell and
Murphy (1997) a money wage Phillips curve with βw1 , βw2 = 0 and a price Phillips
curve that appears to be based on cost-push terms solely, but which (when appro-
priately reformulated, see Chiarella, Flaschel, Groh, Köper and Semmler (2000a))
in fact also makes use of βp1 = 1 implicitly. Furthermore, the parameter βw2 is
in their study about eight times larger than βw1 when the nonlinear wage Phillips
curve measured in this work is linearized at the steady state, which supports Kuh’s
(1967) assertion that the wage Phillips curve is a level relationship rather than
one concerning rates of inflation (and which at the same time stresses the impor-
tance of Phillips loops as already observed by Phillips (1958) himself). Indeed,
if ŵ = βw2 V̂ represents the dominant part of the money wage Phillips curve, we
get by integration w = constV βw2 and thus a wage curve as considered on the mi-
crolevel by Blanchflower and Oswald (1994) for example. In this view the wage
Phillips curve, with derivative control solely, is therefore the important one.
Laxton et al. (1998) use for the Multimod mark III model of the IMF an inte-
grated (or hybrid) PC of the type (6.35) with only βw1 = 0, and thus the most basic
type of PC approach, but stress instead the strict convexity of this curve and the
dynamic NAIRU considerations this may give rise to. In their view, therefore, the
Disequilibrium growth in monetary economies 155
wage Phillips curve with proportional term only is the important one. Stock and
Watson (1997) find evidence for a Phillips curve of the type π̇ = βw3 (V − V̄ ), π = p̂,
which shows that this view is in fact based on an integral control in the money
wage Phillips curve (solely) and possibly also on an implicit treatment of inflation-
ary expectations in addition. Roberts (1997) derives a conventional expectations
augmented price Phillips curve from regional wage curves as in Blanchflower and
Oswald (1994) and thus argues that proportional control is relevant in the aggregate
even if derivative control applies to the regional level.
We thus find in this brief discussion of applied approaches a fairly varied set of
opinions, which is, however, not so varied as to pay attention to inside employment
rates and inventory utilization rates and which only in the case of Fair (1997a,b)
takes account of the possibility that demand pressure on the goods market may
be qualitatively and quantitatively different from demand pressure on the labor
market with respect to extent and implications. Otherwise, however, at least the
possibility for proportional, derivative and integral control is taken into account by
this literature (though not reflected and compared in these terms). It must therefore
be noted that the discussion on Phillips curves is at present again a lively one, a still
unsettled one, but also one with still a very limited horizon. Of course, not all of
the expressions shown in (6.35) must be relevant from the empirical point of view,
at all times and in all locations. But this should be the outcome of a systematic
investigation and not the result of more or less isolated views and investigations of
already very specialized types of PCs. Despite the new approaches to PC analysis
it therefore appears as if the analysis and investigation of these curves should start
anew from the extended perspective we have tried to describe above.
Let us close this section by considering a theoretical approach by Rowthorn
(1980) which makes use of a price Phillips curve with proportional control and
a wage Phillips curve with derivative control in order to provide an IS–LM–PC
model, in his case in fact a monetarist model of inflation and stagflation, which is
formally of the same type as the ones considered in the preceding section, but which
allows for an endogenous determination of the NAIRU rates V̄ , Ūc (based on the
conflict over income distribution). This is an interesting extension of the IS–LM–
PC dynamics considered in Section 6.3 and it furthermore provides a theoretical
example on how the use of various special types of Phillips curves (appropriately
combined) can lead to quite different views of the interaction of unemployment
and inflation as compared to the conventional one.19
The fundamental features and building blocks of Rowthorn’s reformulation of
this interaction (here augmented by IS–LM analysis in the place of his simpler
quantity theoretic approach) are the following ones:
p̂ = βp (Π∗ − Π) + π e (6.36)
∗
Π = Π∗ (Uc ) (6.37)
Π = 1 − u, u = (w/p)ly the share of wages (6.38)
ŵ = βw V̂ + p̂ (6.39)
156 Peter Flaschel
We have a price Phillips curve of the proportional kind (based on a kind of
self-reference to price inflation expected to hold over the medium run) and a wage
Phillips curve of the derivative type with myopic perfect foresight as far as price
inflation is concerned. Price inflation is driven by the gap between the desired
profit share Π∗ and the actual one, Π, with the desired profit share being a positive
function of the rate of capacity utilization Uc of firms. In the background of this
model we have our fixed proportions technology with given labor productivity 1/ly
and thus get from this a strict proportionality between the rate of capacity utilization
and the rate of employment, as a very simple form of Okun’s law, Uc = const · V ,
as shown in Section 6.3. Furthermore, the money wage Phillips curve gives rise to
(by its integration): w/p = const V βw , a functional form that then also applies to
the wage share u in the place of the real wage w/p. Inserting all these expressions
into the price Phillips curve p̂ = βp (Π∗ − Π) + π e gives rise to
p̂ = βp (Π∗ (V ) − (1 − u(V ))) + π e = βp (Π∗ (V ) + u(V ) − 1) + π e
with both Π∗ , u being strictly increasing functions of the rate of employment V .
On the surface this is just an ordinary PC of the monetarist type (as we have
employed in Section 6.3), though now possibly a nonlinear one. Disregarding
this latter possibility and assuming that parameters are such that there is a solution
V̄ ∈ (0, 1) where Π∗ (V̄ ) + u(V̄ ) = 1 holds (which is then uniquely determined), we
then get from these alternative underpinnings of the IS–LM–PC model analyzed in
Section 6.3 an endogenous explanation of the NAIRU rate of employment, which
was there given as a parameter. This NAIRU rate, among others, now depends on
the relationship Π∗ (Uc ), and thus negatively on the steepness of this curve (which
characterizes the strength with which capital owners defend their income shares)
and also negatively on the parameter βw which measures the strength with which
labor defends its income share. Therefore, the stronger the conflict over income
distribution, the lower is the NAIRU rate of employment at which the income
shares demanded, Π∗ (V ) + u(V ), become compatible with what is available for
distribution, thereby allowing for a steady behavior of wage and price inflation. This
provides in simple terms a simultaneous interpretation of both the NAIRU rate of
capacity utilization and the NAIRU rate of employment, surrounded by dynamics
that are of the same type as the one investigated in the preceding section. We stress
again that this has become possible through a simple specialization of the very
general type of PCs we have investigated in the present section. Note however that
Okun’s law, which is based on a positive correlation between the rate of capacity
utilization and the rate of employment, has been used here in order to derive this
specific view on the explanation of steady state rates of factor utilization.
We conclude this section with the observation that much remains to be done in
the theoretical discussion of the form and the implications of PC approaches, where
many more outcomes may be obtained than is generally believed. The same holds
true for empirical studies of Phillips curves, where there is a lack of systematic
investigation of the wealth of possibilities our extended presentation of the wage
price module of integrated dynamical models of Keynesian or other variety can give
rise to. Furthermore, what has been discussed here for PCs can also be applied to the
Disequilibrium growth in monetary economies 157
ACs considered in Section 6.2, see Figure 6.3, where derivative control terms for
the impact of capacity utilization on capital stock and employment growth would
introduce Harrodian accelerator aspects into the growth cycle there considered and
where integral terms (for profitability) would represent one possibility to introduce
medium run aspects into such AC analysis.
ω̂ = ŵ − p̂
1
= [(1 − κp )βw (V − V̄ ) − (1 − κw )βp (Uc − Ūc )] (6.40)
1 − κw κp
which simply states that the adjustment of real wages depends positively on the
demand pressure on the market for labor and negatively on that in the market for
goods. We immediately realize that either wage or price flexibility should bring
instability to the dynamics of this section (normal or abnormal Rose effects),
depending on whether real wage increases increase or decrease economic activity.
The next law of motion concerns the dynamics of the factor endowment ratio
l = L/K, where it is assumed that labor supply L grows with the given natural rate
Disequilibrium growth in monetary economies 159
n and the capital stock (in our Keynesian approach) with the rate of net investment
K̂ = I /K. The essential element in this law of motion is therefore given by the
investment function which is specified as follows:
This function (as all other behavioral equations) is assumed as linear (just as our
simple production function Ld = ly Y , Y p = yp K, ly , yp is given magnitudes) in order
to keep the model first as linear as is possible which allows us to concentrate on
its intrinsic or unavoidable nonlinearities in the beginning of the analysis of KMG
growth dynamics. Note also that the trend term in this investment equation is given
by the natural rate of growth.
Investment per unit of capital thus depends on the expected rate of profit ρe ,
the real rate of interest r − π e and the rate of capacity utilization Uc , representing
Tobin’s q and the capacity effect considered by Malinvaud (1980) and others. All
these magnitudes are actual (short run) values and should be replaced by medium
run averages in applications of this model type, which increases the dimension of
the dynamics without adding too much new structure to the model, see Flaschel,
Gong, and Semmler (1999) for such extensions and applications of the model.
Here, however, only the above simple formulation of investment behavior will be
used and it gives rise to the following law of motion for the labor intensity l = L/K :
l = −i1 (ρe − (r − π e )) − i2 (Uc − Ūc ). (6.41)
The next dynamical law is basically equation (6.12) of Section 6.2, since it is
based on the definitional equation for real balances (per unit of capital now):
m = M /(pK), implying m̂ = µ̄ + n − l̂ − p̂, where it is again assumed that the
money supply M grows with a given rate µ̄ (since policy questions are not of
interest here). Making use of the equation (6.12) for p̂ this gives:
1
m = µ̄ − n + l̂ − π e − [βp (Uc − Ūc ) + κp βw (V − V̄ )] (6.42)
1 − κw κp
We see that increased capacity utilization on both the labor and the goods market
will speed up inflation and thus reduce the growth rate of real balances, leading to
corresponding nominal interest rate changes due to the Keynes-effect as it derives
from the simple LM-curve still present in this model type.
Next we have the law for inflationary expectations which is a simple extension
of the one used in Section 6.3, since we now determine these expectations as an
average of backward and forward looking behavior (time series methods and fore-
casts by means of small theoretical models). Time series methods can in principle
be as complicated and refined as possible, when only numerical simulations of the
model are intended. From the viewpoint of theory they should at first be chosen to
be as simple as possible in order to allow for an analytical treatment of stability
issues as we shall provide below, i.e. they will be of the simple adaptive type made
use of already in Section 6.3 (they can be made a humped shaped average of past
160 Peter Flaschel
observations of inflation by means of nested adaptive expectation schemes in a
next step for example). Forward looking expectations can be based on the p-star
concept of the FED and the German Bundesbank for example, which says that
inflation rates will converge to the difference of µ̄, the growth rate of the money
supply, and Ŷ p , the growth rate of potential output (as long as the velocity of money
can be considered a given magnitude). Made as simple as possible again this shows
that inflationary expectations of this type assume that there is convergence of these
actual inflation rates back to the steady rate µ̄ − n giving rise to:
where α denotes the weight attached to the backward looking type of expectations
and 1 − α the one for the forward looking type. The destabilizing role of the
Mundell effect is clearly visible in this extended equation, since economic activity
depends positively on expected inflation (due to the assumed investment behavior)
and since an increase in economic activity, here measured by two rates of factor
capacity utilization, speeds up the increase in inflationary expectations as shown
by this equation.
There remains the quantity adjustment process on the market for goods which is
driven by the adjustment of sales expectations and the changes in actual inventories,
ye = Y e /K and ν = N /K, both already in per unit of capital form here, see Metzler
(1941) for the original approach. The two laws of motion for these variables read:
where the terms involving l are simply due to the fact that everything is expressed
in per unit of capital terms. Sales expectations of firms, ye , are here assumed to
change in an adaptive fashion, following actual demand yd with some time delay,
while actual inventories changes are given by definition through the difference
between actual output y and actual demand yd , again corrected by a term that takes
account of the intensive form under consideration.
This closes the description of the laws of motion of the state variables of our
basic KMG dynamics which concern income distribution, relative factor growth,
inflation as measured by the change of real balances, inflationary expectations,
sales expectations and actual inventory changes. These dynamical laws do not yet
form a complete system, but must be supplemented by some algebraic equations
which define the statically endogenous magnitudes we used in the above differen-
tial equations. They are given by:
Proposition
1. Assume sufficiently sluggish adjustment for wages, prices, and inflationary
expectations, and a strong Keynes-Effect (h2 small). Then:
The interior steady state of the 6D dynamics (6.40) – (6.45), which is easily
calculated and uniquely determined, is locally asymptotically stable for all
adjustment speeds of sales expectations βye chosen sufficiently large and
speeds of adjustments of inventories, βn , chosen sufficiently low.
162 Peter Flaschel
2. The 6D determinant of the Jacobian of the dynamics at the steady state is
always positive.
3. If βπe , βn , h2 are chosen sufficiently large, the steady state equilibrium be-
comes locally repelling. The system therefore undergoes a (generally non
degenerate) Hopf bifurcation at intermediate value of these (and other) pa-
rameters, which generates persistent fluctuations, that are attractors in the
supercritical case and repellers in the subcritical case.
The periodic fluctuations obtained in this way integrate the growth cycle anal-
ysis of Section 6.2 with the inflationary dynamics of Section 6.3, coupled with
Metzlerian quantity adjustment in the market for goods and they are generated in-
dependently of any kink in the money wage PC. Further details on this proposition
are provided in Chiarella and Flaschel (2000b). We do not go into a proof of this
proposition here, but simply add some explanations to the assertions made. The
steady state of the system is asserted to be locally attracting for all price and quan-
tity adjustment speeds (including the Metzlerian inventory accelerator) sufficiently
low (up to sales expectations which mirror the stable Keynesian multiplier dynam-
ics which improve stability if chosen sufficiently large). In addition we should
have a fairly interest-inelastic money demand function in order to produce large
positively correlated swings of the nominal rate of interest when the price level
rises or falls. Partial insights on the stability of Keynesian dynamics (augmented
with what we know for the Rose effects considered in Section 6.2) or even static
conclusions of Keynesian theory, appropriately combined, thus allow here for a
stability assertion for the full 6D dynamics of the integrated KMG growth model.
Furthermore, since the determinant does not change sign when the parameters of
the model are changed, we know that loss of stability can only occur in a cyclical
fashion since eigenvalues must then cross the imaginary axis excluding 0 (and will
generally do so with positive speed). The resulting situations of Hopf bifurcations
then generally imply that this change in stability of the system is accompanied by
either the ‘birth’ of a stable limit cycle (with increasing amplitude) to the right
of the critical bifurcation value (where a pair of eigenvalues has become purely
imaginary) or the death of an unstable limit cycle (via its shrinking amplitude and
the disappearance of a ‘stable corridor’) to the left of this critical point when this
point is approached.
The above brief considerations of KMG growth dynamics must here suffice to
indicate what results might be expected from an integrated AC-PC analysis with
Keynesian supply rationing and with also sluggish quantity adjustment processes
of Metzlerian type. These topics are further pursued in particular in Chiarella and
Flaschel (2000b) and Chiarella, Flaschel, Groh and Semmler (2000).
Notes
1 See Chiarella and Flaschel (2000h) for the full details of what is sketched in this chapter
and in the concluding section in particular.
2 continuity of this curve is of course assumed in addition.
3 Note that we reinterpret the NAIRU of the literature here in terms of the rate of
employment (or utilization) V of the labor force, not in terms of unemployment.
4 Note that a real wage PC is obtained from the conventional money wage PC (augmented
by inflationary expectations of course) by assuming myopic perfect foresight with
regard to the expected rate of inflation.
5 See Flaschel, Franke and Semmler (1997) for the consideration of cross-dual macro-
dynamics on various levels of generality.
6 See also Flaschel (1984) in this regard.
7 We assume that the parameters of the model are such that both uo and do are positive.
8 See Chiarella, Flaschel amd Semmler (2000) for a much more general treatment of
such growth dynamics.
9 We assume again that parameters are chosen such that all steady state values are
meaningful.
10 These authors employ a special discrete time version of the following model, a differ-
ence which however is not essential for our following discussion of their model.
11 Note that the real wage is constant in this approach which is therefore clearly comple-
mentary to one in the preceding section.
Disequilibrium growth in monetary economies 165
12 which in principle should be well-known from Tobin (1975) and subsequent work, but
which is still unfamiliar, see Groth (1993).
13 Note that – though globally asymptotically stable – the model is still incomplete since
the right half of the phase plane is not an invariant set of this dynamics, i.e. output
can be become negative along trajectories that start in an economically meaningful
domain.
14 See Laxton, Rose and Tambakis (1997) for an empirical discussion of the kind of
nonlinearity that may be involved in the integrated price level – rate of unemployment
Phillips curve.
15 The two systems are identical at the border line ŵ = βw (V − V̄ ) + π e = 0.
16 See here also Flaschel (1994) and Flaschel and Franke (1996,98).
17 See for example Barnett and He (1998) for an exception.
18 Note that π e was set equal to zero in the models considered in Section 6.2.
19 See also Flaschel (1993) and Flaschel and Groh (1996) in this regard.
References
Barnett, W. and Y. He (1998) ‘Bifurcations in continuous-time macroeconomic systems’.
Washington University in St. Louis: Mimeo.
Barro, R. (1994) ‘The Aggregate-Supply Aggregate-Demand Model’. Eastern Economic
Journal, 20, 1–6.
Blanchard, O.J. (1981) ‘Output, the stock market, and interest rates’. American Economic
Review, 71, 132–43.
Blanchard, O. (1997) ‘Is there a core of usable macroeconomics?’ American Economic
Review. Papers and Proceedings, 87, 244–6.
Blanchflower, D.G. and A.J. Oswald (1994) The Wage Curve. Cambridge, MA: The MIT
Press.
Chiarella, C. and P. Flaschel (1999) ‘Keynesian monetary growth in open economies’.
Annals of Operations Research, 89, 35–59.
Chiarella, C. and P. Flaschel (2000a) ‘The emergence of complex dynamics in a ‘natu-
rally’ nonlinear integrated Keynesian model of monetary growth’. In: W. A. Barnett, C.
Chiarella, S. Keen, R. Marks and H. Schnabl (eds) Commerce, Complexity, and Evo-
lution: Topics in Economics, Finance, Marketing and Management. Proceedings of the
Twelfth Symposium on Economic Theory and Econometrics. Cambridge: Cambridge
University Press.
Chiarella, C. and P. Flaschel (2000b) The Dynamics of Keynesian Monetary Growth: Macro-
foundations. Cambridge: Cambridge University Press.
Chiarella, C. and P. Flaschel (2000c) “‘High order” disequilibrium growth dynamics: The-
oretical aspects and numerical features’. Journal of Economic Dynamics and Control,
24, 935–63.
Chiarella, C. and P. Flaschel (2000d) ‘Towards applied disequilbrium growth theory: I. The
starting theoretical model’. Discussion Paper: University of Technology, Sydney.
Chiarella, C. and P. Flaschel (2000e) ‘Towards applied disequilbrium growth theory: II.
Intensive form and steady state analysis of the model’. Discussion Paper: University of
Technology, Sydney.
Chiarella, C. and P. Flaschel (2000f) ‘Towards applied disequilbrium growth theory: III.
Basic partial feedback structures and stability issues’. Discussion Paper: University of
Technology, Sydney.
Chiarella, C. and P. Flaschel (2000g) ‘Applying disequilibrium growth theory. II. Housing
cycles and debt deflation’. Mimeo: University of Bielefeld.
166 Peter Flaschel
Chiarella, C. and P. Flaschel (2000h) ‘Disequilibrium growth theory: Foundations, synthesis,
perspectives’. Discussion paper: University of Techonology, Sydney.
Chiarella, C., P. Flaschel, R. Franke (2001) Integrated Macroeconomic Dynamics: Foun-
dations, Extensions, Perspectives. Book manuscript, in preparation.
Chiarella, C., P. Flaschel, R. Franke and T. Lux (2001) Open Economy Macrodynamics:
The Keynes–Metzler–Goodwin Approach, (forthcoming). Heidelberg: Springer Verlag.
Chiarella, C., P. Flaschel, G. Groh, C. Köper and W. Semmler (2000a) ‘Towards applied dis-
equilibrium growth theory: VI. Substitution, money-holdings, wealth-effects and further
extensions’. Discussion paper: University of Technology, Sydney.
Chiarella, C., P. Flaschel, G. Groh, C. Köper and W. Semmler (2000b) ‘Towards applied
disequilibrium growth theory: VII. Intensive form analysis in the case of substitution’.
Discussion Paper: University of Technology, Sydney.
Chiarella, C., P. Flaschel, G. Groh, W. Semmler (2000) Disequilibrium, Growth and Labor
Market Dynamics: Macro Perspectives. Heidelberg: Springer Verlag.
Chiarella, C., P. Flaschel, W. Semmler (2001) ‘The macrodynamics of debt deflation’. In:
R. Bellofiore and P. Ferri (eds) Financial Fragility and Investment in the Capitalist
Economy Aldershot: Edward Elgar .
Chiarella, C., P. Flaschel, W. Semmler (2001) ‘Price flexibility and debt dynamics in a high
order AS–AD model’ To appear: Central European Journal of Operations Research.
Chiarella, C., P. Flaschel and P. Zhu (2000a) ‘Towards applied disequilibrium growth theory:
IV. Numerical investigations of the core 18D model’. Discussion Paper: University of
Technology, Sydney.
Chiarella, C., P. Flaschel and P. Zhu (2000b) ‘Towards applied disequilibrium growth theory:
V. Extrinsic nonlinearities and global boundedness’. Discussion Paper: University of
Technology, Sydney.
Chiarella, C., P. Flaschel and P. Zhu (2000c) ‘Towards applied disequilibrium growth theory:
VIII. A 22D core dynamics in the case of substitution’. Discussion Paper: University of
Technology, Sydney.
Chiarella, C., P. Flaschel, P. Zhu (2000d) ‘Towards applied disequilibrium growth theory:
IX. Global boundedness in the case of substitution’. Discussion Paper: University of
Technology, Sydney.
Desai, M. (1973) ‘Growth cycles and inflation in a model of the class struggle’. Journal of
Economic Theory, 6, 527–45.
Dornbusch, R. and S. Fischer (1996) Macroeconomics. New York: McGraw-Hill (7th edi-
tion).
Fair, R. (1997a) ‘Testing the NAIRU model for the United States’. Yale University: Mimeo.
Fair, R. (1997b) ‘Testing the NAIRU model for 27 countries’. Yale University: Mimeo.
Fisher, I. (1933) ‘The debt-deflation theory of great depressions’. Econometrica, 1, 337–57.
Flaschel, P. (1984) ‘Some stability properties of Goodwin’s growth cycle model’. Zeitschrift
für Nationalökonomie, 44, 63–9.
Flaschel, P. (1993) Macrodynamics. Income Distribution, Effective Demand and Cyclical
Growth. Frankfurt/M.: Verlag Peter Lang.
Flaschel, P. (1994) ‘A Harrodian knife-edge theorem for the wage–price sector?’ Metroe-
conomica, 45, 266–78.
Flaschel, P. and R. Franke (1996) ‘Wage flexibility and the stability arguments of the Neo-
classical Synthesis’. Metroeconomica, 47, 1–18.
Flaschel, P. and R. Franke (2000) ‘An old-Keynesian note on destabilizing price flexibility’.
Review of Political Economy, 12, 273–83.
Disequilibrium growth in monetary economies 167
Flaschel, P., Franke, R. and W. Semmler (1997) Dynamic Macroeconomics: Instability,
Fluctuations and Growth in Monetary Economies. Cambridge, MA: The MIT Press.
Flaschel, P., Gong, G. and W. Semmler (1999) ‘A Keynesian based econometric framework
for studying monetary policy rules’. University of Bielefeld: Mimeo.
Flaschel, P. and G. Groh (1996) Keynesianische Makroökonomik. Unterbeschäftigung, In-
flation und Wachstum. Heidelberg: Springer.
Flaschel, P. and G. Groh (1998) ‘Textbook Stagflation Theory and Beyond’. University of
Bielefeld: Discussion Paper.
Goodwin, R.M. (1967) ‘A growth cycle’. In: C.H. Feinstein (ed.): Socialism, Capitalism
and Economic Growth. Cambridge: Cambridge University Press, 54–8.
Groth, C. (1993) ‘Some unfamiliar dynamics of a familiar macro model’. Journal of Eco-
nomics, 58, 293–305.
Hicks, J. (1937) ‘Mr Keynes and the Classics’, Econometrica, 5, 147–59.
Keen, S. (1999) ‘The nonlinear economics of debt deflation’. In: W.A. Barnett, C. Chiarella,
S. Keen, R. Marks, and H. Schnabl (eds): Commerce, Complexity and Evolution: Topics
in Economics, Finance, Marketing and Management. Proceedings of the Twelfth Inter-
national Symposium in Economic Theory and Econometrics. Cambridge: Cambridge
University Press.
Keynes, J.M. (1936) The General Theory of Employment, Interest and Money. New York:
Macmillan.
Kuh, E. (1967) ‘A productivity theory of the wage levels – An alternative to the Phillips
curve’. Review of Economic Studies, 34, 333–60.
Laxton, D., D. Rose and D. Tambakis (1997) ‘The U.S. Phillips curve: The case for asym-
metry’. Mimeo.
Laxton, D., P. Isard, H. Faruquee, E. Prasad and B. Turtelboom (1998) MULTIMOD Mark
III. The core dynamic and steady-state models. Washington, DC: International Monetary
Fund.
Malinvaud, E. (1980) Profitability and Unemployment. Cambridge: Cambridge University
Press.
Metzler, L. A. (1941) ‘The nature and stability of inventory cycles’. Review of Economic
Statistics, 23, 113–29.
Orphanides, A. and R. Solow (1990) ‘Money, inflation and growth’. In: B. Friedman and F.
Hahn (eds): Handbook of Monetary Economics. Amsterdam: North Holland, 223–61.
Phillips, A.W. (1954) ‘Stabilisation Policy in a Closed Economy’. The Economic Journal,
64, 290–323.
Phillips, A.W. (1958) ‘The relation between unemployment and the rate of change of money
wage rates in the United Kingdom, 1861–1957’. Economica, 25, 283–99.
Powell, A. and C. Murphy (1997) Inside a Modern Macroeconometric Model. A Guide to
the Murphy Model. Heidelberg: Springer.
Roberts, J.M. (1997) ‘The wage curve and the Phillips curve’. Washington: Board of Gov-
ernors of the Federal Reserve System.
Rose, H. (1967) ‘On the non-linear theory of the employment cycle’. Review of Economic
Studies, 34, 153–73.
Rose, H. (1990) Macroeconomic Dynamics. A Marshallian Synthesis. Cambridge, Mass.:
Basil Blackwell.
Rowthorn, B. (1980) Capitalism, Conflict and Inflation. London: Lawrence and Wishart.
Sargent, T. (1987) Macroeconomic Theory. New York: Academic Press.
Solow, R. and J. Stiglitz (1968) ‘Output, employment and wages in the short-run’. Quarterly
Journal of Economics, 82, 537–60.
168 Peter Flaschel
Solow, R. (1990) ‘Goodwin’s growth cycle: Reminiscence and rumination’. In: K.Velupillai
(ed.): Nonlinear and Multisectoral Macrodynamics. London: Macmillan, 31–41.
Stock, J.H. and M.W. Watson (1997) ‘Business cycle fluctuations in US macroeconometric
time series’. Harvard / Princeton University: Mimeo.
Tobin, J. (1975) ‘Keynesian models of recession and depression’. American Economic
Review, 65, 195–202.
Turnovsky, S. (1977) Macroeconomic Analysis and Stabilization Policies. Cambridge: Cam-
bridge University Press.
Turnovsky, S. (1995) Methods of Macroeconomic Dynamics Cambridge, MA: MIT Press.
7 Schumpeterian dynamics
A disequilibrium theory of
long run profits
Katsuhito Iwai
7.1 Introduction
The subtitle of this chapter may sound a contradiction in terms. In the traditional
economic theory, by which I include both classical and neoclassical economics, the
long-run state of an economy is an equilibrium state and the long-run profits (if they
ever exist) are equilibrium phenomena. Figure 7.1 illustrates this by drawing two
supply curves that can be found in any textbook of economics. In the upper panel
is an upward-sloping supply curve which aggregates diverse cost conditions of
the existing firms in an industry. Its intersection with a downward-sloping demand
curve determines an equilibrium price, which in turn determines the amount of
profits (represented by the shaded triangle) accruing to the industry as a whole. As
long as the supply curve is upward-sloping, an industry is able to generate positive
profits.
In traditional theory, however, this is merely a description of the ‘short-run’state
of an industry. Whenever there are positive profits, existing firms are encouraged
to expand their productive capacities and potential firms are induced to enter the
industry, both making the supply curve flatter and flatter. This process will continue
until the industry supply curve becomes totally horizontal, thereby wiping out any
opportunity for positive profits. The lower panel of Figure 7.1 describes this ‘long-
run’ state of the industry.
This implies that if there are any profits in the long run, it must be the ‘normal’
profits which have already been incorporated into cost calculations. In fact, it is how
to explain the fundamental determinants of these normal profits which divides the
traditional economic theory into classical and neoclassical approaches. Classical
economics (as well as Marxian economics) has highlighted an inverse relationship
between the normal profit rate and the real wage rate, and reduced the problem
of determining the former to that of determining the latter and ultimately to that
of distributional conflicts between classes. Neoclassical economics has identified
the normal profit rate with the interest rate plus a risk premium and reduced the
problem of its determination to that of characterizing equilibrium conditions for
intertemporal resource allocation under uncertainty. But no matter how opposed
their views might appear over the ultimate determinants of normal profits, they
170 Katsuhito Iwai
Price
Quantity
0
Price
normal profit
Quantity
0
Figure 7.1 Industry supply curve in the short run and in the long run
share the same ‘equilibrium’ perspective on long-run profits – any profits in excess
of the normal rate are ‘disequilibrium’ phenomena which are bound to disappear
in the long run.
It is Joseph Schumpeter who gave us a powerful alternative to this deep-rooted
‘equilibrium’ tradition in the theory of long-run profits. According to Schumpeter,
it is through an ‘innovation’or ‘doing things differently’that positive profits emerge
in the capitalist economy. ‘The introduction of new commodities . . . , the techno-
logical change in the production of commodities already in use, the opening-up of
new markets or of new sources of supply, Taylorization of work, improved handling
of material, the setting-up of new business organizations.’ (Schumpeter (1939), p.
84) etc. allow the innovators to charge prices much higher than costs of production.
Profits are thus the premium put upon innovation. Of course, the innovator’s cost
advantage does not last long. Once an innovation is successfully introduced into
the economy, ‘it becomes much easier for other people to do the same thing’.1 A
subsequent wave of imitations soon renders the original innovation obsolete and
Schumpeterian dynamics 171
gradually wears out the innovator’s profit rate. In the long run, there is therefore
an inevitable tendency towards classical or neoclassical equilibrium which does
not allow any positive profits in excess of the normal rate. And yet Schumpeter
argued that positive profits will never disappear from the economy because capi-
talism is ‘not only never but never can be stationary’. It is an ‘evolutionary process’
that ‘incessantly revolutionalizes the economic structure from within, incessantly
destroying an old one, incessantly creating a new one’.2 Indeed, it is to destroy
the tendency towards classical or neoclassical equilibrium and to create a new in-
dustrial disequilibrium that is the function the capitalist economy has assigned to
those who carry out innovations. ‘Surplus values [i.e. profits in excess of normal
rate] may be impossible in perfect equilibrium, but can be ever present because that
equilibrium is never allowed to establish itself. They may always tend to vanish
and yet be always there because they are incessantly recreated’.3
It is the first objective of this chapter to formalize this grand vision of Joseph
Schumpeter from the perspective of evolutionary economics.4 It makes use of a
simple evolutionary model of Iwai (1984a, b) and demonstrates the Schumpeterian
thesis that profits in excess of normal rate will never disappear from the economy
no matter how long it is run. Indeed, it will be shown that what the economy will
approach over a long passage of time is not a classical or neoclassical equilibrium
of uniform technology but (at best) a statistical equilibrium of technological dis-
equilibria which reproduces a relative dispersion of efficiencies among firms in
a statistically balanced form. Although positive profits are impossible in perfect
equilibrium, they can be ever present because that equilibrium is never allowed to
establish itself.
This chapter is organized as follows. After having set up the static structure
of an industry in Section 7.2, the following three sections will develop an evolu-
tionary model of industrial dynamics and examine how the firms’ capacity growth,
technological imitations and technological innovations, respectively, move the in-
dustry’s state of technology over time. It will be argued that while both the dif-
ferential growth rates among different efficiency firms and the diffusion of better
technologies through imitations push the state of technology towards uniformity,
the punctuated appearance of technological innovations disrupts this equilibriating
tendency. Section 7.6 will then turn to the long-run description of the industry’s
state of technology. It will indeed be shown that over a long passage of time these
conflicting microscopic forces will balance each other in a statistical sense and
give rise to a long-run distribution of relative efficiencies across firms. This long-
run distribution will in turn allow us to deduce an upward-sloping long-run supply
curve in Section 7.7. The industry is thus capable of generating positive profits
even in the long run! Hence, the subtitle of this chapter – ‘a disequilibrium the-
ory of long-run profits’. Section 7.8 will then examine the factors determining the
long-run profit rate of the industry.
The present chapter will adopt the ‘satisficing’ principle for the description
of firms’ behaviors – firms do not optimize a well-defined objective function but
simply follow organizational routines in deciding their growth, imitation and in-
novation policies.5 Indeed, the purpose of the penultimate Section 7.9 is to show
172 Katsuhito Iwai
that our evolutionary model is able to ‘calibrate’ all the macroscopic characteris-
tics of the neoclassical growth model without having recourse to the neoclassical
assumption of fully optimizing economic agents. If we look only at the aggregative
performance of our evolutionary economy, it is as if aggregate labor and aggregate
capital together produce aggregate output in accordance with a well-defined ag-
gregate production function with Harrod-neutral technological progress. Yet, this
macroscopic picture is a mere statistical illusion. If we zoomed into the microscopic
level of the economy, what we would find is the complex and dynamic interactions
among many a firm’s capital growth, technological imitations and technological
innovations. It is simply impossible to group these microscopic forces into a move-
ment along an aggregate production function and a shift of that function itself. The
neoclassical growth accounting may have no empirical content at all.
Section 7.10 concludes the chapter.
c
0 cn cn–1 …… c2 c1
by Kt (c1 ), but will be denoted simply as Kt in the following discussion. Next, let
st (ci ) and St (ci ) represent the ‘capital share’ and the ‘cumulative capital share’ of a
unit cost ci at time t. Of course, we have st (ci ) ≡ kt (ci )/Kt and St (ci ) ≡ Kt (ci )/Kt .
As a convention, we set St (c) = St (ci ) for ci ≤ c < ci−1 . Figure 7.2 exhibits a typical
distribution of cumulative capital shares in the industry. It illustrates the ‘state of
technology’ at a point in time by showing us how technologies with diverse unit
costs are distributed among capital stocks of an industry.
The state of technology thus introduced, however, represents merely the pro-
duction ‘possibility’of an industry. How this possibility is actualized depends upon
the price each firm is able to obtain in exchange for its product. Let us assume
that the industry in question is a competitive industry in which a large number of
firms are producing the same homogeneous product and charge the same price for
it.7 Let us denote by Pt the product price (measured in terms of money wage) at
time t. Then, under the assumptions of homogeneous product and fixed proportion
technology, firms with unit costs strictly smaller than Pt decide to produce up to
their productive capacity k/b, and firms whose unit costs are strictly higher than Pt
decide to quit all production. Firms with the unit cost equal to Pt are indifferent to
their production level, as long as it does not exceed their productive capacity. (We
ignore here the cost of shutting-down of a factory as well as the cost of setting-up
of a new production line.)
It follows that when ci−1 > Pt > ci the total supply of the industry product
becomes equal to Kt (ci )/b and that when Pt = ci it takes any value from Kt (ci+1 )/b
to Kt (ci )/b. Hence, if we denote by Yt (P) the industry’s ‘short-run supply curve’
174 Katsuhito Iwai
Pt
c1
cn
yt
0 St (cn ) St (cn –1 ) …… St (c2 ) St (c1 ) = 1
Hypothesis (CG): The capital growth rate of a firm with unit cost ci is linearly
increasing in its current rate of profit rt (ci ), or it is equal to:
This hypothesis needs little explanation. It merely says that a higher profit rate
on the existing capital stocks stimulates capital accumulation, either by influencing
the expected profitability of new investment projects or by directly providing an
internal fund for the projects. The parameter γ (or, more precisely, γ/b) represents
the sensitivity of the firm’s growth rate to the current profit rate, and the parameter
γ0 represents the rate of capital depreciation of the break-even firm. As I have
already indicated in Section 7.1, the present chapter follows the strict evolutionary
perspective in supposing that firms do not optimize but only ‘satisfice’ in the sense
that they simply follow organizational routines in deciding their growth, imitation
and innovation policies. Indeed, one of the purposes of this chapter is to see how
far we can go in our description of the economy’s dynamic performance without
relying on the assumptions of individual optimality. I will therefore assume that
the values of γ and γ0 are both exogenously given.10
We have already assumed that every firm in the industry produces the same
homogeneous product and faces the same price Pt . If we further assume that
the price of capital equipment is proportional to Pt , we can calculate the profit
rate r(ci ) ≡ (Pt yt − ci lt )/Pt kt as b(Pt − ci )/Pt , which we will approximate as
b(log Pt − log ci ) for analytical convenience. Then, by simply differentiating the
cumulative capacity share St (ci ) with respect to time, Hypothesis (CG) allows us to
deduce the following set of differential equations for the dynamics of cumulative
capital shares11
Ṡt (ci ) = γδt (ci )St (ci )(1 − St (ci ))(i = n, n − 1, . . . , 1). (7.6)
176 Katsuhito Iwai
In the above equations, δt (ci ) represents the difference between the logarithmic
average of a set of unit costs higher than ci and the logarithmic average of a set of
unit costs not higher than ci , or:
i−1
(log cj )st (cj )
n
(log cj )st (cj )
δt (ci ) ≡ − > 0. (7.7)
j=1
1 − St (ci ) j=i
St (ci )
Its value in general depends on t and the whole distribution of ci . I will, however,
proceed with the following analysis as if it were an exogenously given constant δ,
uniform both across technologies and over time. This will simplify the exposition
of our evolutionary model immensely without losing any of its qualitative nature.12
Then, we can rewrite (7.6) as:
Ṡt (ci ) ∼
= γδSt (ci )(1 − St (ci )) (i = n, n − 1, . . . , 1). (7.8)
Each of the above equations is a well-known ‘logistic differential equation’ with
a logistic parameter µ, and can be solved explicitly to yield:
1
St (ci ) = (i = n, n − 1, . . . , 1), (7.9)
1 + (1/ST (ci ) − 1)e−γδ(t−T )
where e stands for the exponential and T (≤ t) a given initial time.13
Differential growth rates among firms with different cost conditions never leave
the industry’s state of technology static. As the firms with relative cost advantage
grow faster than the firms with relative cost disadvantage, the distribution of capital
shares gradually shifts in favor of the lower unit costs, thereby reducing the average
unit cost of the industry as a whole. This process then eliminates the relative cost
advantage of the existing technologies one by one until the capital share of the
least unit cost completely overwhelm those of the higher ones. Only the fittest
will survive in the long run through their higher growth rates, and this of course
is an economic analogue of the ‘Darwinian’ natural selection mechanism. The set
of logistic equations (7.9) describes this ‘economic selection’ mechanism in the
simplest possible mathematical form, and its evolutionary dynamics is illustrated
by Figure 7.4. In particular, the equation for i = n shows that the cumulative
capital share of the lowest unit cost St (cn ) moves along an S-shaped growth path.
It grows almost exponentially when it occupies a negligible portion of the industry,
gradually loses its growth momentum as its expansion narrows its own relative cost
advantage, but never stops growing until it swallows the whole industry.
St (cn –2 )
St (cn –1 )
St (cn )
0
T t
Figure 7.4 Evolution of the state of technology under the pressure of either economic
selection or technological diffusion
Hypothesis (IM ): The probability that a firm with unit cost ci succeeds in imitating
a technology with unit cost cj is equal to:
1
St (ci ) = (i = n, n − 1, . . . , 1), (7.12)
1 + (1/ST (ci ) − 1)e−µ(t−T )
where T (≤ t) is a given initial time.
Since the second set of logistic equations (7.12) is mathematically equivalent to
the first set of logistic equations (7.9), Figure 7.4 in the preceding section can again
serve to illustrate the dynamic evolution of the cumulative capacity shares under
the sole pressure of technological diffusion. And yet, the logic behind these second
logistic equations is entirely different from that of the first. ‘If one or a few have
advanced with success many of the difficulties disappear’, so wrote Schumpeter,
‘others can then follow these pioneers, as they will clearly do under the stimulus
of the success now attainable. Their success again makes it easier, through the
increasingly complete removal of the obstacles . . . , for more people follow suit,
until finally the innovation becomes familiar and the acceptance of it a matter of
free choice.’(Schumpeter (1961), p. 228) The logistic equations (7.12) describe
this swarm-like appearance of technological imitations in the simplest possible
form. In particular, the equation for i = n shows that the cumulative capital share
of the lowest unit cost moves along a S-shaped growth path, initially growing
at an exponential rate but gradually decelerating its growth rate to approach unity
asymptotically. In the long run, therefore, the lowest cost technology will dominate
the whole industry, simply because it will eventually be diffused to all the firms
in it. This technological diffusion process is nothing but an economic analogue of
the ‘Lamarkian’ model of biological evolution – the achievement of one individual
are passed directly to the other individuals.
Let us then bring back the Darwinian process of economic selection into our
industry and add (7.6) to (7.11). The result is the third set of logistic differential
equations in the present chapter:
1
St (ci ) = (i = n, n − 1, . . . , 1), (7.14)
1 + (1/ST (ci ) − 1)e−(γδ+µ)(t−T )
Schumpeterian dynamics 179
for t ≥ T . (We refrain from drawing a diagram for the third set of logistic equations
(7.14) which is qualitatively the same as Figure 7.4.)
We have thus shown how the mechanism of economic selection and the pro-
cess of technological diffusion jointly contribute to the logistic growth process of
cumulative capital shares – the former by amassing the industry’s capacities in
the hands of the lowest cost firms and the latter by diffusing the advantage of the
lowest cost technology among imitating firms. While the former is Darwinian, the
latter is Lamarkian. But, no matter how opposed the underlying logic might be,
their effects upon the industry’s state of technology are the same – the lowest cost
technology will eventually dominate the whole capital stocks of the industry.
St (cn –2 )
St (cn –1 )
St (cn )
St (cn +1 )
St (cn +2 )
St (cn +3 )
0
T (cn ) T (cn +1 ) T (cn +2 ) T (cn +3 ) t
Figure 7.5 Evolution of the state of technology under the joint pressure of economic
selection, technological diffusion and recurrent innovations
A new question then arises: is it possible to derive any law-like properties out of
this seemingly erratic movement of the industry state of technology?
In order to give an answer to this question, it is necessary to introduce two
more hypotheses – one pertaining to invention and the other to innovation. The
conceptual distinction between invention and innovation was very much empha-
sized by Schumpeter. Invention is a discovery of new technological possibility
which is potentially applicable to the production processes of the economy. But,
‘as long as they are not carried into practice’, so says Schumpeter, ‘inventions
are economically irrelevant’, and ‘to carry any improvement into effect is a task
entirely different from the inventing of it’.19
Denote then by C(t) the unit cost of potentially the best possible technology
at time t and call it ‘the potential unit cost’. The following is our hypothesis about
the process of inventions:20
Hypothesis (PC): The potential unit cost is declining at a positive constant rate
λ over time.
The declining rate of potential unit cost λ reflects the speed at which the stock
of technological knowledge is being accumulated by academic institutions, private
firms, government agencies and amateur inventors throughout the entire economy.
In the present chapter which follows an evolutionary perspective, however, it is
assumed to be given exogenously to the industry.
Schumpeterian dynamics 181
We are then able to characterize the notion of ‘innovation’ formally as an event
in which the potential unit cost is put into actual use by one of the firms in the
industry. This is tantamount to saying that when an innovation takes place at time
t, it brings in a technology of unit cost C(t) for the first time into an industry.
This also implies that if a technology with unit cost c is presently in use, it must
have been introduced at time t = T (c) where T (c) is the inverse function of C(t)
defined by:
Hypothesis (IN − a): The probability that a firm succeeds in an innovation is equal
to:
vdt, (7.17)
during any small time interval dt, where ν is a small positive constant.
Ŝt (c) = (γδ + µ)Ŝt (c)(1 − Ŝt (c)) + ν(1 − Ŝt (c)), (7.18)
for t ≥ T (c). It turns out that this is the fourth set of logistic differential equations
of this chapter, for each of which can be rewritten as ẋ = (γδ + µ + ν)x(1 − x)
with x ≡ (Ŝt (c) + ν/(γδ + µ))/(1 + ν/(γδ + µ)). It can thus be solved to yield:
1+ ν ν
Ŝt (c) = γδ+µ − , (7.19)
1+ γδ+µ
e−(γδ+µ+ν)(t−T (c)) γδ +µ
ν
for t ≥ T (c).25
Of course, we cannot hope to detect any regularity just by looking at the motion
of expected cumulative shares Ŝt (c) given above, for they are constantly pushed to
the lower cost direction by recurrent innovations. If, however, we neutralize such
declining tendency by measuring all unit costs c relative to the potential unit cost
C(t) and observe the relative pattern of the cumulative capital shares, a certain
regularity is going to emerge out of the seemingly unpredictable vicissitude of
the industry’s state of technology. Let us thus denote by z the proportional gap
between a given unit cost c and the current potential unit cost C(t), or
αβ
0 z
(7.22) by such basic parameters as γδ, µ, ν and λ. The first composite parame-
ter α ≡ ν/(γδ + µ) represents the relative strength between the disequilibriating
force of creative-cum-destructive innovations and the joint equilibriating force
of economic selection mechanism and diffusion process through imitations. The
second composite parameter β ≡ λ/ν, on the other hand, represents the relative
strength between the force of inventions and that of innovations. Since the expected
rate of innovation per unit of time is 1/ν and the reduction rate of the potential unit
cost per unit of time is λ, β can also be interpreted as the expected cost reduction
rate of each innovation.26 It is not difficult to show that:27
∂ S̃(z) ∂ S̃(z)
<0 and < 0. (7.23)
∂α ∂β
As is illustrated in Figure 7.6, an increase in both α and β thus shifts S̃(z) clock-
wise, thus rendering the distribution of efficiencies across firms more disperse than
before.
The long-run cumulative distribution S̃(z) thus deduced is a statistical summary
of the way in which a multitude of technologies with diverse cost conditions are
dispersed among all the existing capital stocks of the industry. It shows that, while
the on-going inventive activities are constantly reducing the potential unit cost,
the unit costs of a majority of production methods actually in use lag far behind
this potential one. The state of technology therefore has no tendency to approach
a classical or neoclassical equilibrium of uniform technology even in the long run.
What it approaches over a long period of time is merely a ‘statistical equilibrium
of technological disequilibria’.
184 Katsuhito Iwai
7.7 The industry supply curve in the long run
Now, the fact that the state of technology retains the features of disequilibrium
even in the long run does have an important implication for the nature of the
industry’s long-run supply curve. For, as is seen by (7.4), the relative form of
industry supply curve yt = Yt (Pt )b/Kt traces the shape of St (c), except for the
portions of discontinuous jumps. Hence, if the expectation of St (c) tends to exhibit
a statistical regularity in the form of S̃(z), the expectation of the relative form of
the industry supply curve should also exhibit a statistical regularity in the same
long-run form of S̃(z). Let us denote by pt the relative gap between a given product
price Pt and the potential unit cost C(t), or
Proposition (SC): Under Hypotheses (CG), (IM’), (PC) and (IN-a), the expected
value of the relative supply curve of the industry yt = Qt (Pt )b/Kt will in the long
run approach a functional form of
1+α
S̃(pt ) ≡ 1+ α − α. (7.25)
1 + (1/α)e− αβ pt
Figure 7.7 exhibits the relative form of the industry’s long-run supply curve as
a function of price gap p. As a matter of fact, it has been drawn simply by turn-
ing Figure 7.6 around the 45 degree line. It therefore moves counter-clockwise as
either of the composite parameters α and β increases. This implies that the long-
run supply curve becomes more upward-sloping, as the disequilibriating force of
creative-cum-destructive innovations becomes stronger than the joint equilibriat-
ing force of economic selection and technological diffusion or as the average rate
of cost reduction of each innovation becomes larger.
What is most striking about this long-run supply curve, however, is not that
it is the ‘sixth’ logistic curve we have encountered in this chapter but that it is an
upward-sloping supply curve!
Let us recall the lower panel of Figure 7.1 of the introductory section. It re-
produced a typical shape of the long-run supply curve which can be found in any
textbook of economics. This horizontal curve was supposed to describe the long-
run state of the industry in which the least cost technology is available to every
firm in the industry and all the opportunities for positive profits are completely
wiped out. However, the relative form of the long-run supply curve we have drawn
in Figure 7.7 has nothing to do with such traditional picture. There will always be
a multitude of diverse technologies with different cost conditions, and the industry
supply curve will never lose an upward-sloping tendency, just as in the case of
the ‘short-run’ supply curve of the upper panel of Figure 7.1. There are, therefore,
always some firms which are capable of earning positive profits, no matter how
competitive the industry is and no matter how long it is run.
Schumpeterian dynamics 185
p
αβ
y
0 1
We can thus conclude that positive profits are not only the short-run phe-
nomenon but also the long-run phenomenon of our Schumpeterian industry. It is
true that the positivity of profits is a symptom of disequilibrium. But, if the industry
will approach only a statistical equilibrium of technological disequilibria, it will
never stop generating positive profits from within even in the never-never-land of
long-run.
e*
p*
r*
y
0 y* 1
p∗ p∗
r∗ = b(p∗ − z)s̃(z)dz = b S̃(z)dz (7.26)
0 0
αβb ∗ y∗
= − log(1 − y ) − α log 1 + > 0.
1+α α
j=i
j=i
Lt = cj st (cj )Kt /b + ci−1 Yt − st (cj )Kt /b
j=n j=n
ci
≡ cdSt (c)Kt /b + ci−1 (Yt − St (ci )Kt /b)
0
whenever S(ci )Kt /b ≤ Yt < S(ci−1 )Kt /b. If we divide this relation by Kt /b and take
its inverse, we can construct a functional relation between the industry-wide labor–
capacity ratio lt ≡ Lt b/Kt and the industry-wide output–capacity ratio yt ≡ Yt b/Kt
as:
yt = ft (lt ), (7.32)
ci
where l ≡ 0 cdSt (c) + ci−1 (ft (l) − St (ci )) whenever St (ci ) ≤ ft (l) < St (ci−1 ).
Figure 7.9 depicts this functional relation in a Cartesian diagram which measures
labor–capacity ratio l along horizontal axis and output-capacity ratio y along a
vertical axis. It is evident that this relation satisfies all the properties a neoclassical
production function is supposed to satisfy.30 Y is linearly homogeneous in L and
K, because y ≡ Yb/K is a function only of l ≡ Lb/K. Though not smooth, this
relation also allows a substitution between Kt and Lt and satisfies the marginal
productivity principle: ∂ȳt /∂lt ≤ 1/Pt ≤ ∂ + yt /∂lt . (Here, 1/Pt represents a real
wage rate because of our choice of money wage rate as the numeraire, and ∂ȳ/∂l
and ∂ + y/∂l represent left- and right-partial differential, respectively.) Yet, the
important point is that this is not a production function in the proper sense of
the word. It is a mere theoretical construct that has little to do with the actual
technological conditions of the individual firms working in the industry. It is in
this sense that we call the relation (7.32) a ‘short-run pseudo-aggregate production
function’, with an emphasis on the adjective: ‘pseudo’.
The shape of the short-run pseudo-production function y = ft (l) is determined
by a distribution of capital shares {St (ci )} across technologies. Hence, as this
distribution changes, the shape of this short-run function also changes. And in our
Schumpeterian industry, the distribution of capital shares is incessantly changing
over time as the result of dynamic interplay among capital growth, technological
innovation and technological imitation. The most conspicuous feature of the short-
run pseudo-production function is, therefore, its instability.
In the long run, however, we know we can detect a certain statistical regularity
in the distribution of capital shares {St (ci )} out of its seemingly unpredictable
190 Katsuhito Iwai
yt
1
xt
0
movement. We can thus expect to detect a certain statistical regularity in the pseudo-
production function as well out of its seemingly unpredictable movement. Let l̂ and
ŷ denote the expectation of labor–capacity ratio l ≡ Lb/K and of output–capacity
ratio y ≡ Yb/K, respectively. Then, we indeed arrive at:31
Proposition (PF): Under Hypotheses (CG), (IM’), (PC) and (IN-a), the func-
tional relationship between the expected labor–capacity ratio and the expected
output–capacity ratio will in the long run take the form of :
ŷ = f˜ l̂eλt , (7.33)
xe λt
0
mere statistical illusion! If we zoomed into the microscopic level of the economy,
what we would find is the complex and dynamic interactions among many a firm’s
capital growth, technological imitation and technological innovation. In fact, as
is seen from (7.34), the functional form of f˜(·) is a complex amalgam of such
basic parameters of our Schumpeterian model as α ≡ ν/(γδ + µ) and β ≡ λ/ν. It
is just impossible to disentangle various microscopic forces represented by these
parameters and decompose the overall growth process into a movement along a
well-defined aggregate production function and an outward shift of the function
itself.33 Indeed, it is not hard to show that both an increase in α and in β shift the
function f˜(·) in the downward direction,34 or
∂ f˜(·) ∂ f˜(·)
< 0 and < 0. (7.35)
∂α ∂β
We are after all living in a Schumpeterian world where the incessant reproduction
of technological disequilibria prevents the aggregate relation between capital and
labor from collapsing into the fixed proportion technology of individual firms. It
is, in other words, its non-neoclassical features that give rise to the macroscopic
illusion that the industry is behaving like a neoclassical growth model. It is for this
reason we will call the relation (7.33) the ‘long-run pseudo aggregate production
function’.
∂r ∗
p∗ ∂ S̃(z)
∂α p∗ = const. =b ∂α dz ≡ −Aα
0
βb
1
= − α(1+α) e−( 1+ y∗ ) / ( αβ) u−1−log
(1+u)2 du < 0;
u
p∗ ∂ S̃(z)
∂r ∗ (7.A1)
∂β p∗ = = b 0 ∂β dz ≡ −Aβ
const. 1
= − (1+β)b
αβ e−( 1+ y∗ ) / ( αβ) (1+u)2 du < 0.
u−1−log u
Schumpeterian dynamics 193
This is nothing but (7.29) of the main text. Note also that since ∂ S̃(z)/∂ν >
∗
0, ∂r ∗
∂ν p = const. > 0.
Next, consider the case of an absolutely inelastic demand curve. As is shown
in Figure 7.A2 which juxtaposes a vertical demand curve on an inverted logistic
shape of the long-run supply curve, an increase in either α or β moves the latter
counter-clockwise and transfers the equilibrium point from e∗ to e∗∗ along the
vertical demand curve. This raises p∗ to p∗∗ , while keeping y∗ the same as before.
The long-run profit rate thus changes from 0e∗ p∗ to 0e∗∗ p∗∗ . We have to examine
whether this amounts to an increase or decrease of r ∗ . To see this, Figure 7.A2
decomposes this change of profit rate into two components – A ≡ 0e∗ e∗∗ and
B ≡ p∗ e∗ e∗∗ p∗∗ . The first component A represents the ‘loss’ of profit rate due to
a universal increase of cost gaps, which corresponds to the profit loss A of the
previous case. In the present case of absolutely inelastic demand curve, however,
an increase in the long-run equilibrium price gap gives rise to a ‘gain’ of profit rate,
as is represented by the second component B. Whether r ∗ increases or decreases
thus depends on whether A is smaller or larger than B. This can be checked by
differentiating (7.26) with respect to α and β, keeping y∗ constant. We have:
S̃−1(y∗ )
∂r ∗ ∂ S̃(z) ∂ S̃ −1 (y∗ )
= b dz + by∗
∂α y= const. 0 ∂α ∂α
≡ −Aα + Bα .
βb ∗ y∗
= − log(1 − y ) − α log 1 +
(1 + α)2 α
y∗ y∗
−(1 + α)α log 1 + − ; (7.A2)
α α + y∗
S̃ −1 (y∗ )
∂r ∗ ∂ S̃(z) ∂ S̃ −1 (y∗ )
= b dz + by∗ ≡ −Aβ + Bβ
∂β y∗ = const. 0 ∂β ∂β
αb y∗
= − log(1 − y∗ ) + α log 1 + > 0.
1+α α
∗ ∗ ∗
Although both − log(1 − y∗ ) − α log(1 + yα ) and (1 + α)α(log(1 + yα ) − α+y y
∗ ) are
positive in the first expression, the former dominates the latter if we let α → 0. Since
α ≡ ν/(γδ + µ) is assumed to be small, it does not seem unreasonable to suppose
the first expression to be positive. The second expression is always positive. Hence,
∗
(7.30) of the main text. Note that we can also calculate ∂r ∗
∂ν y =const. as
−α2 βb αy∗ y∗ y∗
− log(1 − y∗ ) − − log 1 + − < 0.
(1 + α)2 α + y∗ α α + y∗
Finally, let us consider the general case where industry demand curve is neither
perfectly elastic nor absolutely inelastic. As is seen from Figure 7.A3, an increase
in either α or β transfers the equilibrium point upward from e∗ to e∗∗ along this
downward-sloping demand curve. This raises p∗ to p∗∗ but lowers y∗ to y∗∗ , thereby
changing r ∗ from 0e∗ y∗ to 0e∗∗ y∗∗ . We can then decompose this change again into
A ≡ 0e∗ e∗∗ and B ≡ p∗ e∗ e∗∗ p∗∗ . A represents the ‘loss’ of r ∗ due to a universal
194 Katsuhito Iwai
αβ
e ** e*
p*
A
y
0 y ** y*
Figure 7.A1 The case of a perfectly elastic demand curve
e **
p **
αβ
B
e*
p*
A
y
0 y*
Figure 7.A2 The case of an absolutely inelastic demand curve
Schumpeterian dynamics 195
p
αβ
e **
p **
B'
e*
p*
A
y
0 y ** y*
Figure 7.A3 The general case
increase of cost gaps, and B represents the ‘gain’ due to an increase in the long-run
equilibrium price gap. However, B in Figure 7.A3 is not as large as B in Figure
7.A2, for the price elasticity of the demand allows the effect of cost increases to
be absorbed not only by price hike but also by quantity reduction. This means
that when the demand curve is steeply sloped, the gain component B is likely
to outweight the loss component A. But, when the demand curve becomes more
elastic, B becomes smaller, and in the limiting case of perfectly elastic demand
curve it shrinks to zero.
This graphical explanation can be formalized as follows. First write down the
relative form of industry demand function as yt = D̃(pt ). Then, p∗ is determined
by the supply-demand equation: S̃(p∗ ) = D̃(p∗ ). Differentiating this with respect
to α and β and rearranging terms, we have: ∂p∗ /∂α = (−p∗ ∂ S̃(p∗ )/∂α)/(ε +
η) and ∂p∗ /∂β = (−p∗ ∂ S̃(p∗ )/∂β)/(ε + η), where ε and η are the price-
elasticity of the supply curve and of the demand curve, respectively defined by
(∂ S̃(p)/∂p)/(S̃(p)/p) and (∂ D̃(p)/∂p)/(D̃(p)/p). Keeping this in mind and dif-
ferentiating (24), we obtain:
∂r ∗
S̃ −1 (y∗ ) ∂ S̃(z) ∗
∗ −∂ S̃(p ) 1
∂α S̃(p∗ )=D̃(p∗ ) =b 0 ∂α dz + bp ∂α η+ε ≡ −Aα + Bα . ;
∂r ∗
S̃ −1 (y∗ ) ∂ S̃(z) ∗
∗ −∂ S̃(p ) 1
(7.A3)
∂β S̃(p∗ )=D̃(p∗ ) = b 0 ∂β dz + bp ∂β η+ε ≡ −Aβ + Bβ .
Notes
1 Schumpeter (1939), p.100.
2 Schumpeter (1950), p. 83.
3 Schumpeter (1950), p. 28.
4 See, for instance, Nelson and Winter (1982), Dossi, Freeman, Nelson, Silverberg and
Soete (1988), Metcalfe and Saviotti (1991), and Anderson (1994) for the comprehen-
sive expositions of the ‘evolutionary perspective’ in economics.
5 The term ‘satisficing’ was first coined by Simon (1957) to designate the behavior of a
decision maker who does not care to optimize but simply wants to obtain a satisfactory
utility or return. The notion of ‘organizational routines’ owes to Nelson and Winter
(1982). Organizations ‘know’ how to do things. In Iwai (1999) I have provided a legal-
economic-sociological framework for understanding the nature and sources of such
organizational capabilities.
6 Or we can think of this as a one-commodity economy with many competing firms.
7 Our evolutionary model can also accommodate a wide variety of industry structures.
See Appendix A of Iwai (1984b) for the way to deal with the case of monopolistically
competitive industry.
8 It is easy to show from (7.8) below that: K̇t /Kt = γ(log Pt − i (log ci )st (ci )) − γ0 ,
so that the growth rate of the industry’s total capital stock is linearly dependent on
the proportional gap between the price-wage ratio Pt and the industry-wide average
unit cost. If K̇t /Kt is pre-determined (probably by the growth rate of the demand for
this industry’s products), this equation can be used to determine Pt . If, on the other
hand, Pt is pre-determined (probably by the labor market conditions in the economy
as a whole), this equation can be used to determine K̇t /Kt . In either case, the forces
governing the motion of Kt are in general of the different nature from those governing
the evolution of {St (c)}.
9 Schumpeter (1961), p. 154.
10 It is, however, not so difficult to deduce an investment function of this form by explicitly
setting up an intertemporal optimization problem with adjustment costs, as in Uzawa
(1969).
11 The actual derivation is as follows.
n
Ṡt (ci ) ≡ ṡt (cj )
j= i
n
= (k̇t (cj )/kt (cj ) − Kt /Kt )st (cj )
j= i
Schumpeterian dynamics 197
n
n
= (γ(log pt − log cj ) − γ0 ) − (γ(log pt − log ch ) − γ0 )st (ch ) st (cj ) by (7.5)
j= i h= 1
n
n
= γ (log ch )st (ch ) − log cj st (cj )
j= i h= 1
i−1
n
= γ (log ch )st (ch )St (ci ) − (log ch )st (ch ) (1 − St (ci ))
h= 1 h= i
= γδt (ci )St (ci )(1 − St (ci )).
12 This is the simplification I also adopted in Iwai (1984b). However, in a recent article
Franke (1998) indicated that the value of δt (c) may actually vary considerably as the
parameter values of γ, ν and λ as well as the value of c vary. A caution is thus needed
to use this approximation for purposes other than heuristic device.
13 A logistic differential equation: x = ax(1 − x) can be solved as follows. Rewrite it as:
x /x − (1 − x) /(1 − x) and integrate it with respect to t, we obtain: log(x) − log(1 −
x) = log(x0 ) − log(1 − x0 ) + at, or x/(1 − x) = eat x0 /(1 − x0 ). This can be rewritten
as: x = 1/(1 + (1/x0 − 1)eat ), which is nothing but a logistic equation given by (7.9).
14 The reason I have designated this Hypothesis by (IM ) is to differentiate it from a
slightly different hypothesis adopted in Iwai (1984a). Its Hypothesis (IM) assumes
that the probability of imitating a better technology is proportional to the frequency
(rather than their capital share) of the firms using it. On the other hand, Iwai (2000) has
adopted yet another hypothesis which assumes that firms imitate only the best practice
technology and the probability of its success is proportional to the frequency of the
firms using it.
15 p. 615.
16 See, for instance, Mansfield, Schwartz and Wagner (1981), Gorts and Klepper (1982)
and Metcalfe (1988).
17 It is, however, possible to incorporate a trade-off between the resources devoted to
capital growth and the resources devoted to imitative activities into our model. For
instance, the growth parameters γ and/or −γ0 in (7.5) can be made a decreasing function
of the imitation coefficient µ.
18 The actual derivation is as follows. The value of St (ci ) increases whenever one of the
firms with unit costs higher than ci succeeds in imitating one of the technologies with
unit costs ci or lower. Indeed, because of the assumption of the disembodied nature of
technology, it increases by the magnitude equal to the imitator’s capacity share. Note
that St (ci ) is not affected by the imitation of any of the firms with unit costs ci or less,
for it only effects an infra-marginal transfer of capacity share. Let Mt (ci ) denote the
number of firms with unit costs ci or lower. Since the average capacity share of the
firms with unit costs higher than ci is (1 − St (ci ))/(M − Mt (ci )) and the probability
of a successful imitation for each of those M − My (ci ) firms is µSt (ci )dt during a
small time interval dt, we can calculate the expected increase in St (ci ) during dt as
((1 − St (ci ))/(M − M (ci ))(µSt (ci )dt)(M − Mt (ci )) = (µSt (ci )dt)(1 − St (ci ))). If the
number of firms is sufficiently large, the law of large numbers allows us to use this
expression as a good approximation of the actual rate of change in St (ci ). Dividing
this by dt and letting dt → 0, we obtain (7.11).
19 Schumpeter (1961), p. 88.
20 Iwai (2000), however, presents an evolutionary model which does not separate inno-
vators from inventors and assume that each innovation raises the productivity of the
industry’s best technology by a fixed proportion.
21 See, for instance, Kamien and Schwarts (1982), Grilliches (1984) and Scherer and
Ross (1990).
198 Katsuhito Iwai
22 It is, however, possible to incorporate a trade-off between the resources devoted to
capital growth and the resources devoted to innovative activities into our model. For
instance, the growth parameters γ and/or −γ0 in (7.5) can be made a decreasing function
of the innovation coefficient ν.
23 Iwai (1984a, 2000) also develops versions of evolutionary models which assume that
only the firms currently using the best technology can strike the next innovation. In
this case, the process of technological innovations is no longer a Poisson process, so
that it is necessary to invoke the so-called ‘renewal theory’ in mathematical probability
theory to analyze the long-run performance of the state of technology.
24 The derivation is as follows. Whenever one of the firms with unit costs higher than
c succeeds in innovation, the value of St (c) increases by the magnitude equal to the
innovator’s capacity share. (St (c) is, however, not affected by the innovation of any of
the firms with unit costs c or less, because it only effects an infra-marginal transfer of
the capacity share.) As in note 11, let M − Mt (ci ) denote the total number of firms with
unit costs higher than ci . The average capacity share of the firms with unit costs higher
than ci is (1 − St (ci ))/(M − Mt (ci )) and the probability of a successful innovation for
each of those M − Mt (ci ) firms is νdt during a small time interval dt. We can then
calculate the expected increase in St (ci ) due to an innovation as ((1 − St (ci ))/(M −
M (ci )))(νdt)(M − Mt (ci )) = (νdt)(1 − St (ci )). If we divide this by dt and add to it the
effects of economic selection and technological imitations given by (7.13), we obtain
(7.18).
25 In deducing (7.19), we have employed a boundary conditionŜT (c) (c) = ν or ŜT (c) (c) =
0.
26 In our companion paper [1998] which assumes the step-by-step nature of innovations,
it is β that is assumed to be exogenously given.
27 The derivation is as follows.
and
Note that ν appears both in α and in β. But its impact on S̃(z) can be calculated as
References
Anderson, E.S. (1994) Evolutionary Economics: Post-Schumpeterian Contributions, Pinter:
London.
Arrow, K. (1962) ‘Economic welfare and the allocation of resources for inventions’, in R. R.
Nelson (ed.), The Role and Direction of Inventive Activity, Princeton University Press:
Princeton.
Dossi, G., G. Freeman, R. Nelson, G. Silverberg and L. Soete (1988) (eds), Technical
Change and Economic Theory, Pinter: London.
Franke, R. (1998) ‘Wave trains and long waves: a reconsideration of Professor Iwai’s Schum-
peterian dynamics’, forthcoming in D. Dell Gatti, M. Gallegati and A. Kirman (eds),
Market Structure, Aggregation and Heterogeneity. Cambridge University Press: Cam-
bridge.
200 Katsuhito Iwai
Gorts, M. and S. Klepper (1982) ‘Time-paths in the diffusion of product innovations’,
Economic Journal, 92, 630–53.
Grilliches, Z. (1984) (ed.) R and D, Patents and Productivity, Chicago University Press:
Chicago.
Iwai, K. (1984a) ‘Schumpeterian dynamics: an evolutionary model of innovation and imi-
tation’, Journal of Economic Behavior and Organization, 5, (2) 159–90.
Iwai, K. (1984b) ‘Schumpeterian dynamics, Part II: technological progress, firm growth and
“economic selection”,’ Journal of Economic Behavior and Organization, 5, (3) 321–51.
Iwai, K. (1999) ‘Persons, things and corporations: Corporate personality controversy and
comparative corporate governance’, American Journal of Comparative Law, 45, 101–50.
Iwai, K. (2000) ‘A contribution to the evolutionary theory of innovation, imitation and
growth’, Journal of Economic Behavior and Organization, 43, 167–98.
Kamien, M. and N. Schwarts (1982) Market Structure and Innovation, Cambridge Univer-
sity Press: Cambridge.
Maddison, A. (1987) ‘Growth and slowdown in advanced capitalist economies: Techniques
of quantitative assessment’, Journal of Economic Literature, 25 (June 1987), 649–98.
Mansfield, E., M. Schwartz and S. Wagner (1981), ‘Imitation costs and patents: an empirical
study’, Economic Journal, 91, (364), 907–18.
Metcalfe, J.S. ‘The diffusion of innovation: an interpretive survey’, Chapter 25 of G. Dossi
et al. (1988).
Metcalfe, J.S. and P. Saviotti (1991) (eds) Evolutionary Theories of Economics and Tech-
nological Changes, Harwood Publishers: Reading. PA.
Nelson, R. and S. Winter (1974) ‘Neoclassical vs. evolutionary theories of economic growth:
critique and perspective’, 84, 886–905.
Nelson, R. and S. Winter (1982) An Evolutionary Theory of Economic Change, Harvard
University Press: Cambridge, MA.
Sato, K. (1975) Production Functions and Aggregation, North-Holland: Amsterdam.
Scherer, F.M. and D. Ross (1990) Industrial Market Structure and Economic Performance,
Houghton and Mifflin: Boston.
Schumpeter, J.A. (1939) Business Cycles, McGraw-Hill: New York.
Schumpeter, J.A. (1950) Capitalism, Socialism and Democracy, 3rd. ed., Rand McNally:
New York.
Schumpeter, J.A. (1961) The Theory of Economic Development, Oxford University Press:
Oxford.
Simon, H.A. (1957) Models of Man, Wiley: New York.
Solow, R.M. (1956) ‘A contribution of the theory of economic growth’, Quarterly Journal
of Economics, 70, (1), 65–94.
Solow, R.M. (1957) ‘Technical change and the aggregate production function’,Review of
Economics and Statistics. 39, 312–20.
Uzawa, H. (1969) ‘Time Preference and the Penrose Effect in a Two-Class Model of Eco-
nomic Growth’. Journal of Political Economy, 77 (July/Aug.), 628–92.
Part III
Dynamics by interaction
8 Asymmetrical cycles and
equilibrium selection in
finitary evolutionary
economic models
Masanao Aoki
8.1 Introduction
This chapter takes a fresh look at economic phenomena with multiple equilibria,
and some of the issues associated with modeling asymmetrical business cycles. It
summarizes an approach proposed in Aoki (1996, 1998), which is based on using
continous time Markov chains and illustrates the advantages of using models with
at most countable state spaces to answer these questions. The main advantage of
employing models with a finite number of agents is simplicity in clarifying the issue
of asymmetrical oscillation and naturalness of the criterion proposed for equilib-
rium selection. Two examples are presented to illustrate the proposed method. One
is a binary choice model in which agents choose one of two alternative decisions
or behavioral rules. Agents’ decisions are interdependent due to externalities, such
as congestion, bandwagon, group sentiments, network externalites and the like.
The other example is a finite agent version of Diamond’s search model, Diamond
(1982).
which is known as the master equation. See Aoki (1996, 116) for a heuristic
derivation. See Karlin and Taylor (1981, Chapter 14 ), for example, for the proof
of the existence of the time derivative.
Write the sum of the right hand side separately for state X and the rest as
P(X , t)wN (X |X , t) = P(X , t)wN (X |X , t) + P(X , t)wN (X |X , t),
X X =X
and substitute − X =X wN (X |X , t) in w(X |X , t) to rewrite the master equation
as
dP(X , t)/dt = P(X , t)wN (X |X , t) − P(X , t)wN (X |X , t).
X =X X =X
This is the usual form in which the master equation is stated. The first term is
the sum of the probability flows into state X , and the second the probability flow
out of state X .
We assume that x(t) = X (t)/N behaves as an intensive variable, and that the
transition rates depend on X and N through x, except possibly for the scale factor
which may depend on N .3
We make one additional key assumption that the change in the state variable, i.e.
X −X , remains the same for different values of N . This assumption is certainly met
Asymmetrical cycles and equilibrium selection 205
in birth-and-death processes, or birth-and-death with immigration processes, since
jumps are restricted to be ±1 or some fixed integers from any state regardless of the
total number N . This assumption actually is concerned with the scaling properties
or homogeneity properties of the transition rates. Loosely put, it means that each
of N agents contribute approximately equally to the transition events. To make this
explicit, express the transition rate as a function of the starting state, X , and the
jump (vector), r = X − X as
wN (X |X ) = wN (X ; X − X ) = wN (X ; r),
for some function Φ, and where x = X /N . Using the same function we can express
the transition rate in the opposite direction as
wN (X |X ) = N Φ(x; −r).
wN (X ; r) = f (N )Φ(x ; r),
for some positive function f (N ). Actually, this factor f (N ) can be arbitrary, since
it can always be absorbed into the choice of time unit.
More generally, the transition rates may take the form
where higher order terms in N −1 may represent higher order interactions among
microeconomic units beyond those which are captured by the leading term. In
terms of these transition rates the master equation may be rewritten as
dP(X , t) X −r X −r
= f (N ) {Φ0 ( ; r) + N −1 Φ1 ( ; r) + · · ·)}P(X − r, t)dr
dt N N
X X
− f (N ) {Φ0 ( ; r) + N −1 Φ1 ( ; r) + · · ·}P(X , t)dr. (8.2)
N N
Series expansions in terms of some fractional powers of N , such as N −1/2 may
also be used in some problems.
Example 1: A binary choice model Suppose that the total number of agents,
N , is constant and that each of N agents face a binary choice, c1 , and c2 . Merits
or benefits of choices depend on the fraction of agents with particular choices.
Let n be the number of agents with c1 at a point in time. This, or its fractional
form n/N serves as state variable. We suppress time argument for simplicity. If an
agent can make his choice independent of others, then the transition rate from n to
206 Masanao Aoki
n + 1 occurs because one of the agents with choice 2 has changed its mind, while
n becomes n − 1 if one of the agents with choice 1 changes its mind. We assume
no entry and no exit to keep N fixed in this example, for simplicity. To account for
the externality of decisions we posit
n
rn := wN (n + 1|n) = a(N − n)η = aN (1 − x)η(x),
N
and n
ln := wN (n − 1|n) = bn 1 − η = bNx[1 − η(x)].
N
Without the factor η, each of N − n agents changes its mind at the rate a∆t over a
small time interval ∆t, and each of n agents does likewise at the rate b∆. Here a
and b are some constant. To simplify presentation we let them be the same.
The master equation is given, in terms of n(t)
together with boundary conditions at 0 and N , where n(t) is the number of agents
with the first choice. Equivalently it can be written in terms of x(t) = n(t)/N .
N
ln := wN (n − 2|n) = eb(e).
2
The master equation is
together with some boundary conditions. See Aoki and Shirai (2000) for exact
expressions.
Asymmetrical cycles and equilibrium selection 207
8.3 Two types of state variables
In these two examples the number of agents of some type ( with choice c1 or
employed) serves as state variable. In situations with K types of agents, K ≥ 3, we
need a vector n = (n1 , . . . , nK ) as state vector in general, where ni is the number
of agents of type i or choice i. More correctly, the vector with fractions ni /N as
the ith component, i = 1, . . . , K is a state vector. When the total number of agents
is fixed, then a K − 1 dimensional vector may be used.
Although this choice of state variables seems quite natural, many situations can
be described better using an alternative choice of state variables. This alternative
choice is in line with the occupancy problems in statistics or in physics. See Feller
(1970) for example. In this definition, think of types as boxes and agents as indis-
tinguishable balls. That is, agents are treated as exchangeable, the labels attached
to agents being non-intrinsic and for mere convenience. The state vector is defined
by a vector a with components ai being the number of types (boxes) with exactly
i agents
in each of them, i = 1, . . . , K. By definition, then we have i iai = N ,
and i ai ≤ K. This type of state variables are useful in dealing with distributions
of firms by size, for example. Here interpret size broadly such as the number of
employees, amount of capital or output in some convenient units, and so on. We
do not have space to discuss models of this kind of state variable description. See
Aoki (1999, 2000) for examples.
8.4 Dynamics
We use dynamic equations for the probability distributions of the states as the
basic dynamic description of economic models with many agents of several types
or agents with discrete choices. We do not use ordinary differential equations for the
states themselves. This distinction is important, if subtle. That is, we do not derive
differential equations for X (t)/N , such as the fraction of the employed in Diamond
(1982), for example, but rather the differential equation for the probability of the
fraction of the employed. This is what we call the master equation. In Weidlich and
Haag (1983) we find examples of a birth-and-death stochastic process, which has a
simple master equation, being adapted to model diffusion of opinion or information
such as brand choices of consumer goods or voting for political candidates among
a population. Recent examples in which master equations are used in economic
models are Kirman (1993), Weidlich (1994), and Aoki (1995, 1998, and 1999).
holds for all x and y. This is called the detailed balance conditions. See Kelly (1979,
Sec.1.5) for Kolmogorov criteria for stationary Markov chains and processes to
satisfy the detailed balance condition.
Given an irreducible Markov chain, for any state xi there is a finite sequence
of states which reaches it from some initial state, x0 , x1 , . . . , xi . If the detailed
balance condition holds, we have
i−1
Pe (xi ) = Pe (x0 ) [w(xk+1 |xk )/w(xk |xk+1 )].
k=0
with
w(xk+1 |xk )
−βN [U (xi ) − U (x0 )] = ln , (8.3)
w(xk |xk+1 )
where β is a parameter introduced to embody uncertainty or imprecision involved in
making decisions, seeAoki (1996, 138), for example. Note that (8.3) is independent
of paths from x0 to xi , i.e. U (x) is a potential. See the Kolmogorov criterion in
Kelly (1979, 21).
with k−1
k η
g := ln N
k + O(1/N ).
N 1−η N
Parameter β is introduced to indicate the degree of uncertainty about the relative
merits of alternative choices. See Aoki (1996, 138) or Aoki (1998, 2000) for further
discussion on this parameter.
We can show that the local minima of the potential corresponds to locally stable
equilibria of the aggregate dynamic equation which is derived next.
Asymmetrical cycles and equilibrium selection 209
In models with several types of agents or several choices, suppose we use
x(t) = (x1 (t), . . . , xK (t)) as
the vector. The equilibrium distribution Pe (x) in
some cases has product form k πk (xk ). See Kelly (1979) and Pollett (1986) for
several examples of this representation. Even when this product form is not exact,
it may serve as an approximation in some cases. In terms of the alternative state
vector a mentioned at the end of the previous section, there is multi-variate Ewens
distribution which seems to hold much promise in economic modeling.Aoki (1996,
238) has some preliminary account of this distribution. In problems with agents
of many types in which random partitions of agents into types or choices are
important, the Ewens distributions invariably arise. Kingman (1978) defines the
notion of random partitions. See Aoki (1999, 2000) for economic examples.
must take the partial derivative with respect to time by keeping x(t) fixed, i.e. we
must impose the relation
dξ dφ
= −N 1/2 ,
dt dt
and we obtain
dP dΠ ∂Π dξ ∂Π dφ ∂Π
= = = − N 1/2 .
dt dt ∂ξ dt ∂t dt ∂ξ
We also note that we need to rescale time by
τ = N −1 f (N )t.
Otherwise, the random variable ξ will not be of the order O(N 0 ) contrary to our
assumption, and the power series expansion will not be valid. But, f (N ) = N in
this section. In general τ = t. We use τ from now on to accommodate this more
general scaling function.
The master equation in the new notation is given by
∂Π(ξ, τ ) dφ ∂Π ∂ 1 ∂2
− N 1/2 = −N 1/2 {α1,0 (x) · Π} + {α2,0 (x) · Π}
∂τ dτ ∂ξ ∂ξ 2 ∂ξ 2
1 −1/2 ∂ 3 −1/2 ∂
− N α3,0 (x) · Π − N α1,1 (x) · Π
3! ∂ξ 3 ∂ξ
+ O(N −1 ),
where x = φ(τ ) + N −1/2 ξ, and where we define the moments of the transition rates
by
αµ,ν = r µ Φν (x; r)dr. (8.4)
Asymmetrical cycles and equilibrium selection 211
See van Kampen (1992, 253) for the terms not shown here.7
In this expression we note that the dominant term O(N 1/2 ) on the both sides are
equated. In the next section, the first moment of the leading term of the transition
rate wN (X |X ), α10 (φ), will be shown to determine the macroeconomic equation.
A binary choice model continued With the transition rates of Example 1, the
macroeconomic equation becomes
dφ
= (1 − φ)η(φ) − φ[1 − η(φ)]. (8.6)
dt
In terms of g(·) introduced in Example 1, the critical points of the macroeconomic
equation is
φ
exp[βg(φ)] = .
1−φ
Approximate analysis
First we recognize that we need to calculate only the event from one of the equilib-
rium state to the boundary between two basins of attraction, ψ, which is introduced
in the example above. The reason is the same one used by van Kampen (1992) as
quoted in Aoki (1996, 151). The time needed for φ to reach its equilibrium value,
φ1 , or φ2 depending on the initial value, is much shorter than the time needed to
go from one basin of attraction to the other.
A quick way to see this is to solve the deviational equation for φ. To be definite,
suppose that φ is in the domain of attraction to φ1 and let x := φ − φ1 . Then, it is
governed by
dx/dτ = Φ1 (φ1 )x = −A(φ1 )x,
with the initial condition x(0) = φ(0) − φ1 .
214 Masanao Aoki
The solution is x(τ ) = x(0) exp[−A(φ1 )τ ]. Recalling that exp(−4. 5) = 0. 01,
it takes about τ = 4. 5/A(φ1 ) to reduce the distance from φ1 to about 0.01 of the
original value. In the case where a = 1 and p = 0. 2, we have φ1 = 0. 358 and
A(φ1 ) = 0. 92. Thus, it takes about 4.9 or 5 time units to reach the equilibrium
point. As we show later in example 2, the mean first passage time for this example
is of the order of 103 when N = 100. Therefore we are justified in assuming that
φ is initially at one of the equilibrium points in calculating the mean first passage
time. The procedure is as outlined in Appendix in Aoki (1998). We set up a two
state Markov chain, because there are two locally stable equilibria in the example.
Let π1 and π2 = 1 − π1 be the probability that the employed fraction are in basins
of attraction for φ1 and φ2 , respectively. These probabilities evolve according to
the differential equation,
where W1,2 is the transition rate from φ1 to the boundary of the two basins of
attractions, that is, ψ, and W2,1 is that from φ2 to ψ. In the stationary state dπ1 /dτ =
0, we have
1
π1 = .
W1,2 /W2,1 + 1
The mean first passage time is given by
1
τ1,2 = .
W1,2
W1,2 = Pr[ξ ≥ ξc ],
with
ξc
φ1 + √ = ψ.
N
√
Analogously W2,1 is approximated by√the probability that ξ is smaller than N (ψ −
φ2 ), or equivalently it is larger than N (φ2 − ψ).
and
√
W2,1 = Pr ξ < N (ψ − φ2 )
∞
1 ξ2
= √ exp − dξ.
N (φ2 −ψ) 2πσ 2 (φ2 ) 2σ 2 (φ2 )
It is easy to see that both W1,2 and W2,1 approach zero as N is brought to in-
finity. Hence, we can approximate limN →∞ W1,2 /W2,1 by limN →∞ (dW1,2 /dN )/
(dW2,1 /N ). This is given by,
The larger the distance between the critical point and the boundary of basins
of attraction, or the smaller the variance of fluctuation around the critical point, it
is likely that this critical point is selected as an equilibrium in a model with infinite
number of agents.
Notes
1 There are many references on master equations. Van Kampen (1992, 97) tells us the
origin of the word ‘master’. See also Weidlich and Haag (1983), or Kubo (1975).
2 This fraction is an example of variables called intensive variables. This is often implicit
in the search literature, where it is routinely assumed that there are an infinite number
of agents, and fractions of one kind or another are posited from the beginning, such as
the fraction of employed, or fractions of agents which hold some specified assets, and
so on. See Diamond (1982) or Kiyotaki and Wright (1993) for example.
Asymmetrical cycles and equilibrium selection 217
3 When the size N is fixed, this assumption is innocuous because it can be absorbed into
time units. If N is a random variable as in models open to entry and exit by agents, it
must be explicitly incorporated into the transition rates. Kelly (1979) has an example
in which N is a random variable.
4 When these terms are zero, we may want to retain terms of the order N −2 . Then we have
diffusion equation approximations to the master equation. Diffusion approximations
are not discussed here.
5 We need not be precise about the initial condition since an expression of the order
O(N −1 ) or O(N −1/ 2 ) can be shifted between the two terms without any consequence.
Put differently, the location of the peak of the distribution can’t be defined more
precisely than the width of the distribution which is of the order N 1/ 2 .
6 Following the common convention that the parameters of the density are not carried as
arguments in the density expression, we do not explicitly show φ when the substitution
is made.
7 Note that
∂ ∂Π ∂
{α10 (x)Π} = α10 (φ) + N −1/ 2 α10
(φ) (ξΠ) + O(N −1 ).
∂ξ ∂ξ ∂ξ
References
Aoki, M. (1995) ‘Economic Fluctuations with Interactive Agents: Dynamic and Stochastic
Externalities’, Japanese Economic Review, 46, 148–65.
Aoki, M. (1996) New Approaches to Macroeconomic Modeling: Evolutionary Stochastic
Dynamics, Multiple Equilibria, and Externalities as Field Effects , Cambridge University
Press, New York.
Aoki, M. (1998), ‘Simple Model of Asymmetrical Business Cycles: Interactive Dynamics
of a Large Number of Agents with Discrete Choices’, Macroeconomic Dynamics, 2,
427–42.
Aoki, M. (1999), ‘Open Models of Share Markets with Several Types of Participants’,
presented at Workshop on Heterogenous Interacting Agents, Univ. Genoa, June.
aoki, M. (2000) ‘Cluster Size Distributions of Economic Agents of Many Types in a Market’,
Journal of Mathematical analysis and Applications, 249, 32–52.
Aoki, M. (2001) Modeling Agregate Behavior and Fluctuations in Economics: Stochastic
Views of Interacting Agents, Cambridge University Press, Cambridge.
Aoki, M. and Y. Shirai (2000), ‘Stochastic Business Cycles and Equilibrium Selection in
Search Equilibrium’, Macroeconomic Dynamics, 4, 487–505.
Breiman, L. (1968) Probability Theory, Addison-Wesley, Reading, MA.
Cox, D. R., and H. D. Miller (1967) The Theory of Stochastic Processes, Chapman & Hall,
London.
Diamond, P. D. (1982) ‘Aggregate Demand Management in Search Equilibrium’. Journal
of Political Economy, 90, 881–94.
Feller, W. (1970) Introduction to Probability Theory and Its Applications, Vol I, John Wiley
& Sons, New York.
Grimmett, G. R. and D. R. Stirzaker (1992) Probability and Random Processes, Oxford
University Press, New York.
Karlin, S. and H.Taylor (1981) A Second Course in Stochastic Processes, Academic Press,
New York.
Kelly, F. (1979) Reversibility and Stochastic Networks, Wiley & Sons, New York, 1979
Kelly, F. (1976) ‘On Stochastic Population Models in Genetics’, J. Appl. Probab., 13,127–
31.
218 Masanao Aoki
Kendall, D. G. (1975) ‘Some problems in mathematical genealogy’, in J. Gani (ed.) Per-
spective in Probability and Statistics, Academic Press, New York.
Kingman, J. F. K. (1978) ‘Random Partitions in Population Genetics’, Proc. London. Ser.
A, 361, 1–20.
Kirman, A. (1993) ‘Ants, Rationality, and Recruitment’, Quart.J.Econ., 103, 137–56.
Kiyotaki, N., and R. Wright (1993) ‘A Search-TheoreticApproach to Monetary Economics’,
Amer. Econ. Review, 83, 63–97.
Kubo, R. (1975) ‘Relaxation and Fluctuation of Marovariables’ in Lecture Notes in Physics
No.39, H. Araki (ed.), Springer-Verlag, Berlin.
Pollett, P. K. 1986) ‘Connecting Reversible Markov Processes’, Adv. Appl. Probab. 18,
880–900.
van Kampen, N. G. (1965) ‘Fluctuations in Nonlinear Systems’, Chapter V in R.G. Gurgess
(ed.) Fluctuation Phenomena in Solids, Academic Press, New York.
van Kampen, N. G. (1992) Stochastic Processes in Physics and Chemistry: Revised and
enlarged edition North Holland, Amsterdam.
Weidlich, W. (1994) ‘Synergetic Modelling Concepts for Sociodynamics with Application
to Collective Political Opinion Formation’, J.Math.Sociology, 18, 267–91.
Weidlich, W. and G. Haag (1983) Concepts and Models of a Quantitative Sociology,
Springer-Verlag, Berlin/Heidelberg/New York.
Whittle, P. (1976) Probability via Expectation, Second Edition, Springer-Verlag, New York.
9 The instability of markets
Tad Hogg, Bernardo A. Huberman and
Michael Youssefmir
9.1 Introduction
The explosive growth of computer networks and of new forms of financial products,
such as derivatives, is leading to increased couplings among previously dispersed
markets. This increased fluidity raises questions about the stability and efficiency
of the international financial system. On the one hand, it is quite apparent that
increased overall connectivity among markets allows transactions that were not
previously possible, increasing the net wealth of people and leading to more effi-
cient markets. But at the same time, this implies that a transition is taking place,
from a more static scenario in which isolated markets can be considered in equi-
librium on their own, to a global economy that knows of no geographical borders.
One may wonder about the nature of this transition, i.e. is it gradual in the sense of
a smooth change in prices, or punctuated by abrupt changes in the value of certain
commodities, cascading bankruptcies and market crashes?
Underlying these questions is the old problem of the existence and stability
of equilibria in markets, which general equilibrium theory has addressed under
various conditions (Arrow and Hahn 1971, Arrow 1988) . In this chapter, we will
focus on the stability of markets by treating the couplings among them as dynamical
entities in the spirit of other evolutionary approaches (Day 1975, Nelson and Winter
1982, Brock 1988) .
The standard approach to characterizing the stability of markets postulates that
agents opportunistically optimize their portfolios in such a way so as to minimize
their risk while maintaining a given return. But in real life agents are not always
able to perfectly process information about the system in which they are embed-
ded. Under such conditions, adaptive agents continuously switch between different
behavioral modes in response to a constantly changing environment, favoring be-
haviors that can temporarily lead to increased rewards. It follows that while agents
are continuously learning about the relative merits of different commodities and
markets, the couplings between such commodities evolve in rather complicated
ways.
Other examples of such evolving couplings are provided by agents linking
limited baskets of commodities efficiently while ignoring other commodities in the
220 Tad Hogg, Bernardo A. Huberman and Michael Youssefmir
process. Of more recent interest are the couplings that agents introduce through
complicated derivatives created in an effort to optimize and hedge away risks.
Due to their complex nature such derivatives can be poorly understood by the
people who use them, once again leading to extraneous couplings. Finally, highly
leveraged hedge funds can also introduce couplings into the market by being forced
to cover certain leveraged positions with other unrelated positions.
In this chapter, we assume the existence of an equilibrium within a network of
markets and examine its dynamic stability through the use of a general model of
the adjustment processes within the markets. We focus on cases where agents are
not perfectly rational or identical and therefore can introduce couplings that may
not always represent the best possible allocation of their resources. We then discuss
how the stability of such a system scales as different markets become increasingly
coupled. We show that the class of systems that are stable becomes smaller and
smaller as the number of coupled markets scales up. These instabilities are shown
to exist even when couplings are so weak that the markets are near decomposable.
Such instabilities in turn require a heightened degree of learning on the part of
market participants, a process in which the learning itself is marked by periods of
instability until equilibrium is reestablished. These results, which run counter to
prevailing notions of stability in large coupled markets, offer a cautionary note on
treating emerging markets with the tools of equilibrium economics.
This implies that as the matrix gets large enough, its largest eigenvalue will become
positive.
A more realistic case relaxes the requirement that the changes in one price are
on average balanced by changes in others. This corresponds to having a non-zero
mean. In this situation a theorem of Furedi and Komlos (1981) states that, on
average, the largest eigenvalue is given by
E = (n − 1)µ + σ 2 /µ − D (9.4)
E ∼ µ(n − 1) − D (9.5)
Since µ is positive these results imply that even if markets are stable when small,
they will become unstable as their size becomes large enough for E to change sign.
One argument that could be given for the stability of coupled markets in spite
of their size is that not all commodities happen to be coupled to each other. In
terms of our theory, this amounts to a near decomposability of the Jacobian matrix
whereby the entries are such that the further they are from the diagonal the smaller
they become (Simon and Ando 1961). Such systems, sometimes called loosely
coupled, are very relevant to situations when markets are initially weakly coupled
to each other. But as we will now show, even in this case, as the size of the markets
increases the equilibrium will be rendered unstable. In terms of the interactions, this
situation can describe either the fact that a given item’s price is strongly influenced
by a few others and weakly by the rest, or a more structured clustering, where
the commodities appear in groups (e.g. technology stocks, foreign currencies)
wherein their members strongly interact but members of different groups have
weak interactions with each other.
Another way of considering this scenario is to imagine initially isolated mar-
kets that eventually get coupled through the interaction of mediating interactions,
224 Tad Hogg, Bernardo A. Huberman and Michael Youssefmir
such as roads and communications. In this case the coupling between the initially
isolated markets grows in time.
The corresponding matrices for the first case are constructed by selecting off-
diagonal entries at random with large magnitude 1 with probability p and small
value < 1 otherwise. In this case µ = p + (1 − p) and σ 2 = p(1 − p)(1 − )2 .
As shown in (9.5), the largest eigenvalue will become positive when the system
becomes large enough.
In the second case, the commodities are grouped into a hierarchical structure
which we assume to be of depth d and average branching ratio, b. In this representa-
tion, the strength of the interaction between two commodities will decrease based
on the number of levels in the hierarchy that one has to climb to reach a large
enough common group to which they both belong. Specifically, the interaction
strength will be taken to be Rh , with h the number of hierarchy levels that separate
the two commodities, and R characterizes the reduction in interaction strength that
two commodities undergo when they are separated by one further level. The aver-
age size of the matrix is given by n = bd and the mean of the non-diagonal terms
can be computed to be
d
h=1 b
h−1
(b − 1)Rh (Rb)d − 1 R(b − 1)
µ= = (9.6)
d
h=1 b
h−1 (b − 1) bd − 1 Rb − 1
but in slower fashion than the case above. Nevertheless, this subtle difference
in convergence to zero amounts to a qualitative difference in the stability of the
system. To see this, notice that (9.5) implies that the largest eigenvalue of a random
matrix grows as µn, which increases with n for this value of µ, thus leading to
instability when the system gets large enough. The growth in largest eigenvalue
with the size of the system is exhibited in Figure 9.1 for a particular choice of
The instability of markets 225
5
3
largest eigenvalue
3
0 1 2 3 4 5 6 7 8 9 10
d
Figure 9.1 Plot showing the growth of the largest eigenvalue of a hierarchical matrix with
branching ratio b = 2 and R = 0. 55 as a function of d = log2 n. The diagonal
elements were chosen to be D = 3.The points are the theoretically predicted
values of Eq. 9.6 and lie very close to the computed eigenvalues shown by the
solid line.
parameters. Notice that the system becomes unstable for d ≥ 5. Given these results
we see that the size of the matrix for which this instability takes place is much
larger than the one in the absence of a hierarchy of interactions.
A final possibility is for the commodities to include aggregate structures, such
as stock indices. In this case the couplings will correspond to situations where each
commodity interacts with itself, the other components of its aggregate, or its higher
order aggregate. The ensuing Jacobian will have blocks of nonzero elements and
zero entries elsewhere. For this case, we have shown that an even slower growth
of the largest eigenvalue with size is obtained (Hogg,√Huberman and McGlade
1989). Specifically, the eigenvalues grow no faster than ln n, implying that much
larger coupled markets can be stable when they are structured in a highly clustered
fashion. Fluctuation corrections make the eigenvalues grow like ln n, still implying
a higher degree of stability than in the previous cases.1
9.4 Discussion
Recent dramatic fluctuations and losses in the world financial markets have raised
concerns about the inherent stability of the global financial system. These concerns
226 Tad Hogg, Bernardo A. Huberman and Michael Youssefmir
have been prompted by the heightened fluidity of global currency flows and the
emergence of complicated derivatives, which now allow market players to make
financial bets as never before. For example, the derivatives debacle in Orange
County, California points to the fact that some market players do not understand
the full risks that are being taken. At the same time, the result of the current trends
in global finance is that markets are now more and more coupled and individual
governments have less and less power to control these perhaps destabilizing cou-
plings. The recent economic crisis in Mexico, the ‘tequila effect’, resulted in added
volatility and ‘corrections’ in many emerging equity markets all over the world.
In this chapter, we modeled this system as a web of interacting markets much
like a biological ecosystem (Rothschild 1990). By doing so we obtained a gen-
eral result showing that as couplings between previously stable markets grow, the
likelihood of instabilities is increased.
These results allow us to understand phenomena that are likely to arise as a
system grows in diversity, strength of couplings and in the size of the overall number
of the coupled markets. In a sense these results appear to be counterintuitive, for
one expects that as a system gets larger, disturbances in a particular part would
exhibit a kind of decoherence as they propagate through the system, making it very
unlikely that after a given time they would once again concentrate on a particular
node and amplify it. But the properties of random matrices make it probable for
this conspiracy of perturbations to concentrate on given parts of the market. As the
size and diversity of couplings in such matrices grow, the complicated effects of
the couplings are more and more likely to result in instability that leads to motion
away from equilibrium.
One may ask about the fate of the lost equilibrium and the ensuing evolution of
an unstable market. We speculate that once the instability sets in adaptive agents
will modify the respective couplings in such a way as to stabilize the entire system
once more at another equilibrium. If this is indeed the case, the volatility brought
about by instabilities in such large systems is the natural mode by which couplings
are modified to achieve a more efficient market. Due to the lack of information
needed for appropriate centralized control, it is also by no means clear that regu-
latory approaches to controlling the market structures will be effective. Improper
controls could introduce additional couplings in such an uncertain way that the
system may be further destabilized. Accepting the instabilities of these larger sys-
tems as the best way for market participants to learn the correct couplings may,
therefore, be the most reasonable course of action.
Last but not least, these results cast light on the related problem of the dynamics
of multiagent systems in distributed computer networks, which have been shown
to behave like economic systems (Huberman and Glance 1995). In the case of
only two resources their equilibrium is punctuated by bursts of clustered volatility
(Youssefmir and Huberman 1997), a fact which renders the notion of equilibrium
suspect. This paper shows that as distributed computing systems get large and more
coupled, they will also exhibit a loss of stability, on their way to finding a new and
more efficient equilibrium.
The instability of markets 227
Notes
1 A further interesting possibility (Trefethen et al 1993) is that the linearized system can
produce an initially growing transient even when E < 0 so the perturbations eventually
decay. During this transient growth, the perturbations may become large enough to be
sustained by nonlinear corrections, thus giving another possible source of instability.
References
Arrow, K. J. and F. H. Hahn (1971) General Competitive Analysis. Holden-Day, San Fran-
cisco.
Arrow, K. J. (1988) ‘Workshop on the economy as an evolving complex system: Summary’,
in P. W. Anderson, K. J. Arrow, and D. Pines, (eds) The Economy as an Evolving Complex
System, 275–81. Addison-Wesley.
Arrow K. J. and Leonid Hurwicz (1958) ‘On the stability of competitive equilibrium.’
Econometrica, 26:522–52.
Brock, W. A. (1988) ‘Nonlinearity and complex dynamics and economics and finance.’ in
P. W. Anderson, K. J. Arrow, and D. Pines, (eds), The Economy as an Evolving Complex
System, 77–97. Addison-Wesley.
Cohen, J. E. and C. M. Newman (1984) ‘The stability of large random matrices and their
products.’ Annals of Probability, 12, 283–310.
Day, R. H. (1975) ‘Adaptive processes and economic theory.’ in R. H. Day and T. Groves,
(eds), Symposium on Adaptive Economics, 1–35. Academic Press.
Edelman, A. (1989) Eigenvalues and Condition Numbers of Random Matrices. PhD thesis,
MIT, Cambridge, MA 02139, May.
Furedi, Z. and K. Komlos (1981) ‘The eigenvalues of random symmetric matrices.’ Com-
binatorica, 1, 233–41.
Hogg, T. B. A. Huberman and Jacqueline M. McGlade (1989) ‘The stability of ecosystems’
Proc. of the Royal Society of London, B237, 43–51.
Huberman, B. A. and Natalie S. Glance (1995) ‘The dynamics of collective action.’ Com-
putational Economics, 8,27–46.
Juhasz, F. (1982) ‘On the asymptotic behavior of the spectra of non-symmetric random
(0,1) matrices’ Discrete Mathematics, 41, 161–5.
May, R. M. (1972) ‘Will a large complex system be stable?’ Nature, 238, 413–14.
McMurtrie, R. E. (1975) ‘Determinants of stability of large randomly connected systems’
J. Theor. Biol., 50, 1–11.
Mehta,. M. L. (1967) Random Matrices and the Statistical Theory of Energy Levels, Aca-
demic Press, New York.
Metzler, L. A. (1945) ‘Stability of multiple markets: The Hicks condition’ Econometrica,
13, 277–92.
Nelson, R. A. and S. G. Winter (1982) An Evolutionary Theory of Economic Change,
Harvard University Press.
Quirk, J. (1981) ‘Qualitative stability of matrices and economic theory: A survey article.’
in H. J. Greenberg and J. S. Maybee, (eds), Computer-Assisted Analysis and Model
Simulation, 113–64. Academic Press, NY.
Rothschild, M. L. (1990) Bionomics: Economy as Ecosystem. Henry Holt and Company.
Samuelson, P. A. (1941) ‘The stability of equilibrium: Comparative statics and dynamics.’
Econometrica, 9, 111.
228 Tad Hogg, Bernardo A. Huberman and Michael Youssefmir
Simon, H. A. and Albert Ando (1961) ‘Aggregation of variables in dynamic systems.’
Econometrica, 29, 111–38.
Trefethen, L. N., A. E. Trefethen, S. C. Reddy, and T. A. Driscoll (1993) ‘Hydrodynamic
stability without eigenvalues.’ Science, 261, 578–84, July 30.
Wigner, E. P. (1958) ‘On the distribution of the roots of certain symmetric matrices.’ Annals
Math, 67, 325–7.
Youssefmir, M.and B. A. Huberman (1997) ‘Clustered volatility in multiagent dynamics.’
J. of Economic Behavior and Organizations, 914, 101–18.
10 Heterogeneity, aggregation
and capital market
imperfection
Domenico Delli Gatti and
Mauro Gallegati
10.1 Introduction
The representative agent framework has a long tradition (Marshall, 1920) and has
been one of the most successful tools in economics (Hartley, 1997). 1 It is still
the cornerstone of microfoundations in macroeconomics and of aggregation in the
classical literature (see e.g. Lucas, 1975; Kydland and Prescott, 1982; Long and
Plosser, 1983) because the aggregation process allows any difference between the
behaviour of individually optimizing agents and that of aggregate variables to be
ruled out.
Despite its success, economists are growing more and more dissatisfied with
the representative agent framework2 for a number of different reasons.
First of all, the set of assumptions necessary to reach exact aggregation in a
logically consistent way is impressive.3 Martel (1996: 128) lists the following:
• homothetic preferences (i.e. linear Engel’s curve);
• linearly homogeneous production functions, identical for all firms;
• homogeneous and infinitely divisible commodities and factors of produc-
tion;
• a common set of prices with constant relative ratios;
• constant distribution of income and endowments over time.
Of course, none of the requirements above matches real world features. Empiri-
cal analysis at the disaggregated level has shown that heterogeneity matters and
aggregation of heterogeneous agents is of central relevance, since there is sys-
tematic evidence of individual differences in economic behaviour (Stoker, 1993).
In econometrics, several contributions show that aggregate dynamics and indi-
vidual heterogeneity are intertwined. On the one hand, neglecting heterogeneity
230 Domenico Delli Gatti and Mauro Gallegati
in aggregate equations generates spurious evidence of dynamic structure. On the
other hand, aggregation of very simple individual behaviour may lead to aggregate
complex dynamics.4
Apart from the obvious objections to the Representative Agent Hypothesis
(RAH) emerging from the empirical evidence, there are theoretical reasons to
reject the use and misuse of the representative agent.
First of all, Arrow, 1951 (and later Sonnenschein, 1972, and Mantel,1976)
have shown that the proposition according to which the properties of an aggregate
function will reflect those of the individual function has no theoretical foundations.
Second, ignoring heterogeneity may lead to a fallacy of composition (see Weiss,
1967) which may be relevant when agents’decisions are not perfectly synchronized
– so that composition effects arise (Caballero, 1992) – and misleading for empirical
analysis.5
Third, the representative agent framework can be unsatisfactory for a wide
range of economists, even belonging to opposite sides of the academic spectrum.
The very idea of asymmetric information of the new keynesian economics (Mankiw
and Romer, 1991) is inconsistent with the representative agent hypothesis (Stiglitz,
1992). In fact, the asymmetric information context is based on heterogeneity; only
if agents are different from one another may phenomena such as adverse selection
or moral hazard exist.
From a different theoretical perspective, the general equilibrium theorist may
not feel at ease with the representative agent assumption because some of the
building blocks of general equilibrium theory do not hold in the presence of the
representative agent (e.g. the ‘Weak Axiom of Revealed Preferences’ or Arrow’s
‘Impossibility Theorem’, Kirman, 1992: 122).
Therefore there are theoretical as well as empirical reasons to question the reli-
ability of the representative agent as an economic tool. This chapter reviews some
macroeconomic models in which agents’ heterogeneity is the main ingredient.
The literature on heterogeneous agents is burgeoning. We can distinguish a
number of different strands of this literature: models of distribution and growth,6
employment and aggregate supply, 7 aggregate demand, 8 and macrodynamics,9
capital market imperfections and business fluctuations.10 Some authors identify
a law of motion of the average level of the state variable, keeping constant the
distribution; others analyze the dynamics of distribution alone, making inferences
on aggregate dynamics.
The assumption of a constant distribution of income, wealth or endowments
over time is quite implausible when coping with dynamics. In a sense, agents’
heterogeneity is the logical requirement for dynamic analysis. According to the
impulse-propagation approach small shocks may lead to large fluctuations once
a sufficiently large amplification mechanism is activated. If agents are heteroge-
neous, the propagation mechanism is affected by changes in the distribution of
agents. In such a case, business cycles are not ‘all alike’ (Lucas, 1977). Moreover,
if we give up the RAH, fluctuations and cycles may emerge as the result of changes
in the distribution of agents rather than the consequence of aggregate (as well as
idiosyncratic) exogenous shocks to which a ‘representative’ individual reacts, as
Heterogeneity, aggregation and capital market imperfection 231
in the standard literature on the business cycle. In other words, if the RAH is
abandoned, the change in distribution may produce endogenous fluctuations. The
literature we review in the next sections emphasizes these aspects.
In the following we survey extensively a class of models in which heterogene-
ity and aggregation play a crucial role. We deal with macroeconomic models based
on capital market imperfections due to asymmetric information. Greenwald and
Stiglitz, 1988, 1990, 1993 (section 2), Bernanke and Gertler, 1989, 1990 (section
3), Kiyotaki and Moore, 1997 (section 4), have put forward an approach which
emphasizes the role of financial fragility through the interaction of heterogenous
agents. These authors, however, don’t push the analysis to the point of identify-
ing the law governing the distribution of agents. On the other hand, models by
Galor and Zeira, 1993, and Aghion and Bolton, 1997 (section 5) emphasize how
the distribution of income and wealth across agents affects growth, i.e. the dy-
namic behaviour of the system. Section 10.6 discusses an approach which is at the
crossroad of the existing literature: while providing an aggregation mechanism, it
identifies the law of motion of agents’ distribution and shows how aggregation of
heterogenous individuals affects the dynamics. Section 10.7 concludes.
Πt+1 = Pt+1 Yt − (1 + iL ) Bt
i i i i
where R ≡ (1+r) = (1 + iL ) PPt+t 1 is the gross real interest rate, which will turn out to
be equal to the rate of time preference and exogenous. Recalling that bit = wt Nti −Ait
and that Nti = Φ(Yti ) the expression above can be written as:
πt+1 = ut+1 Yt − R wt Φ(Yt ) − At
i i i i i
Heterogeneity, aggregation and capital market imperfection 233
The relative price being stochastic, also profit is a random variable with expected
value:
E(πt+1 ) = Yt − R[wt Φ(Yt ) − At ]
i i i i
In other words the firm goes bankrupt if the relative price in t+1 happens to be
i
lower than a critical threshold ūt+1 , which coincides with debt service per unit of
output. The probability of bankruptcy is13
R[wt Φ(Yti ) − Ait ]
Pr(ut+1 < ūt+1 ) = G(ūt+1 ) = G , G > 0.
i i i
Yti
In other words, the probability of bankruptcy is increasing with the interest rate
and decreasing with the equity base. It is increasing also in the level of output if
the elasticity of the input requirement to the level of output is greater than one.
Finally, GS assume there are bankruptcy costs which are increasing with the
scale of production:14 CBti = c(Yti ), c > 0.
1
Solving the problem yields: Yti = Φ−1 Rw .Therefore Nti = F −1 (Rwt ) . It is easy
to see that for each firm output supply (and labour demand) is a decreasing function
of the real interest rate and the real wage, which are uniform across firms. In
other words employment and production decisions react only to changes of the
real interest rate and the real wage. Heterogeneity of financial conditions, albeit
present, is irrelevant.
Notice that there are two sources of heterogeneity in this framework. The
first one is the random relative price, which can be thought of as an idiosyncratic
shock. This source of heterogeneity is irrelevant in this context when GS consider
the profit maximization problem of the firm. By assumption, in fact, each firm is
risk neutral and its expected relative price is one (expectations are homogeneous
and rational). The second source of heterogeneity is the equity base. Firms can
differ, in fact, according to their degree of financial robustness as measured by
the equity base. In the first best case, however, also this source of heterogeneity is
irrelevant because there are no bankruptcy costs.
In this case, individual agents, albeit heterogeneous, may be modeled as iden-
tical with no loss of generality. Aggregation would yield an aggregate supply and
labour demand functions which keep the basic features of the individual functions.
234 Domenico Delli Gatti and Mauro Gallegati
Let’s now turn briefly to the household sector. The representative infinitely
lived household maximizes the discounted sum of utilities – which are assumed
to be linear in consumption and labour supplied – over an infinite horizon subject
to a budget constraint according to which total wealth consists of labour income
and the return on assets. According to the solution of this dynamic programming
problem:
• in each period, consumption is equal to income (output) of the previous
period;15
• the real interest rate is equal to the (given and constant) rate of time prefer-
ence;
• the supply of labour is an increasing function of the real wage only.
fe fe fe fe
Equilibrium on the labour market yields wt = w (R) and Nt = N (R) where the
fe stands for full employment. As a consequence
fe fe fe
Yt = F N (R) = Y (R)
is full employment output. It is easy to see that full employment output is a de-
creasing function of the real interest rate, provided certain plausible conditions on
the configuration of parameters are met.
The law of motion of the equity base of the i-th firm in real terms can be written
as follows:
At+1 = ut+1 Yt − R wt Φ(Yt ) − At − Dt+1
i i i i i i
(10.1)
where Ait+1 , the equity base in period t+1, is equal to retained profit, which in
i
turn is equal to total profit less dividends Dt+1 . GS assume that dividends are an
increasing function of the equity base Dt+1 = d(Ait ), d > 0.
i
The law of motion of the aggregate equity base can be derived from (10.1)
through aggregation. Assuming that the labour market is in equilibrium, the law
of motion is:
At+1 = Y fe (R) − R wfe (R) N fe (R) − At − d(At ) (10.2)
(10.2) is a non linear difference equation in the state variable At , the aggregate
(average) equity base, parametrized to the real interest rate. The steady state solu-
tion(s) will be generically written as follows: A∗ = A∗ (R). In the steady state the
equity base on average depends negatively upon the real interest rate. Under quite
general assumptions, the steady state is stable.
Yt = Y i (Rwt , At ).
i i
Therefore
Nt = Φ Y i (Rwt , At ) = N i (Rwt , At )
i i i
In this case, for each firm output and labour demand are increasing functions of
the equity base, given the real interest rate and the real wage, which are uniform
across firms.
Notice that firms can differ in their production and employment decisions due
to the different degrees of financial robustness as measured by the level of the equity
base. This is the only source of heterogeneity which persists once the problem of
the firm is solved as it is clear from the formulae above. In the first best case, on the
contrary, production and employment decisions were independent from the equity
base.
According to GS, aggregation would yield an aggregate supply function and
a labour demand function which keep the basic features of the corresponding
individual functions. Therefore aggregate (average) output and labour demand are
increasing functions of the aggregate (average) equity base, given the real interest
rate and the real wage, which are uniform across firms. GS would be perfectly
right if the equity base were the same for each and every firm (i.e. if they had
assumed the representative agent framework from the beginning) – in which case
there would not be any difference between the individual and the average relations
– and/or if the individual relations were linear. If financial robustness is different
from one firm to the other, however, and if the Y i (. ) function is non-linear, also
the higher moments of the distribution of the equity base across agents should be
taken into account in the aggregation procedure. In particular, the variance of the
equity base across agents should show up in the aggregate relation.
If we take GS’ aggregation procedure at face value and ignore the higher
moments of the distribution, equilibrium on the labour market, yields wtfe =
wfe (R, At ) , Ntfe = N fe (R, At ). As a consequenceYtfe = Y fe (R, At ) is full employment
output. If we amend GS’ aggregation procedure to take heterogeneity seriously,
however, the variance of the equity base would be an important part of the story, it
should show up in the full employment formulae for real wage, employment and
output and composition effects would be crucial.
The law of motion of the equity base of the i-th firm in real terms has been
derived above (see eq.(10.1)). The law of motion of the aggregate equity base can
be derived from (10.1) through aggregation. Summation and averaging over firms
yields :
At+1 = Y fe (R, At ) − R wfe (R, At ) N fe (R, At ) − At − d(At ) (10.3)
236 Domenico Delli Gatti and Mauro Gallegati
where, once again, we have borrowed the aggregate formulae from GS, ignoring
the variance. (10.3) is a non linear difference equation in the state variable At ,
parameterized to the real interest rate, which can yield different types of dynamics.
The steady state solution(s) will be generically written as follows: A∗ = A∗ (R).In
the steady state the equity base on average depends upon the real interest rate. As
a consequence in the steady state also output and employment depend on the same
variables: N ∗ = N fe (R, A∗ (R)) = N ∗ (R), Y ∗ = Y fe (R, A∗ (R)) = Y ∗ (R).
A monetary shock affects the steady state only if unexpected (GS policy effec-
tiveness proposition).16
If we amend GS’ aggregation procedure to take heterogeneity seriously, how-
ever, the variance of the equity base would be an important part of the dynamic
story too. In fact, (10.3) should revritten as follows:
At+1 = Y fe (R, At , σA2 ) − R wfe R, At , σA2 N fe R, At , σA2 − At − d(At )
where σA2 is the variance of the distribution of the equity base in t.
πb is the probability of the lower outcome of the investment project, i.e. the prob-
ability of the ‘bad’ state of the world. Alternatively, we can think of πb as the
proportion of investment projects undertaken by entrepreneurs which yield the
lower outcome.
The return on investment is known to the entrepreneur. Non-entrepreneurs can
observe the return on investment only incurring monitoring costs equal to γ units
of capital. Therefore, capital in period t+1 is:
kt+1 = E k̃ − hγ It (10.4)
where I is the number of investment projects per capita undertaken, while h is the
proportion of projects monitored.
Each agent is endowed with 1 unit of labour which is supplied inelastically
(there is no disutility of labour). In equilibrium the real wage will be equal to the
marginal productivity of labour: wt = θ [f (kt ) − kt f (kt )].
Preferences are such that entrepreneurs save all their labour income when
young and consume all the return they receive on their project when old. Therefore
the entrepreneur’s saving (when young) is Ste = wt η.
such that all the entrepreneurs with inefficiency lower (higher) than ω̄t will invest
(not invest).
Thanks to the assumption according to which ωti is a random variable distributed
uniformly with support [0, 1], the critical degree of inefficiency ω̄t represents also
the proportion of entrepreneurs who invest. Therefore the distribution of agents
between entrepreneurs and non-entrepreneurs is exogenous while the distribution
of entrepreneurs between investors and non-investors is endogenous. The higher
the return on investment (the lower the return on storage), the greater the share of
entrepreneurs who invest in the population of entrepreneurs.
At this point of the analysis agents can be classified into three groups:
• non-entrepreneurs (1 − η).
The last two groups do not invest and therefore employ their saving in the storage
technology. In other words they lend their savings (supply loanable funds) to the
entrepreneurs. The distinction between borrowers (entrepreneurs who invest) and
lenders (entrepreneurs who don’t invest and non-entrepreneurs) is at least partially
endogenous.17 In a sense, in this framework, entrepreneurs who are ‘too inefficient’
– i.e. entrepreneurs whose degree of inefficiency is ωti > ω̄t – give up investment
and join the ranks of non-entrepreneurs in lending their funds to entrepreneurs who
are ‘enough efficient’ to invest – i.e. entrepreneurs whose degree of inefficiency is
ω i ≤ ω̄.
The supply of capital is obtained from (10.4) substituting It = η ω̄t and γ = 0.
We obtain:
kt+1 = E k̃ η ω̄t (10.7)
Substituting the definition of ω̄t into (10.7) and solving for qt+1 we get:
rx kt+1 /E k̃ η
qt+1 = (10.8)
E k̃
i.e. if the return on investment in the bad state of the world is high enough
to allow the entrepreneur to reimburse its debt. Manipulating (10.11) one realises
that full collateralization occurs if the volume of internal finance is higher than a
threshold which is increasing with the degree of inefficiency:
qt+1 kb
Ste ≥ x(ωti ) − = S(ωti )
r
In this case the probability of auditing is equal to zero: pa = 0.
Incomplete collateralization occurs if:
qt+1 kb < r x(ωti ) − Ste (10.12)
Entrepreneurs with inefficiency lower than ω t (good entrepreneurs for short) will
invest even if the probability of auditing is one, whatever their level of internal
finance. Fully collateralized good entrepreneurs will never be monitored. As far as
incompletely collateralized good entrepreneurs are concerned, BG show that they
are monitored with probability
r x(ωti ) − Ste − qt+1 kb
pa = = p(ωti , qt+1 , Ste ); ωti < ω t (10.16)
qt+1 [(1 − πb )(kg − kb ) − πb γ]
At this point of the analysis agents can be classified into six groups:
The taxonomy of agents in the imperfect capital markets case is much richer than
in the perfect capital markets case. The first three groups consists of entrepreneurs
who invest. Good entrepreneurs, whose share in total entrepreneurial population
is ω t , will always invest but the return on investment should be diminished to take
into account monitoring costs if the good entrepreneur is not fully collateralized.
The probability that a good entrepreneur is not fully collateralized is
ωt
−
p(ωti , qt+1 , Ste )dωti .
0
The last three groups do not invest and therefore employ their saving supplying
loanable funds to the entrepreneurs.
Therefore, the supply of capital will be:
⎡ ⎤
ωt
kt+1 = E k̃ η ⎣ω t − πb γ p(ωti , qt+1 , Ste )dωti ⎦ +
0
⎡ ⎤
ω̄t
⎢ ⎥
+E k̃ η ⎣ s(ωti , qt+1 , Ste )dωti ⎦ (10.18)
ωt
Imposing E k̃ = 1 and rearranging we get:
⎡ ⎤
ωt ω̄t
⎢ ⎥
kt+1 = ω̄t − ⎣πb γ p(ωti , qt+1 , Ste )dωti + [1 − s(ωti , qt+1 , Ste )]dωti ⎦ η (10.19)
0 ωt
The supply of capital in the imperfect capital markets case depends not only on the
price of capital as in the first best case but also on the volume of entrepreneurial
savings.
The demand for capital is obtained from the usual condition according to
which the marginal productivity of capital must be equal to the real remuneration
of capital (see (10.10) above).
242 Domenico Delli Gatti and Mauro Gallegati
Equilibrium in the market for capital yields19 :
∗ ∗
kt+1 = k(θ, Ste ); qt+1 = q(θ, Ste )
In the imperfect capital markets case, therefore, the price and the quantity of capital
are influenced not only by the state of technology as in the first best case but also
by the financial conditions of investing entrepreneurs captured by their savings.
For instance an increase in entrepreneurial savings would produce an increase in
the quantity of capital and a decrease of its price.
We recall that entrepreneurial savings in period t are equal to Ste = wt η and that
wt = θ [f (kt ) − kt f (kt )]. Therefore, the equilibrium stock of capital in period t+1
will be:
kt+1 = k (θ, ηθ [f (kt ) − kt f (kt )]) (10.20)
(10.20) is a generally non-linear first order difference equation in the stock of
capital. In this case, there are true dynamics and the volume of internal finance
(entrepreneurial savings, i.e. the wage bill of entrepreneurs who invest) has an
important role to play. The steady state of (10.20) will be k ∗ = k ∗ (θ).
where ctF is the farmer’s consumption. Substituting (10.21) into (10.22) we get the
budget constraint:
ctF = (α + c̄)kt−1
F
− µt ktF (10.23)
where µt = qt − qt+R 1 .
Preferences are such that farmers consume only non-tradable output, i.e. ctF =
c̄ kt−1 . The farmer’s demand for land, therefore, is:
F
1 α F
ktF = F
(α + qt )kt−1 − RbFt−1 = kt−1 (10.24)
µt µt
Substituting (10.24) into (10.21), we obtain:
qt+1 1
bFt = F
(α + qt )kt−1 − RbFt−1 (10.25)
R µt
G
The production function of each gatherer is: ytG = f (kt−1 ) where ytG is output of
G
the gatherer in t, f (. ) is a well behaved production function and kt−1 is land of
244 Domenico Delli Gatti and Mauro Gallegati
the gatherer in t-1. The gatherers’ human capital is not inalienable. Therefore,
gatherers face only a flow-of-funds constraint:
Preferences of the gatherer are such that Rµt = f (ktG ). The demand for land,
therefore, is:
−1
ktG = f (Rµt ) (10.28)
• skilled workers who inherit less than the value of the investment in human
capital. They are borrowers;
• skilled workers who inherit more than the investment in human capital. They
are lenders.
GZ assume that preferences are homogeneous across agents. Whatever her type,
each agent has a linear logarithmic utility function defined over consumption (when
old) and bequest.
From utility maximization follows that the (optimal) bequest each agent leaves
to his child at time t – which will be the initial wealth of the newborn in t+1 (say
Zt+1 ) – is a fraction (1 − ζ) of her wealth, which in turn is the sum of labour income
in the first period of life and initial wealth.
The definition of agents’s wealth however is not uniform across agents. There-
fore the bequest received from her parent by each child can be described by the
following laws of motion:
therefore:
Z̄
z = Zs∗ − (Z ∗ − Zu∗ )
Zmax s
This is a simple theory of the persistence of heterogeneity. If the dynamics of
the distribution were such as to converge to a stationary distribution in which only
one type of agents survive, we would have had the restoration of the representative
agent. This is what would happen in case of perfect capital markets.
Aghion and Bolton (1997) (AB hereafter) assume that agents live for one period
during which they work and invest. Income earned by working and investing is
divided between consumption and bequests. Each agent has one parent, whom she
receives a bequest from, and one child, whom she leaves a bequest to. Each agent’s
wealth is equal to the bequest received from the parent. The distribution of wealth
endowments will be represented by G(Z).
Each agent is endowed with one unit of labour which she supplies inelastically
(there is no disutility of labour). She can use her unit of labour to work on a
‘backyard activity’ which yields r̄ with certainty or invest in an ‘entrepreneurial
activity’ which yields r with probability pr or 0 with probability (1 − pr ).
In order to invest, the agent must commit one unit of wealth. If the agent doesn’t
invest, she can employ her wealth in an economy-wide ‘mutual fund’.
Capital markets are imperfect. The borrower–lender relationship is character-
ized by a moral hazard problem with limited wealth constraints. In this setting
AB show that, due to the features of the optimal lending contract, the probability
of success pr and the interest rate charged on loans ρ are functions of individual
wealth, with ∂p ∂ρ
∂Z > 0, ∂Z < 0.
r
Zt+1 = 0 (10.38)
Heterogeneity, aggregation and capital market imperfection 249
According to AB the economy will grow until all investment opportunities are
exploited. In this case the equilibrium interest rate on the mutual fund is rm∗ = 1.
Applying the results on convergence for monotonic Markov processes AB show
that the distribution of wealth converges to a stationary (long run) distribution.
As in GZ, AB have put forward a simple theory of the persistence of hetero-
geneity in wealth endowments in an imperfect capital markets framework.
where Yti and Kti are output and capital of the i-th firm in period t, ν is the output-
capital ratio, uniform across firms.
Firms sell their output at an uncertain price because of their limited knowledge
of market conditions. In order to capture uncertainty, we model the individual
selling price Pti as: Pti = uti Pt where uti is a random idiosyncratic shock iand
Pt
is the average market
price, uniform across firms. We assume that E u t = 1.
Therefore E Pti = Pt .We can also interpret the random shock as the relative
price: uti = Pti /Pt .
Moreover we assume that firms cannot raise funds on the Stock market because
of equity rationing but they have unlimited access to credit. This means that firms
do not issue new equities but can obtain from banks all the credit they need to
finance production at the (exogenous) rate of interest, r, uniform across firms and
time invariant.
Firms differ according to their financial conditions. The financial robustness
of a firm is proxied by the equity ratio, i.e. the ratio of its equity base to capital
ait = Ait /Kti .
Each firm incurs financing costs CFti equal to debt commitments:
where q is the real price of capital and Ait−1 is the net worth or equity base in real
terms inherited from the past.
Assuming that there is no depreciation, capital accumulates according to the
investment equation Iti = Kti − Kt−1 i
.
We assume that the firm incurs capital adjustment costs, CAit which are increas-
ing in the ratio Iti /Kt of its investment Iti to the average capital stock Kt . We make
250 Domenico Delli Gatti and Mauro Gallegati
the additional technical assumption that the adjustment costs function is quadratic:
2 2
γ Iti γ Kti − Kt−1
i
CAit = = (10.41)
2 Kt 2 Kt
Quadratic adjustment costs are well known in the literature on investment. The
novelty of the expression above consists in assuming that adjustment costs are
decreasing with the average capital stock: the higher is the average capital stock
– i.e. investment activity on the part of other firms – the lower is the level of
adjustment costs for the individual firm. This formulation captures an externality
in investment activity.
Real profit is the difference between real revenue and real cost, which in turn
is the sum of financing and adjustment costs:
πti = uti Yti − CFti − CAit =
2
γ Kti − Kt−1
i
= uti Yti − r(qKti − Ait−1 ) −
2 Kt
In this framework, bankruptcy occurs if net worth becomes negative. Net worth
‘today’ is equal to net worth ‘yesterday’ plus retained profit, which in turn is equal
to profit less the flow of dividends (Dti ). The bankruptcy condition therefore is:
Ait = Ait−1 + πti − Dti = Ait−1 + ui Yti − CFti − CAit − Dti < 0 (10.42)
The inequality (10.42) is verified if the sum of financing costs, adjustment costs and
dividends is higher than revenues – thereby generating a loss – and the associated
loss is higher than the equity base inherited from the past. In order to simplify the
argument, in the following we will assume that the flow of dividends is proportional
to net worth inherited from the past:
Dti = rAit−1 (10.43)
In a sense this is tantamount to assuming the net worth is remunerated at the same
rate as bank loans.
Substituting(10.40)(10.41)(10.43) into (10.42) we get:
2
γ Kti − Kt−1
i
At−1 + ut Yt − r(qKt − At−1 ) −
i i i i i
− rAit−1 < 0
2 Kt
Using (10.39) and rearranging, we can write the bankruptcy condition as follows:
2
q Ait−1 γ Kti − Kt−1 i
ut < r −
i
+ (10.44)
ν νKti 2 νKti Kt
In the following, for the sake of analytical convenience, we will adopt a sim-
plified version of the bankruptcy condition, namely:
q Ai
ui < r − t−1i ≡ ūti (10.45)
ν νKt
Heterogeneity, aggregation and capital market imperfection 251
The condition (10.45) essentially ignores capital adjustment costs. This assumption
makes computations much less cumbersome without loss of generality. According
to the bankruptcy condition, bankruptcy occurs if the realization of the random
relative price uti falls below a critical threshold ūti which in turn is a function, among
other variables, of the capital stock and of the equity base lagged one period.
Let’s assume, for the sake of simplicity, that uti is a uniform random variable,
with support (0,2). In this case, the probability of bankruptcy is:
i ūti 1 q Ait−1
Pr ut < ūt = =
i
r −
2 2 ν νKti
where α1 α2 i
r i = rq 1 + − at−1
2 2
α1 α2 i
i
T = At−1 r + − at−1
2 2
r i is the bankruptcy cost augmented interest rate. It is determined as a mark-up
over the interest rate charged by banks on loans. This mark-up is a decreasing
function of the equity ratio: the higher the equity ratio, i.e. the financial robustness
of the firm, the lower the bankruptcy cost augmented interest rate. In a sense, this
mark-up captures the idea of the risk of the borrower. Solving the problem yields:
Iti K i − Kt−1
i
ν − ri
= t = = T0 + T1 ait−1
Kt Kt γ
q α1
where
ν
T0 ≡ −r 1+
γ γ 2
252 Domenico Delli Gatti and Mauro Gallegati
rqα2
T1 ≡
γ2
Since the issue of new equities is ruled out by assumption, in this framework each
firm can increase its net worth inasmuch as it accumulates internal funds. The
change of the equity base, therefore, coincides with retained profits.
The law of motion of the equity base of the i-th firm is:
2
γ Kti − Kt−1
i
Ait = Ait−1 + uti Yti − rqKti − (10.47)
2 Kt
Dividing by the individual capital stock, we can derive the law of motion of the
equity ratio:
2
i
Kt−1 γ Kti − Kt−1i
at = at−1 i + ut ν − rq −
i i i
Kt 2 Kt Kti
i
The expression Kt−1 /Kti can be written as:
i
Kt−1 I i Kt
i = 1− t i
Kt Kt K t
In order to make the analysis manageable, we assume that the ratio Kti ∼
= 1. Thanks
Kt
to this simplifying shortcut we get:
i
Kt−1 Iti
= 1 − = 1 − T0 + T1 ait−1
Kti Kt
2 i 2
Kti − Kt−1
i
It
=
Kt Kti Kt Kti
where
Γ1 = 1 − T0 (1 + γT1 )
Heterogeneity, aggregation and capital market imperfection 253
γ
Γ 2 = T1 1 + T 1
2
γ
Γi0 = uti ν − rq − (T0 )2
2
From (10.48) through aggregation, assuming that the law of large numbers holds
true, we obtain the law of motion of the average equity ratio:
where Vt−1 is the variance of the distribution of the equity ratio in t-1. From the
definition of the variance at time t, we derive the following:
where βt−1 is a parameter capturing the kurtosis of the distribution, while µ3t−1 is
the third moment from the mean.
(10.49) and (10.50) is a system of two non-linear difference equations in the
state variables at and Vt which describes the evolution over time of the first two mo-
ments of the distribution of the equity ratio. This dynamical system is described by
a two-dimensional non-linear map which yields multiple equilibria (steady states),
with different dynamical properties depending upon the chosen configuration of
parameters.23
The map generates a wide range of dynamic patterns: convergence to a sta-
tionary distribution, periodic or aperiodic cycles, chaotic dynamics, divergence. To
each dynamic pattern of the distribution corresponds a dynamic pattern of aggre-
gate production and investment (capital accumulation). Therefore, the framework
can generate convergence to a steady growth path, endogenous regular or irreg-
ular business cycles and growth – which we label fluctuating growth for short –
divergent trajectories, which we label ‘ financial crises’ .24
A fluctuation can be generated also by an exogenous stochastic shock forced
upon the system when it is in a steady state position. In particular, elsewhere we
have shown that the higher the degree of heterogeneity, the larger the effects of a
shock and the longer its persistence.
Heterogeneity is very important in business cycle analysis for empirical as well
as theoretical reasons (Kirman, 1999). When firms are heterogeneous, knowledge
about the distribution of firms is crucial in order to understand the response of the
system to aggregate and idiosyncratic shocks.
10.7 Conclusion
Heterogeneity is a catch word which can mean different things to different people
in different contexts. We have to refine the notion in order to draw meaningful
conclusions on its importance in macroeconomics. In this section we will try such
a refinement with the help of the following table, which summarizes the main
features of the different theoretical frameworks examined so far. The columns are
254 Domenico Delli Gatti and Mauro Gallegati
(Consumption) (Consumption)
(Consumption) (Consumption) (Consumption)
goods goods
goods goods goods
Markets Labour
Capital Non-reproducible
Labour Labour
Labour assets (land)
Credit Credit Credit
Credit Credit
Asymmetric Asymmetric
Asymmetric information on information on Asymmetric
Asymmetric information on effort effort information on
Capital market
information on investment return ¯ ¯ effort
imperfections equity market ¯ Moral hazard Moral hazard ¯
Moral Hazard (inalienable human (take money and Moral hazard
capital) run)
¯ ¯ ¯
Implication of Equity rationing Monitoring costs Financing
Financing (costly state constraint: External financial Possibility of credit
capital market hierarchy verification) credit= collateral premium rationing
imperfections Bankruptcy risk if auditing occurs wealth
Dynamics
Convergence Convergence
Second best Wide range of Wide range of to stationary to stationary
Complex dynamics possible dynamics possible dynamics wealth distribution wealth distribution
Endogenous Investors Skilled workers Investors
Distribution of Non-investors Unskilled Non-investors
net worth workers
Distri-
Heterogeneity bution
of firms,
households,
banks Entrepreneurs Farmers
Exogenous Non-entrepreneurs Gatherers
entitled to the models presented, while the rows are devoted to a classification of
agents (grouped in the two broad categories of borrowers and lenders), markets,
type and implications of capital market imperfections, type of dynamics, nature of
heterogeneity.
First of all, we can draw a distinction between weak and strong heterogeneity
(Rios-Rull, 1995; Gaffeo, 1999). Heterogeneity is strong if agents, differentiated
according to some economically relevant characteristics, interact strategically. In
this case the benchmark theoretical framework is game theory. Heterogeneity is
weak if agents, albeit different, are not aware of – or simply do not take into
account – the fact that the decision taken by each one of them affects the payoff of
the others. The models surveyed so far explore the macroeconomic implications
of weak forms of heterogeneity. In other words none of the models reviewed in
this chapter is embedded in a game theoretical framework.
Heterogeneity, aggregation and capital market imperfection 255
A second distinction can be drawn between exogenous and endogenous het-
erogeneity. Heterogeneity is exogenous if the distribution of agents according to
some economically relevant characteristics is postulated by the modeler, it is en-
dogenous if the distribution if determined in the model. An example can help in
clarifying this distinction.
A fundamental differentiation of agents in models with imperfect capital mar-
kets is the distinction between borrowers and lenders. GS simply assume that
distinction from the beginning of the analysis. They start, in fact, from the fact that
some agents specialize in lending funds to other agents who specialize in borrow-
ing; they don’t bother to assess which characteristics of the agents brought about
that specialization. In a sense, the same can be said of KM. In their framework,
the distribution between financially constrained and unconstrained agents is ex-
ogenous. It turns out that the former are borrowers and the latter are lenders. In
this sense also the distribution between borrowers and lenders is exogenous.
On the contrary BG exogenously assume the distribution of agents between
entrepreneurs and non-entrepreneurs but are able to derive the distribution of
entrepreneurs in investors and non-investors endogenously. It turns out that en-
trepreneurs characterized by a degree of inefficiency lower than a certain – en-
dogenously determined – threshold level of inefficiency carry out their investment
projects and therefore are seeking funds to finance investment activity, i.e. they
specialize in investing and borrowing, while entrepreneurs with a degree of ineffi-
ciency higher than the above mentioned threshold give up investment and ‘store’.
In other words, inefficient entrepreneurs and non-entrepreneurs specialize in lend-
ing. In a sense, the distinction between borrowers and lenders is partly endogenous
due to the fact that the distinction between efficient and inefficient entrepreneurs
is endogenous. Similarly, AB endogenously determine the specialization of agents
as borrowers or lenders. GZ endogenously determine the (long run or stationary)
distribution between skilled and unskilled agents.
There is a strong correlation between the degree of capital market imperfec-
tions, the relevance of heterogeneity for macroeconomic performance, the relative
complexity of the dynamics of the state variables involved in the analysis and the
response of individuals and the aggregate to impulses.
Generally speaking, in the models surveyed so far, we can distinguish between
the general case of imperfect capital markets – imperfections being generally
informational in nature – and the special, simpler case of perfect capital markets.
For instance, if we relax the assumption of equity rationing and bankruptcy costs
in the GS model we go back to a perfect capital markets world in which the
financial conditions of the individual firm do not matter as far as employment
and production decisions are concerned. If we relax the assumption of monitoring
cost in the BG model we go back to a perfect capital markets world in which the
financial conditions of the individual entrepreneurs do not matter as far as capital
accumulation is concerned.
In all these cases, the representative agent assumption is a legitimate working
hypothesis. If employment and production decisions in the GS world or capital
accumulation decisions in the BG world do not depend on individual financial
256 Domenico Delli Gatti and Mauro Gallegati
conditions, but only on characteristics of the economy (for instance: technology)
which are uniform across firms, heterogeneity is irrelevant and the Modigliani–
Miller rule is reinstated. This is the first best case. It turns out that in this case the
dynamics of the state variables under examination are relatively simple to analyze
and determine a stable steady state (this is the GS’ first best case) or the dynamics
are trivial because . . . there are no dynamics, i.e. no feedback of the value of the
state variable in one period to the values the variable will assume in the future (this
is BGs first best case, for instance).
A slightly different story holds true for the KM framework, in which farmers
and gatherers are differentiated not only because of the presence or absence of a
financial constraint, but also because different agents can access different tech-
nologies. In particular farmers produce by means of a constant marginal returns
technology while gatherers are characterized by a well behaved production func-
tion with decreasing marginal returns. This second distinction is inessential as to
the deep characterization of the two groups of agents. In other words, what is re-
ally important in the model is the financial distinction and not the technological
one. There is no reason, moreover, to link the nature of financially constrained
(unconstrained) agent to the type of technology available.
The technological distinction, however, enables us to define in a peculiar and
interesting way the perfect capital markets case. The special case of perfect capital
markets in KM framework is obtained removing the crucial assumption of finan-
cial constraint but not the technological one. The first best is achieved when the
marginal product of the durable non-reproducible asset (land) for the farmer, which
is a given constant thanks to the constant marginal returns assumption, is equal to
the marginal product of land for the gatherer, which is decreasing with the amount
of land used in production by the gatherer thanks to the decreasing marginal returns
assumption. This is tantamount to assuming a no-arbitrage condition.
Thanks to the technological differentiation, from the no-arbitrage condition
KM can derive the amount of land allocated to each group of agents and the
equilibrium values of the other variables without recurring to any laws of mo-
tion. In other words, in KMs framework, the first best case is characterized by
(technological) heterogeneity but there are no dynamics. As we said before, the
technological heterogeneity which persists in KM even when the financial con-
straints are removed is inessential as far as the macroeconomic implications of
capital market imperfections are considered but they are necessary to derive uni-
vocally the allocation of the durable assets. If technology were uniform across
agents the no-arbitrage condition would have left the allocation of land to farmers
and gatherers undetermined.
Summing up, the perfect capital markets – Modigliani–Miller – first best is
a special case characterized by the absence or irrelevance of heterogeneity (in
our case, heterogeneity of financial conditions) which makes the representative
agent assumption legitimate and the dynamics simple or even trivial. In this case
shocks typically have symmetric effects. This is the consequence of homogeneity
of employment, production and/or capital accumulation responses to exogenous
shocks.
Heterogeneity, aggregation and capital market imperfection 257
Symmetrically, in the imperfect capital markets case heterogeneity (in our
case, heterogeneity of financial conditions) is essential, it invalidates the adoption
of the representative agent assumption and generally determines a more interesting,
sometimes complex dynamics. In this case shocks typically have asymmetric ef-
fects. This is the consequence of heterogeneity of employment, production and/or
capital accumulation responses to exogenous shocks.
We now turn to aggregation. We can distinguish two procedures in the models
discussed so far. In GS there is an initial continuous distribution of agents according
to their level of equity base or net worth. This distribution can be characterized
by its moments and in particular by the mean – the average equity base or the
equity base of the average agent – and the variance. GS derive the law of motion
of the average equity base, neglecting the role of the variance and its evolution
over time. In the extension considered in Section 10.7, on the other hand, we show
how to describe simultaneously the laws of motion of the average equity ratio and
its variance. The initial distribution of the equity ratio may converge or not to a
stationary distribution. As a matter of fact a wide range of dynamical behaviors
are possible, within this framework depending upon the values assumed by the
parameters.
The rest of the papers considered follows a different procedure. BG start from
a continuous initial distribution of degrees of inefficiency and boil it down to a
polarized distribution of efficient and inefficient entrepreneurs by means of a simple
criterion of choice. GZ start from a continuous initial distribution of wealth and boil
it down to a polarized distribution of skilled/rich individuals and unskilled/poor
individuals by means of a piecewise linear phase diagram. Aggregate outcomes,
therefore, emerge as weighted averages of polarized outcomes. In other words it
is as if the modeler could compute the expected value of a discrete distribution
consisting of only two characters. AB follow a similar line of inquiry in a more
sophisticated setting. In fact, they treat the evolution over time of the distribution
of wealth by means of the theory of convergence for monotonic Markov processes.
Notes
1 Stoker (1993: 1829) notes that representative agent modelling ‘has proved a tremen-
dous engine for the development of rational choice models over the last two decades,
and their empirical application has developed into an ideology for judging aggregate
data models’.
2 Kirman, 1992, is the locus classicus of this literature.
3 See, among others, Leontief, 1947; Gorman, 1953; Theil, 1954; Eisenberg, 1961;
Fisher, 1969, 1982; Muellbauer, 1975, 1976; Lau, 1977, 1982. On aggregation see
Daal and Merkies, 1984.
4 Lippi, 1988; Lippi and Forni, 1996; Lewbel, 1992.
5 Stoker, 1986,1993; Caballero et al., 1997; Geweke, 1985, demonstrate the pitfalls of
using the representative agent framework to explain aggregate behaviour. Grunfeld and
Griliches, 1960, show that macromodels with large aggregation bias can produce better
prediction than their disaggregated counterparts. Empirical evidence on distribution
of income and wealth in the growth process can be found in Kuznets, 1955, and Levy
and Murnane, 1992.
258 Domenico Delli Gatti and Mauro Gallegati
6 Banerjee and Newman, 1988; Acemoglu, 1997; Benabou, 1993, 1996; Bertola, 1993;
Hopenhayn, 1992; Solon, 1992; Persson and Tabellini, 1994; Zarembka, 1975; Chiu,
1998.
7 Roller and Sinclair-Desgagne, 1996; Bertola and Caballero, 1994; Keane et al., 1988;
Heckman and Sedlacek, 1985.
8 Keane, 1997; Amable and Chatelain, 1997; Guariglia and Schiantarelli, 1998; Zietz,
1996; Fortin, 1995; Caballero, 1995; Gordon, 1992; Perotti, 1993;Attanasio and Weber,
1995.
9 Cho, 1995; Aoki, 1994; Chiaromonte and Dosi, 1993; Caballero and Engel, 1993; Das,
1993; Zeira, 1994; Caplin and Leahy, 1991, 1994; Kydland, 1984.
10 Aghion and Bolton, 1997; Banerjee and Newman, 1988, 1993; Galor and Zeira, 1993;
Piketty, 1997; Bernanke and Gertler, 1989, 1990; Greenwald and Stiglitz, 1988, 1990,
1993; Kiyotaki and Moore, 1997.
11 In other words, there are marginal decreasing returns. The analysis could be carried
out, however, also in the presence of constant marginal returns, i.e. in the case in which
F is linear.
12 When capital markets are affected by informational imperfections such as asymmetric
information, a financing hierarchy (pecking order) can be envisaged. Internal finance is
the most preferred source of finance. As to external sources, credit has a cost advantage
over the issue of new equities (Fazzari, Hubbard and Petersen, 1988). A different
wording is used in the literature on the so-called credit view: bank loans are imperfect
substitutes of the issue of new equities (credit is ‘special’). See, for instance, Bernanke
and Blinder (1988).
13 As a matter of fact, the procedure followed by GS to derive an equation for the prob-
ability of bankruptcy is much more complicated. In this survey we can simplify the
argument without loss of generality.
14 On bankruptcy costs, see Gordon and Malkiel, 1981; Altman,1984; Gilson, 1990;
Kaplan and Reishus, 1990.
15 There is only demand for consumption goods in this framework. Investment is ruled out
by assumption, since production takes place using only labour as an input. Financial
factors do not play any role in the determination of aggregate demand, due to the
peculiar way in which consumer’s preferences and the budget constraint are modelled.
It is clear, however, that if firms use capital together with labour to carry on production,
financial factors can influence aggregate demand through their impact on investment
activity. Even in the absence of investment activity, however, demand can be influenced
by financial factors. If firms pay-out dividends based on their net worth, in fact, and
dividends are part of consumers’income, consumption expenditure is affected by firms’
financial conditions.
A different and somehow more indirect way to explore the impact of financial factors
on aggregate demand is adopted in Delli Gatti and Gallegati (1997).
16 This point is discussed at length in Delli Gatti and Gallegati (1997).
17 BG do not impose equilibrium on the market for loanable funds. They simply assume
that the volume of loans available is sufficient to fill the financing gap of borrowers. The
rate of return on storage therefore is not determined in equilibrium but is exogenous.
18 For the sake of simplicity, we are ignoring the effects on the price and the quantity of
capital of changes of other exogenous variables, such as the interest rate and the share
of entrepreneurs in total population.
19 For the sake of simplicity, we are ignoring the effects on the price and the quantity of
capital of changes of other exogenous variables, such as the interest rate, the share of
entrepreneurs in total population, the cost of monitoring etc.
20 On this issue see Hart and Moore (1994, 1998).
21 As a matter of fact, the paper by Aghion and Bolton has been circulating for some
years as a LSE working paper.
Heterogeneity, aggregation and capital market imperfection 259
22 As a matter of fact, AB show that under appropriate conditions there can be credit
rationing. We will not dig deeper into this issue.
23 If agents were identical – i.e. if the Representative Agent Hypothesis held true – the
law of motion of the equity ratio of the representative firm would be:
at = Γ1 at−1 − Γ2 a2t−1 + Γ0
11.1 Introduction: the big puzzle: why do all rival systems trail so
far behind free market growth rates?
Undoubtedly, the spectacular and unmatched growth rate of the industrialized
free-market economies is what most distinguishes them from all other economic
systems. In no other system, current or in the past, has the average income of the
general public risen anywhere nearly as much or as quickly as it has in North
America, Western Europe and Japan. Though the former Soviet Union planned its
economy and forced its population to invest heavily in factories and hydroelectric
dams, its failure to produce enough to raise the standard of living of its population to
that of the free-market economies undoubtedly played a major role in its downfall.
There have been great civilizations with extraordinary records of invention and
engineering – medieval China and ancient Rome are clear examples. But none
has approached the growth record of modern free-market economies. What is the
secret of their extraordinary success? That is the economic puzzle that is absolutely
critical to the degree of prosperity our future is able to achieve. Its solution is what
the world’s poorer countries are anxious to learn.
Toward the microeconomics of innovation 267
11.2 The free market’s growth record
The growth of per-capita income and productivity in the free-market economies
is so enormous that it is virtually impossible to comprehend. In contrast, average
growth rates of per-capita incomes were probably approximately zero for about
1,500 years before the Industrial Revolution at the time of George Washington. In
1776, even the wealthiest consumers in England, then the world’s richest country,
had only a half-dozen consumer goods that had not been present in ancient Rome.
These new products included (highly inaccurate) hunting guns, (highly inaccurate)
watches, paper, window glass and very little else. And, remarkably enough, Roman
citizens enjoyed a number of amenities, such as hot baths and good roads, that had
long disappeared at the time of the American Revolution.
In contrast, in the past century and a half, per-capita incomes in the typical
capitalist economy have risen by amounts ranging from several hundred to several
thousand percent. Recent decades have yielded an unmatched outpouring of new
products and services: color television, the computer, jet aircraft, the VCR, the
microwave oven, the hand-held calculator, the cellular telephone, and so on and
on. And the flood of new products continues. When, a few years ago, many of the
world’s communist regimes collapsed and when even the masters of China turned
toward capitalist enterprise, surely part of the reason was the public’s desire to
participate in the growth miracle of the capitalist economies that Marx and Engels
– those high priests of anticapitalist movements – were among the first economists
to discern (as the opening quotation demonstrates).
The explanation of this miracle must be sought in the activities of industries
and the business firms of which they are constituted, for they are the producers of
the increasing outpouring of goods and services that constitutes the growth record
of capitalism. It must be something about business firms and the decisions they
make that plays a vital part in this prime attribute of our economy. Yet the standard
core microeconomic theory of firms and industries, while it has included some
outstanding contributions on the theory of innovation, has not provided anything
suggesting what features of business behavior and decision-making can account for
all this growth. Indeed, as we know, mainstream microeconomics offers reasons to
expect that the capitalist economy will be characterized by a growth performance
that is far from optimal.
Here, I will describe some features of competitive markets to which the growth
performance of business firms can be attributed, features that literally force busi-
nesses to do all they can to contribute to the growth miracle. I will then provide
some hints of a microeconomic model, using the most elementary of microeco-
nomic tools to analyze this process.
The poor in the towns and countryside lived in a state of almost complete
deprivation. Their furniture consisted of next to nothing. . . . Inventories
made after death . . . testify almost invariably to the general destitution . . . a
few old clothes, a stool, a table, a bench, the planks of a bed, sacks filled with
straw. Official reports for Burgundy between the sixteenth and the eighteenth
centuries are full of references to people (sleeping) on straw . . . with no
bed or furniture, who were only separated from the pigs by a screen. . . .
Paradoxically the countryside sometimes experienced far greater suffering
(from famines than the townspeople). The peasants . . . had scarcely any
reserves of their own. They had no solution in case of famine except to turn to
the town where they crowded together, begging in the streets and often dying
in public squares. . . .
(Fernand Braudel, The Structures of Everyday Life, Vol. I, New York: Harper
and Row, 1979, pp. 73–75 and 284–286).
Only the growing outputs that innovation, first in agriculture and mining and then in
manufacturing and transportation, made feasible produced the enormous increases
in productive plant and equipment and in education (and other forms of investment
in human capital) that are widely judged to have contributed greatly to economic
growth. Thus, it can be argued not only that innovation has facilitated the growth
process, but that without it the process would have been reduced to insignificance.
Two of the leading analysts of economic growth conclude:
As yet, no empirical study proves that technology has been the engine of
modern-day growth. Still, we ask the reader to ponder the following: What
would the century’s growth performance have been like without the invention
and refinement of methods for generating electricity and using radio waves to
transmit sound, without Bessemer’s discovery of a new technique for refining
iron, and without the design and development of products like the automobile,
the airplane, the transistor, the integrated circuit, and the computer?
(Gene M. Grossman and Elthanan Helpman, ‘Endogenous Innovation in the
Theory of Growth,’ Journal of Economic Perspectives, Vol. 8, Winter, 1994,
p. 32).
Toward the microeconomics of innovation 269
The computer industry hasn’t made a dime . . . Intel and Microsoft make
money, but look at all the people who were losing money all the world over.
It is doubtful the industry has yet broken even,’ said Peter Drucker in a recent
interview . . . but is it true? Paul Gompers of the Harvard Business School and
Alon Brav of the University of Chicago . . . looked at companies that went
public from 1975 to 1992, most of which were high-tech firms, and found
their rate of return to be about average [i.e., zero economic profit], once they
adjusted for risk and company size (‘The Rewards of Investing in High Tech,’
Federal Reserve Bank of Boston, Regional Review, Vol.6, Fall 1996, p. 14).
Acknowledgement
I am grateful to the Russell Sage Foundation and the C. V. Starr Center at New
York University for their support of this work.
Notes
1 According to the late British economist, T.S. Ashton, in his classic book, The Industrial
Revolution, 1760–1830, London: Oxford University Press, 1948, that phrase is how
one schoolboy (quite appropriately) described the Industrial Revolution.
2 See R.E. Lucas, Jr (1988) ‘On the Mechanics of Economic Development’ Journal of
Monetary Economics, 22, 3–42
3 For a discussion of possible reasons for failure of the economies of Ancient Rome
and Medieval China to achieve outstanding growth records, see my Entrepreneur-
ship, Management and the Structure of Payoffs, Cambridge, Mass.: MIT Press, 1993,
Chapter 2.
4 For further materials on technology sharing in practice, with a number of concrete
examples, see Chapter 10 of my recent book (op.cit., 1993).
5 My own value judgment is summed up in George Bernard Shaw’s dictum that there
is no crime greater than poverty. Consequently, I am inclined to prefer a fairly high
spillover ratio, perhaps not far from its current value.
6 ‘This pattern of industrialization without wage gains is what it would take to ensure that
the industrialist captures all of the benefits he creates when he introduces machinery.
. . . [this] cannot be a historically accurate description of the process of development in
industrial countries, for if it were, unskilled labor would still earn what it earned prior
to the industrial revolution’ (Paul Romer, ‘New Goods, Old Theory and the Welfare
Costs of Trade Restrictions,’ Journal of Development Economics, Vol. 43, 1994, p. 29).
Part IV
Principle 2: The model economy being used to measure something should not
have a feature which is not supported by other evidence even if its introduction
results in the model economy better mimicking reality.
When modifying the standard growth model to address a business cycle question,
Hansen (1985) introduced another feature, a labor indivisibility, and permitted
people to enter into mutually beneficial insurance contracts as in Arrow–Debreu
theory. This increases the magnitude of the response to TFP shocks. There is empir-
ical evidence that justifies incorporating this feature. First, there is unemployment
insurance and people have a close substitute for insurance, namely the holding of
liquid assets. Second, and most important, most of the cyclical variation in the
labor input is the result of variation in the number of people that work in a given
week and not in the average workweek length. Like Kydland and my non-time-
separable utility function, the introduction of this feature results in leisure having
a high intertemporal elasticity of substitution.
Hansen finds that for his model economy, cyclical fluctuations induced by
technology shocks are as large as the observed fluctuations. Given the strong
empirical support for labor indivisibility, this mechanism should be part of the
model or measurement instrument used to answer the question of the importance
of technology shocks for business cycle fluctuations. A problem, however, is that
not all fluctuations in hours are the result of the variations in the number of people
working. An important fraction is the result of variation in the workweek length
as in Kydland and my model economy. This number is somewhere between the
Hansen and the original Kydland and Prescott estimates.
What was needed was better theory. The reason is the following. Why is the
workweek fixed? If it is permitted to vary and aggregate hours is the labor input to
production, then all work and all variation is in the workweek length. There was a
major inconsistency between observation and theory. Better theory was developed
that reduced this inconsistency. I begin with the aggregate production function and
its inadequacy for the purpose of understanding the determination of workweek
length.
The aggregate production function is the solution to the following program, where
M+ (T ) is the set of measures on the Borel sigma algebra of T ,
F(X ) = max f (x)z(dx)
z∈M+ (T )
subject to
xi z(dx) ≤ Xi i = 1, 2, . . . , n.
T
Given the assumptions, the constraint set is compact and non-empty and the ob-
jective function is continuous in the weak star topology. Therefore the program
has a solution. Two well-known results are the following.
12.1.3 An example
The Cobb–Douglas production function has come to dominate in aggregate applied
general equilibrium analysis. The reason is that both over time and across countries,
labor’s share of product is surprisingly constant at about 70 per cent.1 The Cobb–
Douglas production function, with its unit elasticity of substitution, is about the
Business cycle research 289
only aggregate production function with the property that factor income shares are
the same even though relative factor prices are very different.
An example of an underlying set of plant technologies for the Cobb–Douglas
production function is the following one. Suppose that the factor inputs are capital
k and labor n and that the plant technologies are g(n)k θ . In addition, the function
g is such that the function g(n)nθ−1 has a maximum. This maximum is denoted
by A and a maximizing n by n∗ .
Outline of proof: Given there are two constraints and an optimum exists, there
is an optimum with at most two types of plants operated. Consider one such
optimum. Let (Ki , Ni ) be the aggregate factor inputs used to operate type i plants
for i = 1, 2. Allocate (Ki , Ni ) equally to m plants of type (Ki /m, Ni /m). The m for
which Ni = mn∗ is an optimum. Thus, all operated plants for this optimum have
n = n∗ . Output maximization requires marginal products of capital be equated
across operated plants. Thus, operating N /n∗ plants each with n∗ workers and
K/(N /n∗ ) units of capital is optimal. Using this result,
N ∗ K
F(K, N ) = ∗ g(n ) = AN (1−θ) K θ
n Nn∗
y ≤ Ahk θ ,
where 0 < θ < 1 and where y is the output produced by an individual, k is the
capital that that individual uses and h is the length of that individual’s workweek.
Element h belongs to the set H ⊂ (0, 1].
Principle 3: A model that better fits the data may be a worse measurement
instrument. Indeed, a model matching the data on certain dimensions can be the
basis for rejecting that model economy as being a useful instrument for estimating
the quantity of interest.
Kt+1 = (1 − δ(x))Kt + It .
The function δ(x) is increasing in x.
Introducing this feature results in aggregate observations being in greater con-
formity with theory. However, there is a problem. The problem is the lack of
micro-observations to back up the depreciation assumption. Does capital depreci-
ate more in boom periods? Until other evidence is provided for this depreciation
assumption, this feature is best not incorporated into the measurement instrument.
A suitable commodity vector is {ct (s), it (s), kt (s), n1t (s), n2t (s)}s∈S,t∈{0,1,2, . . . } .
Here s is the state, c and k are scalars, and the ni are signed measures on the
Borel sigma algebra of [0,1]. Restrictions on a type i consumption set are that ni
be a probability measure and nj=i be the null measure.
The plant production function is
hf (k, x1 , x2 )
where h is hours the plant is operated, k is the capital employed and the xi are
the number of type i workers. Suppressing the t and s index, a period aggregate
production possibility set is
294 Edward C. Prescott
12.2.9 The role of varying number of shifts that plants are operated
A margin of adjustment that has not been introduced into any applied general
equilibrium analysis of business cycles is the option to vary the number of shifts
a plant is operated. This is an important margin of adjustment in the automobile
industry.
Acknowledgement
Prepared for the International School of Economic Research’s XI Workshop, ‘Cy-
cles, Growth and Technological Change.’ 29 June 1998, to 7 July 1998. Certosa di
Pontignano-Siena (Italy). I thank Dirk Krueger and Jessica Tjornhom for useful
comments and assistance in preparing the list of references. The views expressed
herein are those of the author and not necessarily those of the Federal Reserve
Bank of Minneapolis or the Federal Reserve System.
Notes
1 See Gollin (1997) for the cross-country numbers. He uses the Kravis’ (1959) economy-
wide assumption for assigning proprietor’s income and indirect business taxes to capital
and labor.
2 See Chari, Kehoe and McGrattan (2000) for a critical evaluations of mechanisms to
propagate the effects of monetary shocks.
298 Edward C. Prescott
References
Alvarez, F. and U. Jermann (2000) ‘Efficiency, equilbrium and asset pricing with risk of
default’, Econometrica, 68, (4), 775–98.
Atkeson, A. and R. Lucas (1992) ‘On efficient distribution with private information, Review
of Economic Studies, 59, (3), 427-53.
Atkeson, A. and R. Lucas (1995) ‘Efficiency and equality in a simple model of efficient
unemployment insurance’, Journal of Economic Theory, 66, (1), 64-88.
Backus, D., P. Kerhoe and F. Kydland (1992) ‘International Real Business Cycles’, Journal
of Political Economy 101, 745-75
Baxter, M. and M. Crucini (1995) ‘Business cycles and the asset structure of foreign trade’,
International Economic Review, 36, (4), 821-54.
Boyd, J. and E. Prescott (1986) ‘Financial intermediary coalitions’, Journal of Economic
Theory, 38, (2), 211-32.
Bresnahan, T. and V. Ramey (1993) ‘Segment shifts and capacity utilization in the US
automobile industry’, American Economic Review, 83, (2), 213-18.
Brock, W. and L. Mirman (1972) ‘Optimal economic growth and uncertainty, the discounted
case’, Journal of Economic Theory, 4, (3), 479-513.
Burns, A.F. and W.C. Mitchell (1946) Measuring Business Cycles, New York: National
Bureau of Economic Research.
Burnside, C., M. Eichenbaum and S. Rebelo (1993) ‘Labour Hoarding and the Business
Cycle’, Journal of Political Economy 101, 245–73.
Carlstrom, C. and T. Fuerst (1997) ‘Agency costs, net worth, and business fluctuations: a
computable general equilibrium analysis’, American Economic Review, 87, (5), 893-910.
Cass, D. (1965) ‘Optimum growth in an aggregative model of capital accumulation’, Review
of Economic Studies, 32, (3), 233-40.
Chari, V.V., P. Kehoe and E. McGrattan (1998) Monetary shocks and real exchange rates
in sticky price models of international business cycles, Federal Reserve Bank of Min-
neapolis Staff Report 223.
Chari, V.V., P. Kehoe and E. McGrattan (2000) ‘Can the contract multiplier solve the per-
sistence problem?’, Econometrica, 68, (5), 1151–79.
Christiano, L. and M. Eichenbaum (1995) ‘Liquidity effects, monetary policy, and the
business cycle’, Journal of Money, Credit and Banking, 27, (4), 1113-36.
Christiano, L. and M. Eichenbaum (1997) ‘Sticky price and limited participation models of
money, a comparison’, European Economic Review, 41, (6), 1201-49.
Cole, H. and E. Prescott (1997) ‘Valuation equilibrium with clubs’, Journal of Economic
Theory, 74, (1), 19-39.
Cooley, T., G. Hansen, and E. Prescott (1995) ‘Equilibrium business cycle with idle resources
and variable capacity utilization’, Economic Theory, 6, (1), 35-49.
Cooley, T. and V. Quadrini (1998) ‘Monetary policy and the financial decisions of firms’,
mimeo.
Devereux, M., A. Head, and B. Lapham, (1996) ‘Aggregate fluctuations with increasing
returns to specialization and scale’, Journal of Economic Dynamics and Control, 20,
(4), 627-56.
Diamond, P. (1965) ‘National debt in a neoclassical growth model’, American Economic
Review, 55, (5), 1126-50.
Diáz-Giménez, J., E.C. Prescott, T. Fitzgerald and F. Alvarez (1992) ‘Banking in Com-
putable General Equilibrium Economies’, Journal of Economics Dynamics and Control
16, 533–59.
Business cycle research 299
Fagnart, J., O. Licandro, and F. Portier (1999) ‘Firm heterogeneity, capacity utilization and
the business cycle’, REview of Economic Dynamics, 2, (2), 433–55.
Fitzgerald, T. (1996) ‘Reducing working hours’, Economic Review, Federal Reserve Bank
of Cleveland, 32, (4), 13-22.
Fitzgerald, T. (1998) ‘Work schedules, wages, and employment in a general equilibrium
model with team production’, forthcoming in Review of Economic Development.
Freeman, S. and F. Kydland (2000) ‘Monetary aggregates and output’, American Economic
Review, 90, (5), 1125–35.
Friedman, M. and A.J. Schwartz (1963) A Monetary History of the United States, 1867–
1960, Princeton: Princeton University Press.
Fuerst, T.S. (1992) ‘Liquidity, Loanable Funds and Real Activity’, Journal of Monetary
Economics, 29, 3–24.
Gollin, D. (forthcoming) ‘Getting income shares right’, Journal of Political Economy.
Greenwood, J., Z. Hercowitz and G.W. Huffman (1988) ‘Investment, capacity utilization
and the real business cycle’, American Economic Review 78, 402–17.
Hansen, G.D. (1985) ‘Indivisible labor and the business cycle’, Journal of Monetary Eco-
nomics, 16, 309-27.
Hart, O. (1979) ‘Monopolistic competition in a large economy with differentiated com-
modities’, Review of Economic Studies, 46, (1), 1-30.
Hodrick, R. and E. Prescott (1980 and 1997) ‘Postwar US business cycles: an empirical
investigation’, Journal of Money, Credit, and Banking, 29, (1), 1-16 (1997), Discussion
Paper 451, Northwestern University (1980).
Hornstein, A. and J. Praschnik (1997) ‘Intermediate inputs and sectoral comovement in the
business cycle’, Journal of Monetary Economics, 40, (3), 573-95.
Hornstein, A. and E. Prescott (1993) ‘The plant and the firm in general equilibrium theory’,
in R. Becke, M. Boldrin, R. Jones and W. Thompson (eds) General Equilibrium and
Growth: The Legacy of Lionel McKenzie, New York: Academic Press.
Ireland, P. (1995) ‘Endogenous financial innovation and the demand for money’, Journal
of Money, Credit and Banking, 27, (1), 107-23.
Johnson, H. (1960) ‘The cost of protection and the scientific tariff’, The Journal of Political
Economy, 68, (4), 327-45.
Jones, L. (1984) ‘A competitive model of commodity differentiation’, Econometrica, 52,
(2), 507-30.
Kehoe, T. and D. Levine (1993) ‘Debt constrained asset markets’, Review of Economic
Studies, 60, (4), 865-88.
Kehoe, P. and F. Perri ‘International business cycles with endogenous market incomplete-
ness’, Econometrica (forthcoming).
Kehoe, T. and E. Prescott (1995) ‘Introduction to the symposium, the discipline of applied
general equilibrium’, Economic Theory, 6, (1), 1-11.
Kollmann, R. (1996) ‘Incomplete asset markets and the cross-country consumption corre-
lation puzzle’, Journal of Economic Dynamics and Control, 20, (5), 945-61.
Koopmans, T. (1965) ‘On the concept of optimal growth’, The Econometric Approach to
Development Planning, Rand McNally, Chicago.
Kravis, I. (1959) ‘Relative income shares in fact and theory’, American Economic Review,
49, (5), 917-49.
Krusell, P. and A. Smith (1998) ‘Income and wealth heterogeneity in the macroeconomy’,
Journal of Political Economy, 106, (5), 867–96.
Kuznets, S. (1946a) National Income: A Summary of Findings, National Bureau of Eco-
nomic Research (NBER), New York.
300 Edward C. Prescott
Kuznets, S. (1946b) National Product Since 1869, National Bureau of Economic Research
(NBER), New York.
Kydland, F. and E. Prescott (1982) ‘Time to build and aggregate fluctuations’, Econometrica,
50, (6), 1345-70.
Kydland, F. and E. Prescott (1991) ‘Hours and employment variation in business cycle
theory’, Economic Theory, 1, 63–81.
Kydland, F. and E. Prescott (1996) ‘The computational experiment: an econometric tool’,
Journal of Economic Perspectives, 10, (1), 69-85.
Lucas, R. (1977) ‘Understanding business cycles’, Journal of Monetary Economics, 5, (0),
7-29.
Lucas, R. (1980) ‘Equilibrium in a pure currency economy’, Economic Inquiry, 18, (2),
203-20.
Lucas, R. (1981) Studies in Business-Cycle Theory, MIT Press, Cambridge, MA.
Lucas, R. (1988) ‘Money demand: a quantitative review’, in K. Brumer and B. McCallum
(eds) Money, Cycles and Exchange Rates: Essays in Honor of Alan Meltzer, Carnegie-
Rochester Conference Series on Public Policy, vol. 29, North Holland, Amsterdam.
Lucas, R. and N. Stokey (1987) ‘Money and interest in a cash-in-advance economy’, Econo-
metrica 55, 491–514.
Marcet, A. and R. Marimon (1992) ‘Communication and growth’, Journal of Economic
Theory, 58, 219–49.
Mas-Colell, A. (1975) ‘A model of equilibrium with differentiated commodities’, Journal
of Mathematical Economics, 2, 263-96.
McCallum, B. and M. Goodfriend (1987) ‘Money, theoretical analysis of the demand for
money’, NBER Working Paper No. 2157.
Prescott, E. (1986) ‘Theory ahead of business cycle measurement’, Carnegie-Rochester
Conference Series on Public Policy, vol 24. Reprinted in Federal Reserve Bank of Min-
neapolis Quarterly Review 10, 9–22.
Prescott, E. and R. Townsend (1984) ‘Pareto optima and competitive equilibria with adverse
selection and moral hazard’, Econometrica, 52, (1), 21-45.
Saving, T. (1971) ‘Transaction costs and the demand for money’, American Economic
Review, 61, (3), 407-20.
Scheinkman, J. and L. Weiss (1986) ‘Borrowing constraints and aggregate economic activ-
ity’, Econometrica, 54, (1), 23-45.
Schreft, S. (1992) ‘Transaction costs and the use of cash and credit’, Economic Theory, 2,
(2), 283-96.
Schumpeter, J. (1954) A History of Economic Analysis, Oxford University Press, New York.
Thomas, J. and T. Worrall (1990) ‘Income fluctuations and asymmetric information, an
example of a repeated principal-agent problem’, Journal of Economic Theory, 51, (2),
367-90.
Townsend, R. M. (1979) ‘Optimal contracts and competitive markets with costly state
verification’, Journal of Economic Theory, 21, (2), 265-93.
Veracierto, M. (1998) ‘Plant level irreversible investment and equilibrium business cycles’,
mimeo.
13 Complexity-based methods
in cycles and growth
Any potential value-added?
William A. Brock
13.1 Introduction
This chapter is written in the style of both a brief survey and a research proposal in
order to present a set of analytical and statistical tools, some of which are related
to what might be called ‘complexity theory’, and to explore whether these tools
have anything to contribute to the study of cycles, growth, and structural change.
The reader is warned at the outset that much of what is said here is extremely
speculative and may end up being a research dead end. Furthermore, this article
will keep mathematics and technical details to a minimum in order to attempt to
reach a wide audience.
I shall give a lot of references to the literature for details. I apologize in ad-
vance for the literary license with subtle ideas that I take here. Defining whether
an analytical, theoretical, or statistical tool is ‘complexity-based’ or a research
methodology is ‘complexity-based’ is hard to do within the scope of a short arti-
cle. So we shall just ask the reader to take a look at ‘complexity-based’ references
to see what the word ‘complexity’ refers to. Many ‘complexity-based’ research
groups have been operating for a long time.
There are, for example, the Brussels School (cf. Day and Chen (1993) for a
sampling from the Brussels/Austin school as well as others), Day’s two volume
work (Day (1997) which covers not only the work of himself and his co-workers
but also work of many others), Rosser’s book ((1991), which gives an exceptionally
broad based coverage of the work of different research groups in this area), and
the Stuttgart School (cf. Weidlich (1991)).
The current article will draw on my association with the Santa Fe Institute (SFI).
The books, Anderson, Arrow, and Pines (1988), (SFI (I)), and Arthur, Durlauf, and
Lane (1997), (SFI (II)) give nice descriptions of ‘complexity theoretic’ approaches
to economics from the ‘SFI’ point of view. For example, let us quote from the
introduction to SFI (II):
But just what is the complexity perspective in economics? That is not an easy
question to answer. Its meaning is still very much under construction. . . . To
describe the complexity approach, we begin by pointing out six features of
302 William A. Brock
the economy that together present difficulties for the traditional mathematics
used in economics.
We list them as follows: (i) dispersed interaction; (ii) no global controller; (iii)
cross-cutting hierarchical organization; (iv) continual adaptation; (v) perpetual
novelty; (vi) out-of-equilibrium dynamics. Notice that all six being present together
is what prevents one from treating them by simply re-interpreting the usual Arrow–
Debreu–McKenzie apparatus of general equilibrium theory with enlarged state
spaces of date-event contingent goods and the like with disequilibrium dynamics
grafted onto this apparatus. For example, Norman Packard argued in his piece
in Anderson, Arrow, and Pines (1988) that the state space’s dimensionality is
constantly changing so we cannot use conventional dynamical systems theory to
analyze the dynamics generated by his complex system.
Readers can be quite assured that the piece they are reading is ‘complexity-
based’ if they encounter the following terms and ideas: (a) path dependence; (b)
self-organized criticality; (c) edge of chaos; (d) power law scaling; (e) renormal-
ization group; (f) fractals and other types of self similarity; (g) genetic algorithms,
emergent computation, (h) adaptive neural networks, complex adaptive systems;
(i) chaos theory, embedding theorems, correlation integrals; (j) interacting particle
systems, statistical mechanics, mean field theory, non-ergodic systems, breakdown
of the law of large numbers.
My paper, (1999), compares and contrasts, for applications to economics, the
SFI (I,II) version of complexity approaches to economics with another closely
related approach which emphasizes a hierarchy of dynamical systems at temporal
and ‘spatial’ scales and uses bifurcation theory from dynamical systems theory.
The latter research style is common in ecology. Rather than repeat what has
been stated already in SFI (I,II) and Brock (2000), we shall just sketch the needed
highlights here and refer to these sources for details. The econometric and statistical
tools reviewed by Brock and Durlauf (1999) will be used to propose a different
twist to the problem of detecting and measuring the ‘endogeneity’ in ‘endogenous’
growth theories. I shall also outline some proposed research strategies to help
narrow the dispute between those who prefer rational expectations models of cycles
and growth and those who prefer ‘boundedly rational’ approaches.
where y(t) denotes observable per capita income, A(t) denotes the ‘level of tech-
nology’ at date t, d log(A(t))/dt = g, b and l are parameters that depend upon other
economic parameters such as the rate of population growth, the rate of depreciation
on capital stock, savings rate, and the like.
DQ show how (DQ.19) emerges from the main received growth theories and
they use it to organize their discussion of forces that shape the ‘convergence com-
ponent’ (the term involving exp(lt)) and the ‘levels component (the rest of the
right-hand side).’ They point out that since log(b), log(A(0)) are unobserved just
about any pattern of ‘cross-country growth and convergence is consistent with the
model.’
Thus we need to model the forces that shape dlog(A(t))/ dt = g, A(0), b, and
l. Hence, while reading the part below, readers might view it as a discussion of
some economic forces that shape g = d log(A(t))/ dt. Readers should imagine that
they have data on a cross section of countries where they are estimating a system
like (DQ.19) with one equation like (DQ.19) for each country. Theory like that
presented below will be used to suggest regressors to be inserted into the RHS of
(DQ.19). Economic forces that lead to diffusion of technologies across different
countries such as international trade lead to cross equation dependence for systems
like (DQ.19). More will be said about estimation of such systems later.
where PV (t) denotes the present value of the stream of net benefits from t on. This
quantity is computed by integrating the present value flow, exp(−rt)S(t) from t
to infinity. One obtains
I2: Monopoly
For monopoly, profit, P(t) = exp(at)[max{D(q)q − cq} = P(0) exp(at). Copy
the procedure above to obtain
Here, as one would expect, building takes place earlier. The ‘winning’ monopolist
builds earlier and captures no PV profit for herself. She does this to avoid being
displaced in the market by an earlier builder if there were still positive net PV
to be had. Spillovers of the amount exp(at)[S(q∗ ) − D(q∗ )q∗ ] are captured by the
public from the building date on.
Here we have an earlier date of building the facility than competing monopolists
for the market in (13.5). Furthermore if F measures the true cost to society as a
whole of undergoing the cost of building, the entire capitalized net benefit from
building is absorbed by the cost. This is truly a society that is enslaved by a rat
race of building too early to service emerging demands. The discussion in Evans
(1983, p. 219) shows how the welfare cost of this kind of rat race can be very large
to the society as a whole. The ‘dual’ case of a rat race of adopting RCRs too early
from a social welfare benchmark will be examined below.
rL − a L − vL = 0 (13.13)
The roots of the characteristic equation for (13.13) are both real with one positive
and one negative. Since a Laplace transform is always positive and bounded above
by unity only the positive root is sensible. Furthermore if S(0) is already at the
target barrier S, the Laplace transform is unity. Hence the solution of (13.13) is
given by
13.2.2 A summing up
We have seen enough variations on the ‘adoption problem’ to sum up what we have
learned. First, the structure I5 stresses the importance of designing an accounting
system upon which award to bonuses to the management of the enterprise for the
unpleasantness of bearing cost F in order to get the gain ‘g’ from an RCR or
identifying and entering a new market. Much of the wave of privatizations and
‘corporatizations’ during the 1980s and 1990s was driven by the recognition that
Complexity-based methods in cycles and growth 317
incentives to innovate were severely distorted by the regulatory framework in many
countries. See, for example, Lewis Evans et al. (1996) for a review of the New
Zealand ‘experiments’ in restructuring to induce efficiency enhancing incentives.
Second, the instantaneous variance in the rate of gain to new market entry or
gains to RCR as well as the instantaneous variance in the rate of fall of adoption
cost is a modelling way of capturing the presence of latent options the size of which
increase with these instantaneous variances. The performance of the four different
structures I2–I5 in capturing these opportunities (measured relative to the public
interest standard I1) can vary widely. Perhaps the worst structure from the social
welfare perspective is a variation of I5 where the management gains a trivial part
of ‘g’ but bears a nontrivial part of F if they adopt.
Third, the list of parameters in the structures I1–I5 can be given more concrete-
ness by looking at a particular industry such as agriculture where cost and benefit
parameters to an individual farm of adopting a technology are influenced by the
regulatory and tax framework; reference group adoptions; activities by government
such as agricultural experiment stations and extension offices; relative input price
shifts which induce innovations in the direction of economizing on relatively more
expensive inputs as in the Ruttan theory of induced innovations; volatility in the
ratio of benefits/costs to adoption of RCRs; and more. See Carlson et al. (1993)
for an excellent review of agricultural and resource economics issues.
y = A + cx + DE(x|g) + e, (13.22)
where A = a/(1 − b), D = (bc + d)/(1 − b)
Notice that a social multiplier, [1/(1−b)−1] = b/(1−b), exists for the regres-
sor E(y|g). I.e. if, for example, the constant term ‘a’ increases by an amount f for
328 William A. Brock
one member of a group g of size |g|, the group mean, E(y|g) increases by a factor
f /[(1 − b)|g|] for a net gain of bf /[(1 − b)|g|] over the case b = 0 of no multiplier.
This is a measure of the size of spillover. We see the source of the econometric
problem raised by (13.21) and (13.22) right away. There are only three regressors
in the reduced form (13.22) but there are four parameters to estimate in the primary
form (13.21). The coefficient of interest, ‘b’, cannot be separated from d. Taking
first differences does not solve the problem because the constant is lost. We still
have one more parameter relative to regressors in the reduced form.
It is easy to see how this kind of situation can arise in a growth regression
exercise inspired by the spillover models of Romer (spillovers to our economic
entity’s output, y, across outputs in reference group g due to learning by doing
spillovers) and Lucas (spillovers from average human capital in reference group
g) (discussed in Barro and Sala-i-Martin (1995), Barro (1997), and Durlauf and
Quah (1999)).
Following Brock and Durlauf (1999), we consider four avenues of potential
escape from this problem. First, one might be able to use economic theory to argue
that a term like cx + dE(x|g) should not appear in the regression equation and that
it should be replaced by a term like cx + dE(w|g) where w may be correlated with
x but is not identical to x. This escape route leads to a reduced form with the same
number of regressors as parameters but the ‘X X ’ matrix of the regression may be
near singular if the correlation between x, and w is strong enough.
A second avenue is to attempt to use economic theory to argue that the linear
combination of x, E(x|g), should be replaced by, for example, a linear combi-
nation of nonlinear functions of x, E(x|g). In this case the ‘X X ’ matrix would
be nonsingular except for ‘hairline’ cases. However the X X matrix may be near
singular if the nonlinearity was weak or the support of x, E(x|g) was narrow so
that there was not enough variation to let the nonlinearity speak loudly enough to
the nonsingularity of X X . A third route of escape is to replace the current values
of E(y|g), E(x|g) in the regression (13.21) above and replace them by their lagged
values or, perhaps, some function of their lagged values. See Manski’s paper in
SFI (II) for a critique of this route of escape as well as some other routes in other
applications than cycles and growth applications. His critique may, however, be
of concern to researchers in cycles and growth. A fourth route of escape is to con-
sider other ways of adducing evidence for spillovers than growth regressions. For
example if one were studying TFP dynamics at the individual firm or individual
industry level, then one could define a relevant comparison group g and attempt
to estimate spillovers from the average adoption time of a particular innovation of
the reference group g upon the time of adoption of the individual firm. Turn now
to one possible way to implement this strategy.
13.6 Summary
This chapter has attempted to bring to the reader’s attention some ‘complexity-
based’ tools and has attempted to assess whether they might add some value to
discussion of issues raised at the Siena School of Summer 1998. These issues
included (i) explaining the wide differences in growth performance across regions,
countries, and periods; (ii) identification of ‘spillover’ effects in growth contexts;
(iii) and the desirability of ‘backing off’ from fully rational expectations modeling.
We addressed these issues by first setting out a little bit of optimal timing theory
and applying it to uncovering forces that lead to faster adoption of RCR’s and faster
development of new markets. We then used this framework to set out five sample
industrial and regulatory settings to stress how to go point-by-point through how
the structure of incentives (but not all of Harberger’s 1001 ways) impact optimal
Complexity-based methods in cycles and growth 335
adoption times. Much was said about the many things that government can do
to help or to hurt. After spending a fair amount of time setting out this context,
we turned to a discussion of channels for potential spillovers. Then we turned to
discussion of the basic econometric problem, following Manski’s paper in SFI (II),
of identifying spillovers in growth regressions. We suggested some ways based on
Brock and Durlauf (1999) to deal with this problem. We then turned to a sketch
of econometrics of hazard function estimation with spillover effects following
Brock and Durlauf (1999) which is close to the spirit of the optimal timing theory
developed above.
The bottom line is this. Complexity-based methods offer stimulating sugges-
tions of potentially useful research strategies for work in the area of cycles and
growth.
References
Acar, E., Satchell, S. (eds), (1998) Advanced Trading Rules, Butterworth-Heineman: Ox-
ford.
Aghion, P., Howitt, P. (1998) Endogenous Growth Theory, MIT Press: Cambridge, Mass.
Akdeniz, L., Dechert, W., (1997), ‘Do CAPM Results Hold in a Dynamic Economy? A
Numerical Analysis,’ Journal of Economic Dynamics and Control, 21, 981–1003.
Alesina, A. (1997) ‘Institutions for Fiscal Stability,’ NBER Reporter, Winter, 1997/8.
Allen, C., Forys, E., Holling, C. (1998) ‘Body Mass Patterns Predict Invasions and Extinc-
tions in Transforming Landscapes,’ Department of Zoology, University of Florida.
Altug, S., Labadie, P. (1994) Dynamic Choice and Asset Markets, Academic Press: New
York.
Altug, S., Ashley, R., Patterson, D. (1999) ‘Are Technology Shocks Nonlinear?’ Macroeco-
nomic Dynamics 3(4), 506–33..
Amendola, M., Gaffard, J. (1998) Out of Equilibrium, Clarendon Press, Oxford: Oxford.
Anderson, P., Arrow, K., Pines, D. (eds) (1988) The Economy as an Evolving Complex
System, Addison-Wesley: Redwood City, California.
Arthur, W., Durlauf, S., Lane, D. (eds) (1997) The Economy as an Evolving Complex System:
II, Addison-Wesley: Redwood City, California.
Baak, S. (1999) ‘Tests for Bounded Rationality with a Linear Dynamic Model Distorted by
Heterogeneous Expectations,’ Journal of Economic Dynamics and Control, 23(9–10),
1517–43.
Barro, R., (1997), Determinants of economic growth, MIT Press: Cambridge, Mas-
sachusetts.
Barro, R., Sala-i-Martin, X. (1995) Economic Growth, McGraw-Hill: New York.
Baumol, W. (2001) ‘Towards the microeconomics of innovation: growth engine of market
economics’, chapter 11 of this volume.
Black, F. (1995) Exploring General Equilibrium, MIT Press: Cambridge, Mass.
Böhm, B., Punzo, L. (2001) ‘Productivity–investment fluctuations and structural change,’
Chapter 3 of this volume.
Brock, W. (1993) ‘Pathways to Randomness in the Economy: Emergent Nonlinearity and
Chaos in Economics and Finance,’Estudios Economicos, 8, 1, 3–55. reprinted in Dechert,
W. (1996).
336 William A. Brock
Brock, W. (1998) ‘Website: http://www.ssc.wisc.edu/ wbrock’, Department of Economics,
The University of Wisconsin, Madison.
Brock, W. (2000) ‘Whither Nonlinear?’ Journal of Economic Dynamics and Control, 24,
663–78.
Brock, W., Dechert, W., Scheinkman, J., LeBaron, B. (1996) ‘A Test for Independence
Based Upon the Correlation Dimension,’ Econometric Reviews, 15, (3), 197–235.
Brock, W., Durlauf, S. (1995[2001]) ‘Discrete Choice with Social Interactions,’ NBER W.P.
5291. [Review of Economic Studies, forthcoming].
Brock, W., Durlauf, S. (1999) Interactions-Based Models, in J. Heckman and E. Leamer,
(eds), Handbook of Econometrics, North Holland: Amsterdam.
Brock, W., Hommes, C. (1997) ‘A Rational Route to Randomness,’ Econometrica, 65, (5),
1059–95.
Brock, W., Hommes, C. (1998) ‘Heterogeneous Beliefs and Routes to Chaos in a Simple
Asset Pricing Model,’ Journal of Economic Dynamics and Control, 22, 1235–74.
Brock, W., Lakonishok, J., LeBaron, B. (1992) ‘Simple Technical Trading Rules and the
Stochastic Properties of Stock Returns,’ Journal of Finance, XLVII, (5), December,
1731–64.
Brock, W., Rothschild, M. (1986) ‘Comparative Statics for Multidimensional Optimal Stop-
ping Problems,’ in Sonnenschein, H., (ed.), Models of Economic Dynamics, Springer-
Verlag Lecture Notes in Economics and Mathematical Systems, 214, Springer-Verlag:
Berlin, 124–38.
Brock, W., Rothschild, M., Stiglitz, J. (1989) ‘Stochastic Capital Theory,’ in Feiwel, G.,
(ed.) (1989), Joan Robinson and Modern Economic Theory, New York University Press:
Washington Square, New York.
Carlson, G., Zilberman, D., Miranowski, J. (1993) Agricultural and Environmental Resource
Economics, Oxford University Press: Oxford.
Chavas, J. (1999) ‘On the Economic Rationality of Market Participants: The Case of Ex-
pectations in the US Pork Market,’ Journal of Agricultural and Resource Economics,
24, 19–37.
Clark, W. (1985) ‘Scales of Climate Impacts,’ Climate Change, 7, 5–27.
Coleman, J. (1990) Foundations of Social Theory, Harvard University Press: Cambridge,
Massachusetts.
Day, R. (1997) Complex Economic Dynamics, MIT Press: Cambridge, Mass.
Day, R., and Chen, P. (eds) (1993) Nonlinear Dynamics and Evolutionary Economics,
Oxford University Press: Oxford.
Day, R. and Pavlov, O. V. (2001) ‘Qualitative dynamics and evolution in the very long run,’
Chapter 4 of this volume.
Dechert, W. (ed.) (1996) Chaos Theory in Economics: Methods, Models and Evidence,
Edward Elgar: Cheltenham.
Dechert, W. (1998) ‘Website: http://dechert.econ.uh.edu‘, Department of Economics, The
University of Houston.
de Fontnouvelle, P. (2000) ‘Informational Strategies in Financial Markets: The Implications
for Volatility and Trading Volume Dynamics,’ Macroeconomic Dynamics, 4(2).
DeLong, B. (1998) ‘Website: http://econ161.berkeley.edu/’, Department of Economics, The
University of California, Berkeley.
Dixit, A., Pindyck, R. (1994) Investment Under Uncertainty, Princeton University Press:
Princeton.
Dosi, G. (2000) Innovation, Organization and Economic Dynamics, Edward Elgar: Chel-
tenham.
Complexity-based methods in cycles and growth 337
Duffie, D. (1988) Security Markets: Stochastic Models, Academic Press: New York.
Durlauf, S. (1993) ‘Nonergodic Economic Growth,’ Review of Economic Studies, 60, (2),
349–66, April.
Durlauf, S., Quah, D., (1999) ‘The New Empirics of Economic Growth,’ Handbook of
Macroeconomics North Holland: Amsterdam.
Evans, D., (ed.) (1983) Breaking Up Bell, North Holland: Amsterdam.
Evans, L., Grimes, A., Wilkinson, B., with Teece, D., (1996) ‘Economic Reform in New
Zealand 1984–95: The Pursuit of Efficiency,’ Journal of Economic Literature, XXXIV,
(4), December, 1856–1902.
Flaschel, P., (2001) ‘Disequilbrium growth in monetary economics: basic components and
the KMG working model,’ Chapter 6 of this wolume.
Glaeser, E., (1997a) ‘The Economics of Cities,’ NBER Reporter, Winter 1997/8.
Glaeser, E., (1997b) ‘Learning in Cities,’ Harvard Insitute of Economic Research, D.P.
Number 1814.
Granger, C., Terasvirta, T. (1993) Modelling Nonlinear Economic Relationships, Oxford
University Press: Oxford.
Gunderson, L., Holling, C., Light, S. (eds) (1995) Barriers and Bridges: To The Renewal
of Ecosystems and Institutions, Columbia University Press: New York.
Hahn, F. (2001) ‘The “exogenous” in “endogenous” growth theory,’ Chapter 15 of this
volume.
Hahn, F., Solow, R. (1995) A Critical Essay on Modern Macroeconomic Theory, MIT Press:
Cambridge, Mass.
Hall, R. (1996) ‘Economic Fluctuations and Growth,’ NBER Reporter, Autumn.
Hansen, L., Sargent, T. (1991) Rational Expectations Econometrics, Westview Press: Boul-
der, Colorado.
Harberger, A. (1998) ‘A Vision of the Growth Process,’ American Economic Review, 88,
(1), 1–32.
Heckman, J., Singer, B. (eds) (1985) Longitudinal Analysis of Labor Market Data, Cam-
bridge University Press: Cambridge, UK.
Horvath, M. (1998) ‘Business Cycles and the Failure of Marginal Firms,’ Department of
Economics, Stanford University.
Iwai, K. (2001) ‘Schumpeterian dynamics: a disequilibrium theory of long run profits,’
Chapter 7 of this volume.
Jog, V., Schaller, H. (1994) ‘Finance Constraints and Asset Pricing: Evidence on Mean
Reversion,’ Journal of Empirical Finance, 1, (2), January, 193–209.
Jovanonic, B., Lach, S. (1997) ‘Product Innovation and the Business Cycle,’ International
Economic Review, 38, (1), February, 3–22.
Judd, K. (1998) Numerical Methods in Economics, MIT Press: Cambridge, Mass.
Kreps, D., Wallis, D. (eds) (1997) Advances in Economics and Econometrics: Theory and
Applications, Cambridge University Press: Cambridge.
Krugman, P. (1996) The Self-Organizing Economy, Basil Blackwell: Oxford.
Kurz, M. (ed.) (1997) Endogenous Economic Fluctuations: Studies in the Theory of Rational
Beliefs, Springer-Verlag: New York.
LeBaron, B. (1998) Website: http://www.unet.brandeis.edu/ blebaron, Department of Eco-
nomics, Brandeis University.
LeBaron, B. (2000) ‘Agent Based Computational Finance: Suggested Readings and Early
Research,’ Journal of Economic Dynamics and Control, 24, 679–702.
Malliaris, A. (1982) Stochastic Methods in Economics and Finance, North Holland: Ams-
terdam.
338 William A. Brock
Mankiw, N., Romer, D., Weil, D. (1992) ‘A Contribution to the Empirics of Economic
Growth’, Quarterly Journal of Economics, CVII, 407–37.
Olson, M. (1996) ‘Big Bills Left on the Sidewalk: Why Some Nations are Rich, and Others
Poor,’ Journal of Economic Perspectives, 10, (2), Spring, 3–24.
Pesaran, M., Potter, S. (eds) (1992) Nonlinear Dynamics and Econometrics, Journal of
Applied Econometrics, Special Issue, Volume 7, Supplement, reprinted as a book by
John Wiley and Sons: New York.
Potter, S. (1995) ‘A Nonlinear Approach to US GNP,’ Journal of Applied Econometrics, 10,
109–25.
Putnam, R. (1993) Making Democracy Work: Civic Traditions in Modern Italy, Princeton
University Press: Princeton.
Prescott, E. (1998) ‘Needed: A Theory of Total Factor Productivity,’International Economic
Review, 30, (3), August, 525–90.
Romer, P. (1998) ‘Website: http://www-leland.stanford.edu/ promer’ Graduate School of
Business, Stanford University.
Rosser, J. (1991) From Catastrophe to Chaos: A General Theory of Economic Discontinu-
ities, Kluwer Academic Publishers: Dordrecht, The Netherlands.
Sargent, T. (1993) Bounded Rationality in Macroeconomics, Oxford University Press: Ox-
ford.
Scheinkman, J., Woodford, M. (1994) ‘Self-Organized Criticality and Economic Fluctua-
tions,’ American Economic Review, 84 (May) 417–21.
Solomou, S. (1998) Economic Cycles: Long Cycles and Business Cycles Since 1870, Manch-
ester University Press: Manchester.
Sullivan, R., Timmerman, A., White, H. (1999) ‘Data-Snooping, Technical Trading Rule
Performance, and the Bootstrap,’ Journal of Finance, 54, 1647–92.
Turnovsky, S. (1995) Methods of Macroeconomic Dynamics, MIT Press: Cambridge, Mass.
Weidlich, W. (1991) ‘Physics and Social Science: The Approach of Synergetics,’ Physics
Reports, 204, (1), May, 2–163.
Wheeler, C. (1998) PhD Thesis, Department of Economics, The University of Wisconsin,
Madison.
14 Information, complexity
and the MDL principle
Jorma Rissanen
14.1 Introduction
In intuitive terms, the objective of statistical modeling is to separate a given data
sequence into useful learnable information and the rest, which may be viewed
just as noninformative noise. The difficulty is in the formalization of the two
constituents: the ‘useful information’ and the ‘noise’. Traditionally, modeling is
done by envoking a metaphysical ‘true’ data generating distribution, which is to
be estimated from the data by minimization of an appropriate mean performance
criterion, which itself is to be estimated from the data. Since the basic issue of
how to formalize the useful information and the noise is not addressed such an
approach cannot provide a rational explanation of why the best approximation of
the ‘truth’ is not the most complex model fitted to the data. To avoid this disastrous
conclusion one has to add an ad hoc term to the criterion to penalize the model
complexity. But because the added term lacks any deeper meaning it does not
reflect adequately the model complexity nor its effect to models’ performance,
and such a metaphysical assumption does not provide a sound basis for a fruitful
theory of modeling.
There are alternative purely empirical approaches to model building, which are
based on reuse of the available data, such as in cross-validation and bootstrapping.
In these a desired data dependent performance criterion is minimized in a portion
of the data and the result tested either on the rest or on a reshuffled version of the
data. Although computationally demanding such approaches may work reasonably
well in individual cases, but, being empirical processes, they cannot serve as a basis
for a theory of modeling.
It seems to us that no solid theory of modeling can be achieved without a formal
definition of ‘information’ in a data sequence, which can be done by a generaliza-
tion of Shannon’s information measure along the lines of the algorithmic theory of
complexity (Hansen and Yu 1998), Kolmogorov 1965). The basic concepts have
undergone a somewhat tortuous evolution, (Rissanen 1986, 1987, 1996), and in this
chapter we outline some of the most recent developments (Barron et al 1998, Bala-
subramanian 1996, Grünwald 1998, Rissanen 2000a, 2000b). Briefly, the stochas-
tic complexity of a data sequence xn = x1 , . . . , xn (or (yn , xn ) = (y1 , x1 ), . . . , (yn , xn )
340 Jorma Rissanen
as in regression problems), relative to a class of parametric probability models
M = {P(xn ; θ)}, is defined to be the shortest code length with which the data xn
can be encoded, when advantage is taken of the models. Such a code length may
be identified with the negative logarithm of a probability distribution P(xn ; M),
which factors as follows
P(xn ; M) = H (xn |θ̂(xn ))Q(θ̂(xn )), (14.1)
where the first factor has no useful information about the data while the second
factor, describing the maximum likelihood estimate θ̂(xn ), has it all. To such a
factorization there often corresponds a decomposition of the data of the kind xn =
x̂n + en , where en is the ‘noise’ part that cannot be compressed with the given
models, and x̂n = F(xn ; θ̂(xn )) is the purely predictable part. We define the code
length for encoding x̂n the amount of information in the data xn that can be retrieved
with the model class M. A similar factorization and decomposition exist also for
data of type (yn , xn ) = (y1 , x1 ), . . . , (yn , xn ), where yn , given another data sequence
xn , is to be modeled.
We may then state that the objective in building models for data is to ob-
tain a decomposition of the kind described, which so far has been done only for
probabilistic models. Hence the goodness of any of the numerous model selec-
tion criteria proposed in the literature can be judged on how well they accomplish
such a decomposition. Also, principles like Occam’s razor and its variants express
only the obvious: redundancy cannot add to the useful information, and since
such a model cannot achieve the stochastic complexity it should be removed. The
reader will recognize the factorization (14.1) as a generalization of the ordinary
sufficient statistics factorization, which includes parameters, to a universal suffi-
cient statistics factorization, which has no parameters. This, in turn, is an analog
of Kolmogorov’s sufficient statistics decomposition in the algorithmic theory of
complexity (Cover and Thomas 1991) . A perfect separation of the noise from the
useful information is possible only for special model classes, but asymptotically
it can be done for all the usual model types. A perfect nonasymptotic separation
exists in the linear quadratic regression case, (Rissanen 2000a), which we describe
below.
14.2 Models
We begin with a brief discussion of the formal definition of a model. We consider
sets X and Y and their cartesian product X × Y together with the extension X n × Y n
to sets of strings of length n. Perhaps the most common type of model is defined
by a function F : X n → Y n , together with an error function δ(yn , F(xn )), for which
we take the logarithmic one δ(yn , F(xn )) = − log f (yn |x̂n ), where x̂n = F(xn ) and
f (yn |x̂n ) is a conditional density function. Many of the usual error functions, above
all the quadratic one, define conditional density functions. Of particular interest to
us are the parameteric models, Mγ = {f (yn |xn ; γ, θ)}, where γ is a structure index,
such as the pair of orders p, q in ARMA models, and θ = θ1 , . . . , θk ranges over
some, usually compact, subset of the k-dimensional euclidean space, k depending
Information, complexity and the MDL principle 341
,
on γ, such as k = p + q in the ARMA models. Put M = γ Mγ , an example
of which is the set of all ARMA models. The parameters θ often include both
parameters in the function F and others in the conditional probability or density
function, such as the variance. Finally, an important special case is the one where
the data sequence xn is absent, in which case we write xn rather than yn for the
single data sequence, and f (xn ; γ, θ) for the models. The theory is similar for both
types of data.
f (X n ; θ̂(X n ), γ)
min max Eθ log , (14.2)
q θ q(X n )
where q(xn ) is any density function and the expectation is with respect to f (xn ; θ, γ).
It is clear that the models in the class Mγ cannot express all the statistical
properties in real world data sequences xn , no matter how large n is. For instance,
if we generate the data with a density function g(xn ) which is outside the class,
the data will have statistical properties different from those expressible with the
models in the class. Notice, that we do not want to make the claim that the data are
a sample from any distribution. Rather, we are simply using density functions to
describe statistical properties in the data. This suggests that we should generalize
the minmax problem (14.2) as follows
342 Jorma Rissanen
f (X n ; θ̂(X n ), γ)
min max Eg log , (14.3)
q g∈G q(X n )
where G is a class larger than Mγ . In fact, we can let G consist of all distributions
such that G = {g : Eg log(g(X n )/f (X n ; θ̂(X n ), γ) < ∞}. This excludes the singular
distributions, which clearly do not restrict the data in any manner and hence do not
specify any properties in them. Also, both the minimum and the maximum will be
reached.
Theorem 14.1 If Ω is such that the integral
Cn (γ) = f (yn ; θ̂(yn ), γ)dyn (14.4)
θ̂(yn )∈Ω
is finite, the solution to the minmax problem (14.3) is the universal NML (normal-
ized maximum likelihood) model
f (xn ; θ̂(xn ), γ)
fˆ(xn ; γ) = . (14.5)
Cn (γ)
Clearly,
f (X n ; θ̂(X n ), γ)
Eg log = log Cn (γ) (14.6)
fˆ(X n ; γ)
for all g.
The proof is given in Rissanen 2000b.
Interestingly, for discrete data the solution P̂(xn ; γ) also solves Shtarkov’s
minmax problem (Shtarkov 1987),
P(xn ; θ̂(xn ), γ)
min max log = log Cn (γ). (14.7)
Q n x Q(xn )
Notice the important fact that the best model fˆ involves only the models in the class
Mγ . Under certain conditions, satisfied for the classes of exponential distributions
(Rissanen 1996),
k n
log Cn (γ) = log + log |I (θ)|dθ + o(1), (14.8)
2 2π Ω
One can show that − log fˆ(xn ; γ) and − log fw̄ (xn ; γ) behave similarly for large n.
to be the stochastic complexity of the data string xn , relative to the model class
Mγ . We can rewrite it as
fˆ(X n ; γ̂(X n ))
min max Eg log , (14.19)
q g q(X n )
where the indices in γ pick out the components of the regressor variables and define
the k × n matrix Xγ = {xit : i ∈ γ}. Write Zγ = Xγ Xγ = nΣγ , which is taken to be
positive definite. The development for a while will be for a fixed γ, and we drop
the subindex γ in the matrices above as well as in the parameters. The maximum
likelihood solution of the parameters is given by
β̂(yn ) = Z −1 X yn (14.22)
1
τ̂ (yn ) = (yt − xt β̂ (yn ))2 . (14.23)
n t
We then have the NML density function itself for 0 < k < m
n k R n−k k 4 n
− log fˆ(y|x; γ, τ0 , R) = ln τ̂ (y)+ ln −ln Γ( )−ln Γ( )+ln 2 + ln(nπ).
2 2 τ0 2 2 k 2
(14.31)
In order to get rid of the two parameters R and τ0 , which clearly affect the
criterion in an essential manner, set them to the values that minimize (14.31): R =
R̂(y) = n−1 β̂ (y)Σβ̂(y) and τ0 = τ̂ (y). However, the resulting fˆ(y|x; γ, τ̂ (y), R̂(y))
is not a density function. We rectify this by the same normalization process as
above:
fˆ(y|x; γ, τ̂ (y), R̂(y))
fˆ(y|x; γ) = , (14.32)
Y f (z|x; γ, τ̂ (z), R̂(z))dz
ˆ
where the range
Y = {z : τ1 ≤ τ̂ (z) ≤ τ2 , R1 ≤ R̂(z) ≤ R2 }
will be defined by four new parameters. Again the integration can be performed,
(Rissanen 2000a), and the negative logarithm of fˆ(y|x; γ) is given by
n−k k n−k k
− ln fˆ(y|x; γ) = ln τ̂ (y) + ln R̂(y) − ln Γ( ) − ln Γ( )(14.33)
2 2 2 2
2 n τ2 R2
+ ln + ln(nπ) + ln ln . (14.34)
k 2 τ1 R1
This time the last term involving the new parameters does not depend on γ nor
k, and we do not indicate the dependence of fˆ(y|x; γ) on them. Almost the same
criterion was obtained in Hanson and Yu 1998 by evaluation of a mixture density
for Zelner’s prior.
As a final step we wish to extend the density function fˆ(y|x; γ) to the larger
class of models, defined as the union over all the index sets γ, and to obtain a
criterion for finding the optimal index set and the associated optimal model. We
begin with the MDL estimator γ̂(·), obtained by minimization of the ideal code
length for the data − ln fˆ(y|x; γ) with respect to γ. Although the result fˆ(y|x; γ̂(y))
is not a density function we get one by the normalization process
fˆ(y|x; γ̂(y))
fˆ(y|x; Ω) = , (14.35)
γ̂(z)∈Ωfˆ(z|x; γ̂(z))dz
348 Jorma Rissanen
where Ω is a set of indices such that it includes γ̂(y). The denominator, call it C,
is given by
C= P̂n (γ), (14.36)
γ∈Ω
where
P̂n (γ) = fˆ(z|x; γ̂(z))dz. (14.37)
{z:γ̂(z)=γ}
In analogy with fˆ(y|x; γ) we call fˆ(y|x; Ω) the NML density function for the
model class with the index sets in Ω, and we get the final decomposition
where we include in Const all the terms that do not depend on the optimal index
set γ̂ of size k̂. The terms other than the first define the length of a code from
which the optimal normal model, defined by the ML parameters, can be decoded,
while the first term represents the code length of the part of the data that adds no
further information about the optimal model. It may be viewed as noise. Hence
this decomposition is similar to Kolmogorov’s sufficient statistics decomposition
in the algorithmic theory of information, and it is also seen to extend the ordinary
sufficient statistics, as defined for certain parametric families, to parameter free
universal sufficient statistics.
By applying Stirling’s approximation to the Γ-functions we get the NML cri-
terion for 0 < k ≤ m
n
min{(n − k) ln τ̂ (yn ) + k ln(nR̂(yn )) + (n − k − 1) ln − (k + 1) ln k},
γ∈Ω n−k
(14.39)
where k denotes the number of elements in γ.
It seems that in order to find the minimizing index set γ we must search through
all the subsets of the rows of the m × n regressor matrix X . However, this can be
avoided if we make a linear transformation AX of the regressor matrix such that
AX XA /n = I . Let the new ML parameters for the maximal number of rows k = m
be α̂ = α̂1 , α̂2 , . . . , α̂m and σ̂ 2 . Further, let α̂(1)
2
≥ α̂(2)
2
≥ . . . ≥ α̂(m)
2
so that (i) is
the index of the i’th largest parameter in absolute value. Then the criterion (14.39)
is equivalent with
n
min{(n − k) ln(α̂ α̂ − R̂) + k ln(nR̂) + (n − k − 1) ln − (k + 1) ln k},
k n−k
(14.40)
where R̂ is the sum of either the k largest or the k smallest squares α̂i2 (Rissanen
2000a). Hence, the optimum γ can be found with no more than m evaluations of
the criterion.
For the so-called denoising problem an orthonormal regressor matrix is easily
obtained with wavelets, and the criterion (14.40) provides a natural separation of
Information, complexity and the MDL principle 349
noise from the data as its incompressible part, which appears to be superior to any
ad hoc criterion; for numerical examples see Rissanen 2000a.
14.7 Conclusions
We have described a formal definition of complexity and (useful) information in a
data sequence, as a foundation for a theory of model building. These notions can
be obtained to a good approximation from a universal NML (Normalized Maxi-
mum Likelihood) model for parametric model classes as a solution to a minmax
problem. Theorems exist which demonstrate that such a universal model provides
an extension of the usual sufficient statistics decomposition to a parameter free
universal sufficient statistics decomposition, and accomplish the desired decom-
position of the data. For a collection of model classes the best can be found with
the MDL principle. We illustrate such a decomposition and the resulting criterion
for the Gaussian family in the basic linear regression problem, for which they can
be computed exactly even for small amounts of data.
References
Akaike, H. (1974), ‘Information Theory and an Extension of the Maximum Likelihood
Principle,’ Second international symposium on information theory, ed. B.N. Petrov and
F. Csaki, Akademia Kiedo, Budapest, 267–81
Balasubramanian, V. (1996)‘Statistical Inference, Occam’s Razor and Statistical Mechanics
on the Space of Probability Distributions’, Neural Computation, 9, (2), 349–268, 1997
http://arxiv.org/list/nlin/9601
Barron, A.R., Rissanen, J. and Yu, B. (1998) ‘The MDL Principle in Modeling and Cod-
ing’, special issue of IEEE Trans. Information Theory to commemorate 50 years of
information theory, IT-44, (6), October 1998, pp 2743–60
Clarke, B. S. and Barron, A. R. (1990) ‘Information-Theoretic Asymptotics of Bayes Meth-
ods’, IEEE Trans. Information Theory, IT-36, (3), 453–471, May 1990.
Cover, T. M. and Thomas, J. A. (1991) Elements of Information Theory, Wiley, New York,.
Dom, B. (1996) ‘MDL Estimation for Small Sample Sizes and Its Application to Linear
Regression’, IBM Research Report RJ 10030, June 13, 1996.
Grünwald, P.D. (1998) The Minimum Description Length Principle and reasoning under
Uncertainty, PhD thesis, Institute for Logic, Language and Computation, Universiteit
van Amsterdam.
Hanson, M. and Yu, B. (1998) ‘Model Selection and the Minimum Description Length
Principle’, to appear in Journal of American Statistical Association, June, 2001; see also
http://cm.bell-labs.com/stat/doc/mdl.pdf
Kolmogorov, A.N. (1965) ‘Three Approaches to the Quantitative Definition of Information,’
Problems of Information Transmission 1, 4–7.
Li, M. and Vitanyi, P.M.P. (1997) An Introduction to Kolmogorov Complexity and Its Ap-
plications, second edition, Springer-Verlag, New York.
Merhav, N. and Feder, M. (1995) ‘A Strong Version of the Redundancy-Capacity Theorem
of Universal Coding’, IEEE Trans. Information Theory, IT-41, (3), 714–22.
Myung, I., Balasubramanian, V. and Pitt, M. (2000) ‘Counting Probability Distributions:
Differential Geometry and Model Selection’ Proceedings National Academy of Science,
97, 21 (October 2000), 11170–5.
350 Jorma Rissanen
Rissanen, J. (1986) ‘Stochastic Complexity and Modeling’, Annals of Statistics, 14, 1080–
100
Rissanen, J. (1987) ‘Stochastic Complexity’, J. Royal Statistical Society, Series B, 49, (3)
(with discussions) 223–65
Rissanen, J. (1996) ‘Fisher Information and Stochastic Complexity’, IEEE Trans. Informa-
tion Theory, IT-42, (1), pp 40–7
Rissanen, J. (2000a) ‘MDL Denoising’, IEEE Trans. on Information Theory, IT-46, (7),
November 2000. Also http://www.cs.tut.fi/ rissanen/.
Rissanen, J. (2000b) ‘Strong Optimality of Normalized ML models as Universal Codes and
Information in Data’, http://www.cs.tut.fi/ rissanen/ (to appear in IEEE Trans. Informa-
tion Theory, 2001
Shtarkov,Yu. M. (1987) ‘Universal Sequential Coding of Single Messages’, Translated from
Problems of Information Transmission, 23, (3), 3–17.
White, H. (1994) Estimation, inference, and specification analysis, Cambridge University
Press, Cambridge, UK.
15 The ‘exogenous’ in
‘endogenous’ growth
theory
Frank Hahn
15.1 Introduction
In this chapter I examine the claim of the new growth theories that they provide
an ‘endogenous account’ of economic growth. The first question is: what would
count as an endogenous account? Or perhaps better: what would not count? For it
is easier to answer this.
The Solow model [1956] is perhaps the paradigm of an exogenous growth
model. In that model the rate of growth of Harrod-neutral technical progress is
taken as given. Of course the type of technical improvement is also fixed. On
the other hand economic theory is used to show that every equilibrium path of
the economy seeks the steady state. Many if not most of current theories do not
even ask or answer the question of convergence. Yet again Solow does not explain
why we need only consider equilibrium paths (Harrod had distinguished between
the ‘warranted’ and the ‘ actual’ growth rate). But the same is true of current
‘endogenous’ theories. One can continue . . . for instance the saving rate is given
in Solow while it is deduced from the maximisation of a Ramsey integral of a given
intertemporal utility function in many ‘endogenous’ theories.
So it seems clear that while it is easy to distinguish between endogenous and
exogenous variables it is uncertain how to distinguish theories by the same criteria.
On reflection one reaches the conclusion that all theories in all subjects must base
themselves on some exogenously given elements. But in that case what is it about
recent growth theories that entitles them to be called ‘endogenous’? It seems that
the answer is simply that the exogenously given rate of technical progress has
been abandoned in favour of a stochastic profit maximising use of resources to
search for technical improvements. This can be combined with an intertemporal
utility maximising choice of training. Since both can involve increasing returns
and externalities some fairly drastic changes in received theory may result: e.g.
multiple steady states and non-perfect competition. But all arguments are derived
from equilibrium conditions and functional forms are chosen to ensure that some
steady state equilibrium exists.
Clearly the scope of the theory has been enlarged. Has this been sufficient to
earn the character of an ‘endogenous’ theory? It is time to ask what we mean by the
352 Frank Hahn
question. In an endogenous theory all the elements which explain and determine
growth are explained as an outcome of the rational decisions of agents. So one
may be willing to accept rationality of agents as exogenously given. Or more
ambitiously it may be deduced from less immediate postulates, e.g. a Darwinian
struggle for survival. How the theory proceeds on this has important consequences
for its scope; to how many societies can it be applied?
Let us stick to a less ambitious interpretation to see what is needed. Rationality
of decisions (or of beliefs) by itself will not get us very far. We shall surely need
to be given the preferences of agents, their distribution, their endowments and
their distribution. Of course we shall need to know what information is available.
In preferences one includes attitudes to uncertainty and beliefs. One also must
find a way to describe the educational endowments and the production process of
education. The structure of production i.e. the mix of industry and agriculture as
well as the mix of competitive and imperfectly competitive activities all need to
be specified or explained. So will the endowment of public goods.
The list is large but I have included in it only what seems most relevant for
growth and have as yet not even mentioned an account of the manner of technical
progress.1 This will be discussed below. At this stage I hope that it will be clear
why I regard the somewhat hopeless task of a general definition of ‘endogenous’ to
be unnecessary. It is clear to the naked eye that functions describing the outcome
of the choice of education, or the acquisition of new technological knowledge by
a choice of R and D will not suffice for a model of growth in which the most
important elements are themselves explained, by the model. That of course does
not mean that these partially endogenous models are not valuable. But it does
suggest that it would be useful for consumers of these theories to be warned that
they are not being offered a theory of economic history. This chapter constitutes
such a warning.
where nt+1 is the amount of labour in R and D after the (t + 1)th innovation and
πt+1 is the flow profit of the (t + 1)th intermediate good monopolist. They interpret
(r + λφ(nt+1 )) as the ‘obsolescence adjusted interest rate’ and define
There must be no profit that can be made (in equilibrium) by reallocating labour
between the production of intermediate goods and research. Hence
n + 0 (w) = L
This can easily be shown to yield a unique steady state growth rate which will be
an increasing function of λ and γ and L and a declining one in r.
Schumpeter believed that competition increased the rate of innovations. If we
write F(x) = xα (0) the parameter α is a measure of the degree of competition and
the derived demand curve faced by the monopolist has an elasticity 1/(1 − α). So
now
π = wx(1 − α)/α ≡ w(L − n)(1 − α)/α
So A becomes
So n∗ is declining in α and competition is bad for growth. (It is not clear that this
is a faithful interpretation of Schumpeter.)
The above are the bare bones of A. H and the simplest. By introducing many
sectors, capital into R and D and a sequential adoption rate they can attain an-
swers which make competition good or bad for growth depending on parameter
values. They are to be praised for isolating some of the key issues which include
obsolescence and the foresight of it.
It is not the lack of realism of the construction which is of interest here, but
rather the question: is it an endogenous model of growth. It is clear that some
of the more important parts are exogenous. Not only λ, γ and α which play a
354 Frank Hahn
considerable part but the functional forms, the postulate of universal risk neutrality,
the homogeneity of the ability composition of labour, the perfect information of
agents and the postulate that there is either a steady state or that markets clear at
all dates. Of course even pretty closed theories depend on parameter values: for
instance in particle physics, the spin of the electron. But physicists can measure
these parameters, to put it no more strongly, better than we can the ones relevant to
us. It is precisely our relative inability to measure that leads to exogenous variables.
That and the immense complexity of an interdependent group of variables.
I hope that it will not be regarded as presumptuous if I sketch a model I
published in 1990 (for a 1988 conference) to illustrate the latitude plausibility
gives us in functional specification of R and D. Nowadays this approach is rather
commonplace.
Unlike AH I took the production function of firm as Cobb–Douglas:
yt = at lt kt
cr (r, R) > 0 all (r, R), crr (r, R) > 0 when r = R, crr + crR = 0 if r = R.
One can find many plausible arguments for those kinds of externalities. These
include those given by AH but also the likelihood that a higher R will lead to a larger
community of research workers and a greater familiarity with research methods.
For the rest I do not at all improve on A H. R and D is carried out inside the
producing firm and there are no specialised research firms. I then use the calculus
of variations to find the firm’s present value profit maximising plans. To this I add
the equilibrium condition of self-finance which is clearly implausible. Moreover I
take an exogenous Solowian saving ratio. That too is limiting, but the model can be
rectified in the Chicago direction rather easily but at the Chicago cost that savers
have perfect foresight. (Actually in a deterministic setting such as here this would
not be as silly as usual.) There is only one good produced.
It turns out that there are three steady states: one with no research at all so that
α is the Solowian rate of technical progress. In all steady states r = R but there is
more research in one than the other. Of course in steady state all firms take R as
given when they decide on r, (and only two can be stable).
I think that I prefer the AH model with its spirit of struggle for existence and
its explicit attention to obsolescence.3 On the other hand in my version the role
The ‘exogenous’ in ‘endogenous’ growth theory 355
of R is also plausible. It indicates whether one lives in a world where research is
a habit or not, and my formulation suggests that it may or may not be the case.
This gives just a little greater flavour of endogeneity. On the other hand again I,
no more than others, distinguish labour by its ability to research and produce. But
my point is not realism. It was to underline how much that is taken as endogenous
here really is not. This brings me to my next point.
15.3 Equilibrium
When we say that a variable z is endogenous to a model we mostly mean that it is
determinable by equilibrium conditions. In growth theory this involves expecta-
tions in an essential way. In my view they also involve risk attitudes in an essential
way. For instance it seems that in nineteenth century England risk aversion was less
than in Germany. One also needs to make postulates concerning the information
flows and the ability to use them. So when one writes down equilibrium equations
one is bringing into the story many elements not explained by any theory.
If we pay explicit attention to these then we may find our way to a theory which
can indeed explain our observation – say economy A grows slowly because it has
many risk-averse producers – say farmers. But it is not clear whether economists
can say why this should be so.
My point as usual here is not the usual grouse concerning lack of realism. It
is that when we use the canonical paradigm of economic theory we are rarely in a
position to attain purely endogenous results. Or rather what seems like those are
not genuinely so.
As a good example take the influential Chicago method. It needs to postulate
that expectations are rational and look at rational expectations equilibria. The time
horizon for these expectations is infinity. There are possible learning theories –
which lead to rational expectations over finite time but, in the nature of the case,
such expectations over the infinite future could not be proved. But equally seriously,
almost all of the authors make no allowance for risk attitudes or, rather, postulate
risk-neutrality. In our theories of course this means that utility functions are taken
to be exogenous. This in my view is the right attitude to take by economists.
But there is almost surely a distribution of such functions so that the more subtle
hypothesis that tastes etc. are stable through infinite time needs to be invoked. This
is certainly an exogenous piece of theory. For long run growth theories it does not
seem persuasive. In addition Chicago always postulates perfect competition and
production possibilities which allow this. All improvements in technique are either
in intermediate goods or due to improved education but, as in Marshall, external
to the competitive firm. This feature is not explained by the theory and hence an
exogenous element. I do not know whether it is claimed that the theory fits the
facts (so far), but I doubt it, since it is too aggregated to be confronted with some of
the most important episodes in growth, e.g. the introduction of the steam engine,
the internal combustion engine, electricity, computers, etc.
This discussion mirrors one in evolution. Gould maintains if exogenously given
stochastic geological events had been different than they have been, present day
356 Frank Hahn
creatures would not only be different but intelligence might never have arisen. The
Cambridge professor of evolution believes that whatever the external geological
etc. events intelligence would have evolved – I assume because of the advantage it
confers. In our context this is something of a pro-Chicago argument, since it argues
that competition and intelligence are almost bound to lead to growth whatever
particular form it takes. But that is a good deal more modest claim. In particular
it could be made without the full equilibrium paraphernalia since in this context it
is bound to be ill-defined without imports of exogenous factors.
As a matter of fact there are theories which give a fully endogenous account of
why a (Nash) equilibrium should occur. But they rarely include technical progress
and, what is worse, the equilibrium is not unique. This then requires initial condi-
tions and a more or less exogenously specified process of the evolution of strate-
gies to get us where we want to be. A variety of these have been proposed but
the outcome is more in the spirit of Gould since some processes converge to the
risk-dominant rather than the Nash equilibrium.
15.5 Information
Risk-attitude and competence characterised the Schumpeterian entrepreneur. Their
number Schumpeter believed to depend on the culture including the religion of the
economy. Interestingly enough he thought that the ‘routinisation’of R and D would
not only lead to the entrepreneur’s obsolescence but to a slow down in technical
progress. But in any case R and D does depend essentially on people capable of
generating new knowledge not only inside the R and D outfit but outside it.
The DNA revolution has transformed R and D in pharmaceutical industries
and agro-businesses. But this revolution in knowledge was brought about by ex-
ceptional people the fraction of which in the population is surely at present exoge-
nously given. It also, of course, depended on these people having the opportunity to
do their revolutionising, on the transmission of this knowledge and on sufficiently
educated people to recognise the importance of what has been accomplished, (see
Arrow [1974]). Clearly this brings education into the picture (see below) and the
whole is permeated by externalities.
There are many examples of revolutionary inventions which for a long time
were ignored by those who could have profited from their use. The jet engine is
one, as is the fluorescent bulb. I do not know the facts sufficiently well to make
the point completely convincing: to recognise the benefit of an advance is almost
as important as the advance itself. For instance it is reported (Nelson and Wright
[1992] in their excellent survey) that when electronic computers first emerged to
satisfy the US military, the general view, including at IBM, was that it would have
few civilian uses and the invention languished for quite some time. This seems to
me more than a case of mistaken expectations but rather a lack of imagination.
People with that kind of imagination are rare and certainly their representation in
the economy is a matter of culture and largely exogenous for the economist. It is
not clear whether one could regard it as an output of education. (See below.)
But the basic point is (i) finding new information, (ii) whether and how it is
transmitted and (iii) how widely it can be used. I doubt that Crick and Watson
358 Frank Hahn
needed the incentive of monetary rewards to generate DNA information. How-
ever, there were other rather obvious incentives (which I have not seen discussed
in the literature). Certainly adoption by commercial firms or even feeding new
information into R and D depends on their calculation of probable rewards. These
in turn, as Schumpeter argued, depend on the estimated length of money rents (and
on patent law). Clearly that is part of an ‘endogenous’ argument, but only a part.
Many of the relevant elements seem clearly exogenous.
Of course much information is available by publication and word of mouth.
But because information is a public good its transmission, if it is valuable, will
be imperfect. Some information when it is directly or indirectly revealed will
be a surprise to agents who cannot be assumed to have already adapted to it in
their plans. Except, that is, for one thing: the positive probability given to the
appearance of new innovations. This aspect, ‘creative destruction’ is well treated
by AH and of course enters growth as a negative element. But the probability of
such obsolescence will vary from case to case, e.g. the biro does not, and computers
do, seem to run this risk with low and high probability. Once again exogenously
given risk attitudes play as large a part as probabilities. These in turn depend on
the time horizon (effective discount rate) of the agents concerned.
Nelson and Wright [1992] report the predominant view of historians that the
early US productivity superiority (from about 1840) to 1960 had a great deal to
do with superior organisation, abundant raw material and a large domestic market.
(The US for long was a high tariff economy.) None of this seems to be captured
by existing models of endogenous growth.4 Organising and managing etc. is not
a matter of R and D but rather the fruit of a culture which led to the emergence
of a ‘correct’ way to organise and manage. They emphasise that the education
system did not produce many outstanding scientists who had to go to Europe to
study. It is true that high real wages made general education advantageous and that
a generally educated workforce helped to sustain the US system of production.
But later, immigration of ill-educated workers does not seem to have led to a slow
down. Certainly information was of almost a collective sort – one knew how others
organised and so the usual caveats do not apply.
However after the second world war and the rise of ‘science based’ industries,
this picture changes. In particular there was a great growth in resources devoted
to R and D in the US and secrecy of certain aspects of a firm’s activities became
important. But it was never watertight. The rise of science based industries itself
depended on the progress of science which, in turn, depended on the evaluation of
the benefits it would confer. It is interesting to note that the government played a
significant role in this (partly through universities, partly through subsidies).
As a description these (and other aspects in particular education) matters are
very clear. How to convert this into an endogenous theory of growth is not. The
very fast growth in scientific knowledge seems like an exogenous cumulative
process. Once the genetic code was discovered for instance, many researchers
were attracted to this field and new knowledge rapidly accumulated. This sort of
thing is hard to embody persuasively in a functional form and even if it could be,
it would be quite unclear whether it also embodies other scientific advances, e.g.
The ‘exogenous’ in ‘endogenous’ growth theory 359
in superconductivity. Least clear is whether these various theoretical descriptions
can be made to yield a steady state with exponential growth.
But of course post-war scientific advances, for a time, led to greater encour-
agement of fundamental science and the hiring of scientists by firms. So of course
there is something ‘endogenous’ to economic models of growth. My contention
is that it is only ‘something’. For instance, many fundamental sciences by now
receive greatly reduced subsidies from government and firms often use scientists
as managers. There are still considerable risks in new science based economic ven-
tures, for instance in bio-technology. Everyone sees ever more scientific knowledge
emerging, but there is not only uncertainty as to timing but as to kind. This brings
me back to risk attitudes.
Acknowledgement
I am grateful to Kenneth Arrow and Robert Solow for comments.
Notes
1 Kenneth Arrow (in conversation) interprets ‘endogenous’ in this way, that is, to get a
growth equation on the basis of rational choice. He takes Solow’s given rate of growth
as the point of departure. There is no monopoly in terminology and if ‘endogenous’ is
taken to have the Arrow meaning this chapter has the wrong title. I however regard the
word to apply to a theory which does not treat as exogenous what economic theory is
in principle equipped to explain. Moreover one is likely to be led into error when the
full set of factors taken to be exogenous is not enumerated.
2 If one postulates risk neutrality, stochastic elements present no technical problems.
3 Solow has pointed out to me that some innovations may be complementary to existing
technology.
The ‘exogenous’ in ‘endogenous’ growth theory 365
4 What is sometimes discussed is the relative efficiency of various organisation forms
(largely connected with the absorption of information) but it is not integrated into what
we are used to call the production set. Of course these findings suggest that there were
important increasing returns to scale. But it may also be that a large market reduces
risk.
5 This may be taken to mean that the productivity of educational resources increases with
the general educational level of society or that the cost of acquiring extra education by
an individual is a decreasing function of his educational level.
References
Aghion, P. and Howitt, P. (1998) Endogenous Growth Theory, M.I.T. Press.
Arrow, K. J. (1962) ‘The Economic Implications of Learning by Doing’, Review of Economic
Studies, 29, 155–73.
Arrow, K. J. (1974) The Limits of Organisation, Norton and Company.
Hahn, F. (1990) ‘Solowian Growth Models’ in Growth Productivity Unemployment, (ed.)
by P. Diamond, MIT Press.
Jorgenson, D.W. (1961) ‘Development of a Dual Economy’, Economic Journal, 71, June,
309–34.
Kaldor, N. (1957) ‘A Model of Economic Growth’, Economic Journal, 57, 591–624.
Lucas, Robert. E. (1988) ‘On the Mechanics of Economic Development’, Journal of Mon-
etary Economics, 22, 3–42.
Nelson and Wright, G. (1992) ‘The Rise and Fall of American Technological Leadership’,
Journal of Economic Literature, 30.
Phelps, E. (1966) ‘Investment in Humans, Technological Diffusion and Economic Growth’,
American Economic Review, 61, 69–75.
Radner, R. (1993) ‘The Organisation of Decentralised Information Processing’, Economet-
rica, 62, 1109–46.
Romer, P. M. (1986) ‘Increasing Returns and Long-Run Growth’, Journal of Political Econ-
omy, 94, 1002–37.
Romer, P. M. (1990) ‘Endogenous Technological Change’, Journal of Political Economy,
98, S71–S102
Solow, R. M. (1956) ‘A Contribution to the Theory of Economic Growth’, Quarterly Journal
of Economics, 70, 65–94.
Spence, M. (1973) ‘Job Market Signalling’, Quarterly Journal of Economics, 87, 355–74.
366 Index
Index 367
Index