International Review of Financial Analysis 51 (2017) 69–81

Contents lists available at ScienceDirect

International Review of Financial Analysis

Asymmetric effects of the international transmission of US financial

stress. A threshold-VAR approach
Anastasios Evgenidis a,⁎, Athanasios Tsagkanos b
a
Monetary Policy Division, Central Bank of Ireland, Dublin, PO Box 559, Ireland
b
Department of Business Administration, University of Patras, PO Box 1391, Greece

a r t i c l e i n f o a b s t r a c t

Article history: During period of financial stress, the effect of financial stress shocks on economic activity might be different from
Received 1 August 2016 what is usually observed in normal times. This paper investigates the transmission of financial stress episodes on
Received in revised form 15 March 2017 the macroeconomy during stressed and normal periods and it explores how these events are transmitted to the
Accepted 30 March 2017
Eurozone. We find that, a detrimental US financial shock leads to a worsening in economic and financial condi-
Available online 31 March 2017
tions both domestically and in the Eurozone. In addition, during turmoil times, financial accelerator mechanism
JEL codes:
amplifies and propagates the transmission of US financial stress shocks to the Eurozone by reducing its economic
C32 activity. Moreover, small financial stress shocks, rather than infrequent large ones, are able to create large fluctu-
E44 ations in inflation rates. Last, the effect of a detrimental shock in financial conditions has larger negative effects in
F42 the economy compared with the positive effects that would be generated by a beneficial shock in financial
conditions.
Keywords: © 2017 Elsevier Inc. All rights reserved.
Financial stress
Financial stability
Threshold-VAR
Monetary policy

1. Introduction shocks on the macroeconomy during periods of intense financial stress

and explore how these events are transmitted to the Eurozone.
The continuous increase in the degree of interdependence among fi- Certainly, during periods of economic downturns or financial stress,
nancial markets during the last decades has amplified the horizons of the effect of financial stress shocks on economic activity might be differ-
potential investors and created the conditions for benefits (either direct ent from what is usually observed in normal times. During such “bad”
or indirect) to economies through the exploitation of transmission periods, the quality of the assets of financial institutions deteriorates ei-
channels. However, there is the other side of the coin related to the ad- ther because the number of non-performing loans increases or/and be-
verse effects of financial integration. A very recent example of such an cause negative economic sentiment in financial markets reduces the
event is the Lehman Brother's collapse in 2008 which triggered an eco- value of other financial assets. Hence, it is relevant to examine whether
nomic turmoil in the global financial system. This clearly demonstrated and how the effects of financial stress on economic activity and inflation
that, during the recent global financial crisis, financial integration has differ in times of financial instability. In light of the above discussion,
increased the countries' vulnerability to financial crises originating else- this paper discusses and uncovers how a deterioration in financial con-
where (Moshirian, 2008). In the Eurozone, the transmission of financial ditions affects the US economy, whether the Eurozone is affected by a
distress in the real economy evolved at record speed, with tighter credit US shock in financial stability and if there is an asymmetric transmission
standards and sagging confidence hitting business investment and of the shock according to different states of the economy (i.e. normal
household demand. In terms of the European Commission analysis and financial stress periods).
(European Commission, 2009), “cross border transmission was really In this paper we contribute to the financial conditions related vector
rapid, due to the tight connections within the Eurozone financial system autoregression literature by estimating an empirical Threshold VAR
itself and also the strongly integrated supply chains in global product model with exogenous variables (E-TVAR) applied in the international
markets”. It is then important to assess the effects of US financial stress context, using an endogenously determined threshold level of financial
stress. Our framework accounts for the presence of different regimes of
financial stress and produces generalized impulse responses (GIR) that
⁎ Corresponding author.
are based on the assumption that a shock in financial conditions may
E-mail addresses: anastasios.evgenidis@centralbank.ie (A. Evgenidis), lead to switch between regimes during the forecasting horizon. The
atsagkanos@upatras.gr (A. Tsagkanos). GIR derived from the E-TVAR framework allow us to take into account

http://dx.doi.org/10.1016/j.irfa.2017.03.003
1057-5219/© 2017 Elsevier Inc. All rights reserved.
70 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81

differences in the characteristics between shocks, i.e. positive versus possibility of financial instability even in conditions of low inflation
negative and small versus large shocks. Therefore, we set and discuss and growth assuming that there is a combination of supply shocks
three main questions of great interest for macro-financial purposes and asset price booms with overoptimistic assessments of risk.
and policy. The first question is to investigate in which state of the econ- Additionally, some other studies extended the research by noting
omy, a positive (detrimental) shock in the US financial conditions the importance of some other transmission channels. For instance, the
would have the largest impact. We would expect that a positive finan- transmission of financial stress via the banking channel is the subject
cial stress shock would have a more detrimental impact when the econ- under consideration in Haas and Horen (2013). They documented sub-
omy is vulnerable. Second, given that the US economy lies within a high stantial heterogeneity in the extent to which different banks retrenched
financial stress regime, we investigate possible non-linearities in the from the same country. Banks reduced credit less to markets that were
form of the shocks by examining whether large detrimental financial geographically close, more experienced and integrated into a network
shocks have a disproportionately larger impact on the economy com- of domestic co-lenders. Balakrishnan, Danninger, Elekdag, and Tytell
pared to small detrimental shocks. Third, given that the US economy (2009), emphasize on the impact of securities markets to financial stress
lies in the high financial stress regime, we investigate whether US neg- episodes, rather on banking, currency, and debt crises. Gilchrist, Yankov,
ative financial shocks have a greater impact than positive shocks. As and Zakrajsek (2009), find that credit channel is the main transmission
soon as financial stress shocks can differ in respect to their size and di- channel of financial distress. Gertler, Gilchrist, and Natalucci (2007)
rection, the last two questions help us disentangle which kind of point out the significance of exchange rate channel in the transmission
shock mostly affects macroeconomic variables. of financial distress by noting that financial accelerator effects are much
Our analysis unveils important results which have been overlooked stronger under fixed rates than under flexible rates.
by the existing studies that considered linear VAR models which as- Second, in relation to the measures of financial stress, there are a lot
sumed no regime separation and no distinction in the nature of the of studies which use a single variable to measure financial stability con-
shock. First, we find that the effect of a detrimental US financial shock ditions, such as the probability of default used by Alves (2005). However
in the macroeconomy is more pronounced when the economy lies in a measures of financial stability based on one observable variable are in-
high financial stress period rather than to a normal period. In addition, adequate since the analysis of financial stability must include various
during such turmoil periods, the financial accelerator amplifies and sources of risks and vulnerabilities. This requires a precise and efficient
propagates US detrimental financial stress shocks to the Eurozone by re- monitoring of the individual parts of the financial markets and the rela-
ducing Eurozone's economic activity. In addition, this study reveals that tionships among them (Schinasi, 2006). What is more, a suitable mea-
small financial stress shocks, rather than infrequent large ones, are able sure of financial stability should include a wide range of quantitative
to create large fluctuations in inflation rates. Last, the effect of a detri- variables that allows financial system vulnerabilities to be captured
mental shock in financial conditions has larger negative effects in the (Gadanecz & Jayaram, 2009).
economy compared with the positive effects that would be generated The use of a financial stress index that combines different quantities
by a beneficial shock in financial conditions. of variables into a single measure to proxy financial stress has been used
The rest of the paper is organized as follows. Section 2 reviews the by several studies. Hanschel and Monnin (2005) construct a composite
related literature. Section 3 provides a detailed description of the data financial stress index using four different types of information, market
sources and draws some initial insights from a preliminary analysis. prices aggregates, balance sheet data, non-public information and
Section 4 discusses the E-TVAR framework, including the modelling as- other structural data; by employing the variance equal weight method.
sumptions. Section 5 illustrates the empirical findings, together with a Illing and Liu (2006) by using the same method, compile an index that
discussion of the results. Finally, Section 6 concludes. comprises a number of variables suitable to describe a small open econ-
omy. They include a corporate bond spread, a liquidity measure and a
2. Related literature stock market volatility measure. Cardarelli et al. (2011), construct an
FSI index based on a uniform set of indicators across seventeen ad-
This paper is related to the following strands of literature. First, a vanced economies. The FSI for each country is constructed as a vari-
large number of studies have addressed the topic of the financial stress ance-weighted average of three sub-indices, which are associated
transmission in different dimensions of the economy. Accordingly, there with the banking, securities, and foreign exchange markets. Mallick
is a group of papers that investigates the effect of financial stress in the and Sousa (2013) use the same index to investigate the link between fi-
real economy. Davig and Hakkio (2010) identified significant fluctua- nancial stability and output while Balakrishnan et al. (2009), use a sim-
tions of the US economy between episodes of low/high financial stress ilar approach to construct an index reflecting market responses in
and high/low economic activity. Mallick and Sousa (2013) found that securities and exchange markets as well as the banking sector.
a contractionary monetary policy has a negative impact on financial On the logic that financial stress is only important when it is system-
stress conditions and on real output while Cardarelli, Elekdag, and Lall ic; a method that chooses weights to explain the maximum amount of
(2011) supported that financial turmoil periods characterized by bank- variation of the sub-components collectively, would be more efficient
ing distress are highly associated with severe and protracted downturns than weighting the components in the variance-equal weighting meth-
and that economies with more arms-length financial systems appear to od. This is exactly what the St. Louis Financial Stress Index (FSI) does.
be particularly vulnerable to sharp contractions. In addition, there is a This index is constructed by using principal components analysis,
debate among the researchers about the direction of the causal relation- which is a statistical method of extracting factors responsible for the
ship between financial stress and economic activity. For instance, Bloom co-movement of a group of variables. Principal components analysis as-
(2009) advocated that higher uncertainty delays firms' investment sumes that each of the variables used to construct the FSI captures some
plans and as a result, it slows down the growth of the economy. In con- aspects of financial stress. Therefore, as the level of financial stress in the
trast, Klomp and De Haan (2009) found that the causality flows in the economy changes, the variables used to construct the FSI are likely to
opposite direction and concluded that economic growth weakens finan- move together. In the case of the FSI, it is assumed that financial stress
cial stability. is the most important factor in explaining the co-movement of the eigh-
Another branch of literature discusses the link between financial in- teen variables and by extracting the first principal component one is
stability, prices and monetary stability. De Graeve, Kick, and Koetter able to create an index with a useful economic interpretation.
(2008) found that an unexpected change in monetary policy has a sig- Third, in terms of the methodological framework used in our study,
nificant impact on the probability of distress. They concluded that the VAR models have become an increasingly popular tool for investigating
balance between monetary and financial stability is the basic purpose the effects of monetary policy while only recently it has been used to ex-
of a central bank. Borio and Lowe (2002) argue that there is the amine financial stress transmission. However, the focus of the relative
 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81 71

literature is mainly on the use of standard VAR (or Structural VAR) propagation of monetary policy shocks in the macroeconomy. Consis-
models to examine the propagation of financial stress shocks. tent with the relevant literature, the use of short term rates is consid-
Therefore, the relevant research fails to capture possible non-linear- ered as a proxy for the (conventional) monetary policy stance.
ities that arise during turmoil periods, in which, the effect of financial Accordingly, Hubrich and Tetlow (2015) for the US and Boeckx,
stress shocks on macroeconomy might be different from what is usually Dossche, and Peersman (2014) for the Eurozone, use both short term
observed in normal times. For instance, Apostolakis and Papadopoulos rates and financial stress indices within a VAR framework to examine
(2015), examine US financial stress spillovers among the banking, secu- how does monetary policy reacts to financial stress shocks. The use of
rities and foreign exchange markets, while Hubrich and Tetlow (2015) capital flows is motivated by the discussion of the impact of financial
used a time-varying Panel VAR model to analyze the transmission of fi- stress episodes in advanced economies on capital flows. Together the
nancial shocks to real variables for many individual euro area econo- studies of Kaminsky and Reinhart (1999) and Moser (2003), imply
mies. In the international context, Park and Mercado (2014) use that during times of crisis, cross-border transmission of high financial
structural VAR models to investigate the interdependence between fi- stress events is often manifested in co-movements of capital flows
nancial stress indices across emerging market economies. In the same when the economies share similar fundamentals and have strong mac-
vein, Fink and Schüler (2015) analyze the international transmission roeconomic interdependence through trade and financial linkages.
of US systemic financial stress shocks to eight emerging market econo- Global capital flows had consistently increased to over 20% in 2007,
mies while, Apostolakis and Papadopoulos (2014) highlight the signifi- leading to a dramatic expansion of flows to and from advanced econo-
cant role of the US as a financial stress transmitter by using a generalized mies. When the crisis burst these flows simply evaporated, turning
VAR framework. sharply negative in late 2008 on heavy selling of foreign assets.
The literature dealing with the effects of financial stress in a non-lin- Balakrishnan et al. (2009) for instance find that banking sector stress
ear framework is relatively scarce, but growing. Balcilar, Thompson, in advanced economies led to large and enduring reductions in capital
Gupta, and van Eyden (2016) employed a nonlinear logistic smooth flows to emerging economies. Fink and Schüler (2015), show that an
transition VAR to investigate the transmission of a financial conditions adverse shock to the US financial system dries up capital flows from
shock in South Africa. However, in this framework, the upper and the US to emerging economies. Considering the very recent, high level
lower regimes are not necessarily determined by the nature of the financial stress periods in the advanced economies, a similar process
switching variable itself, but rather by the asymmetric and dynamic in- of significant declines might happen in the US or/and the Eurozone as
teractions of all the variables in the system. Davig and Hakkio (2010), a response to a detrimental shock in the US financial conditions.
Mittnik and Semmler (2013) and Hubrich and Tetlow (2015), used In addition, to capture broader global financial market conditions,
Markov-switching VAR models to investigate the interaction between the model is augmented by an exogenous block. The vector of exoge-
financial stress and basic macroeconomic indicators in the US. Contrary nous variables includes the following variables; VIX and VSTOXX indices
to our model, Markov-switching models assume that regime shifts to capture the degree of economic uncertainty in the US and the
evolve according to a Markov chain and therefore the state variable is Eurozone respectively, S&P 500 and Euro Stoxx 50 to include stock mar-
not directly observable. Moreover, these models often impose exoge- ket effects and oil and gold prices to incorporate the impact of world
nous switches. The latter assumption is quite unrealistic in a business commodity markets. In addition, we take into account agents' expecta-
cycle context where endogenous changes should be expected to result tions about future inflation by considering the Survey of Professional
in regime-switches. Forecasters (SPF) inflation expectations index from the Federal Reserve
Therefore, and as far as we can tell, there are no studies that examine Bank of Philadelphia and an inflation expectations index based on
the transmission of financial stress shocks in both the financial sector European Central Bank (ECB) SPF.
and the macroeconomy, where the propagation of financial shocks is We proxy the US financial stability conditions by using the St. Louis
extended in the international context, and within a framework which Financial Stress Index. The index comprises of interest rates (both gov-
clearly separates the economy between high and low financial stress ernment and corporate bond rates), yield spreads and other indicators.
periods, and that is precisely what we do in this paper. In the latter case, the variables are: J.P. Morgan Emerging Markets bond
index, VIX (Chicago Board options market volatility index), Merrill
3. Data description and preliminary statistics Lunch bond market volatility index, 10-year nominal Treasury yield
minus 10-year Treasury inflation protected security yield and Vanguard
Our dataset comes from three databases, Federal Reserve Economic Financials Exchange-traded fund (equities). All these series are crystal-
Database (FRED), Statistical Data Warehouse (SDW) and Datastream. lized into one value for the overall level of financial market stress
Our sample consists of monthly data for a period spanning from through the following procedure. First, each of the data series is de-
2001:01 to 2015:06. The sample period embraces the latest major pe- meaned. The de-meaned series are then divided by their respective
riods of high financial stress, i.e. the global financial crisis of 2008– sample standard deviations. With the variables now expressed in the
2009 and the European debt crisis of 2010. All variables are considered same units, principal components method is used to calculate the coef-
both for the US and the Euro area. Accordingly, the vector of the endog- ficients of the variables in the FSI. Then, the coefficients are scaled so
enous variables in our TVAR model includes the following variables. In- that the standard deviation of the index is one. Finally, each data series
dustrial production index (IP) as a measure of economic activity, is multiplied by its respective adjusted coefficient, and the observation
consumer price index (CPI) - or HICP for the Euro area - as a measure of FSI at time t is the sum of each series multiplied by its respective ad-
of prices, capital flows (CF) to investigate the impact of financial stress justed coefficient. A positive value indicates that financial stress is above
shocks on international trade and financial flows,1 and last, short term the long-run average, while a negative value signifies that financial
rates, which are, the US Federal funds rate (FFR) and the one month stress is below the long-run average.
Euribor rate as monetary policy instruments. The first two indicators, Financial stress in the Eurozone is captured by using the CISS (Com-
i.e. IP and CPI are two standard measures used in the VAR literature to posite Indicator of Systemic Stress, see Hollo, Kremer, Lo, & Duca, 2012)
investigate the transmission of various types of shock to the indicator. This index measures the current state of instability in the fi-
macroeconomy. For instance, Peersman and Smets (2001) and Kim nancial system in the Euro area as a whole or, equivalently, the level
(2001), for the Eurozone and the US correspondingly, include these of systemic stress, where systemic stress is interpreted as an ex post
two indicators in the vector of endogenous variables to examine the measure of systemic risk, i.e. the risk which has materialized already.
The CISS is designed in a way such that it both captures the idea of wide-
1
Our measure for capital flows includes foreign direct investment flows, portfolio and spread financial instability and the importance of financial stress for the
bank-related flows. real economy. The CISS includes 15 raw, mainly market-based financial
72 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81

Table 1
Descriptive statistics.
Descriptive statistics of the endogenous variables included in E-TVAR model.

usip uscpi usflows ffr fsi eurip eurcpi euroflows eurrate ciss

Mean 0.001 0.002 0.044 1.440 −0.217 0.000 0.002 0.002 1.997 0.213
Median 0.001 0.002 0.053 0.920 −0.571 0.001 0.002 0.001 2.000 0.137
Max 0.015 0.014 0.711 5.030 5.112 0.022 0.013 0.125 4.750 0.779
Min −0.044 −0.018 −0.191 0.010 −1.607 −0.043 −0.016 −0.110 0.050 0.033
Std. dev. 0.007 0.003 0.090 1.665 1.137 0.010 0.004 0.035 1.349 0.178
Skew −2.205 −1.345 1.929 0.949 1.949 −0.951 −0.562 0.507 0.333 1.363
Kurt 13.233 11.483 19.470 2.501 8.276 5.701 5.044 5.119 1.965 4.150
JB 910.518 580.778 2098.540 28.252 315.471 80.034 39.904 40.465 11.106 64.220
p-Value 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

stress measures that are split equally into five categories, namely the fi- these indices cannot be considered as a sufficient statistic for assessing
nancial intermediary sector, money markets, equity markets, bond mar- various states of economic activity.
kets and foreign exchange markets. The main methodological All variables are seasonally adjusted. In addition, both Augmented
innovation of the CISS is the application of basic portfolio theory to the Dickey Filler (ADF) and Kwiatkowski–Phillips–Schmidt–Shin (KPSS)
aggregation of five market specific sub-indices created from the total se- tests (results are not reported here for the sake of brevity) show that
ries. The aggregation method takes into account the time-varying cross- all variables are I(1) except for capital flows in US and Euro area. A
correlations between the sub-indices. As a result, the CISS puts relatively threshold VAR with trending variables cannot be estimated in levels
more weight on situations in which stress prevails in several market since stationary variables are needed in order to identify a threshold.
segments simultaneously.2 In line with this, industrial production and inflation are transformed in
Before we move to the results, we offer some preliminary analysis in log deviations from a linear trend, FSI and CISS in deviations from the
our variables. Accordingly, Table 1 reports the descriptive statistics for trend, while no transformation is implemented for interest rates and
all the endogenous variables. capital flows.
Judging from the standard deviation, the most volatile series are FSI
and the short term rates in the US and the Euro area. Moreover, the 4. Model set-up
growth rates of the main macroeconomic indicators for both regions
(industrial productions and inflation) are negatively skewed while all 4.1. Threshold VAR with exogenous variables
other growth rates are positively skewed. The skewness coefficient is
not close to zero for any of the variables which indicates that the distri- Assuming a no regime framework, the standard VAR (p) model of
butions are not-symmetric. The Jarque-Bera (JB) test statistic clearly in- order p is given by:
dicates a rejection of the null hypothesis at the 5% level, that is, all series
are not normally distributed. Y t ¼ AY t þ ΦðLÞY t−1 þ εt ð1Þ
Figs. 1 and 2 depict the FSI and CISS together with NBER (for the US)
and OECD (for the Euro area) recession periods. Recession periods are where Yt is a vector of endogenous variable, εt is the vector of structural
represented by the green lines. Both indices show signs of periods errors, Φ(L) is the lag polynomial matrix and AYt represents the con-
with moderate fluctuations, combined with shorter periods of high temporaneous terms.
levels and volatility. In this paper, we investigate the effects of shocks in US financial
In the US, FSI fails to capture accurately the aftermath of the dotcom stress conditions by accounting for the presence of differing financial
bubble in 2001 as indicated by the NBER dating (in green). However, it stress periods. To evaluate the role of financial stress shocks on the
is clear that the period which can be seen as a high financial stress (in- macroeconomy, we rely on a structural vector threshold autoregressive
dicated by the very large values of FSI), is associated with the post-2007 model. In addition, we extend this framework by adding a vector of ex-
financial crisis period. Similarly, this is exactly the case when we look at ogenous variables. Therefore, the threshold VAR model with exogenous
the CISS for the Euro area. There are two periods of what the unaided variables (E-TVAR) can be expressed using the following equation:
eye sees as high stress, the one associated with the recent financial crisis
that begun in the US and the second is linked with the more recent Y t ¼ ½A1 Y t þ Φ1 ðLÞY t−1 þ Γ1 ðLÞΧt Iðyt−d ≤γÞ
Eurozone debt crisis. Again, both recessions as noted by the OECD indi- þ ½A2 Y t þ Φ2 ðLÞY t−1 þ Γ2 ðLÞΧt Iðyt−d NγÞ þ εt ð2Þ
cator are associated with financial stress periods. An important thing to
notice here is that the 2010 European debt crisis generated much dis- yt−d is the threshold variable and d is the delay parameter which typi-
cussion and a great deal of activity at the US Treasury; however there cally denotes the delay of the threshold variable in determining the re-
was barely any movement in FSI. This means that every headline eco- gimes. We follow the literature (Afonso, Baxa, & Slavik, 2011; Calza &
nomic event has not manifested itself in high stress, thus the level of Sousa, 2006; Van Robays, 2012) and set d equal to one. I(.) is an indica-
tor function that takes the value one if the threshold variable is higher
than the threshold value γ and zero otherwise. Φ1(L) and Φ2(L) are
2
A technical issue that might arise here is the potential correlation between the vari- lag polynomial matrices and A1Yt , A2Yt represent the contemporaneous
ables that constitute each financial stress index and the variables used in the VAR models terms since contemporaneous effects might also differ across regimes.
(especially interest rates) which could potentially affect the results. However, the results
Our framework is expanded to account for a multicountry setting. Ac-
of our analysis are only due to the actual impact of financial conditions and not due to this
technical issue. This is because the construction of both financial stress indices comprises cordingly, Yt includes two blocks, the first block YUS contains the US var-
many variables, mostly market based financial stress measures, which are crystallized into iables and the second block Yeur contains the eurozone variables. The lag
one value for the overall level of systemic stress through either principal components anal- length of the endogenous variables is determined by the Bayesian infor-
ysis (FSI) or basic portfolio theory (CISS). Therefore the link with the individual, underly- mation criterion (BIC) and is equal to one. Beyond that, threshold VAR
ing variables is not so strong. To verify the weak correlation links, we check the correlation
values between both financial stress indicators and the other variables included in the vec-
studies usually characterized by the fact that a very low number of ob-
tor of endogenous variables. The correlations verify that there is no strong link between FSI servations exists in the high stress regime; therefore it seems that the
(or CISS) with any of the other variables in the system. choice of one lag (or maximum two) is the only plausible option
 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81 73

Fig. 1. FSI and recession dates. This figure plots the St. Louis Financial Stress Index (FSI, in blue) within NBER recession periods (in green).

(Afonso et al., 2011; Schmidt, 2013). Following the threshold VAR liter- unknown threshold is not identified under the null. Therefore, the null
ature (Afonso et al., 2011; Calza & Sousa, 2006), specification (2) is iden- hypothesis of no breaks against the alternative of a single breakpoint
tified by a recursive ordering of the variables based on the Cholesky is implemented by a sup-F test. The distribution of this statistic is non-
decomposition. Accordingly, the ordering of the variables is as follows: standard, but Bai and Perron (2003b) provide critical values for various
trimming parameters, numbers of regressors, and numbers of breaks.
Y t ¼ ½Y eur ; Y US 
where YUS ¼ ½IPUS ; CPI US ; FFR; CF US ; FSI  and Yeur ¼ ½IPeur ; CPI eur ; rateeur ; CFleur ; CISS:
4.2. Generalized impulse responses
The ordering of output and prices before the policy rate is conven-
tional and in line with the majority of the VAR literature in the monetary In the linear model, the impulse responses can be derived easily
transmission (Bernanke, Boivin, & Eliasz, 2005; Kim, 2001). It implies from the Wold decomposition. The estimated responses are constant
that monetary transmission responds contemporaneously to changes over time and they are symmetric not only in the sign, but also in the
in macroeconomic conditions. The financial stress index is ordered last magnitude of the structural shocks. In addition, linear responses are
by following the relative literature (Fink & Schüler, 2015; Hubrich & not being suitable to assess the macroeconomic effect of a shock if the
Tetlow, 2015). This identification scheme of the domestic variables as- likelihood of regime switching over the duration of the response is
sumes that the economy as a whole does not react contemporaneously non-negligible. In contrast, these assumptions do not hold within non-
to sudden changes in the financial conditions. The ordering of the vector linear models such as the TVAR model (Koop, Pesaran, & Potter,
Yt implies that the Eurozone does not react to a US financial stress epi- 1996). For this reason Koop et al. (1996) consider GIR which rely on
sodes within the month. the assumption that some shocks may lead to switches between re-
In our model, the threshold value is not known a priori, thus we de- gimes over the forecasting horizon. For instance, in our case it is highly
termine the threshold endogenously by a grid search over the possible probable that a large deterioration in the financial conditions may lead
values of γ. The trimming parameter, which indicates the minimal per- the economy to switching away from its starting regime once its effect
centage of observations in each regime grid, is chosen at 15%. From the pass through and, over time, responses may switch many times be-
grid, the estimated value of γ corresponds to the model with the tween the two regimes.
smallest determinant of the covariance matrix, Ωετ of the estimated re- The approach relies on the simulation of data depending on the re-
siduals of (2): gime the system is at the point of time that the shock hits the economy.
Except the fact that GIR allow for regime dependent responses, they also
γ ¼ argmin logjΩετ ðγ Þj ð3Þ allow us to study the effects of shocks of different magnitudes and direc-
tions. Therefore, GIR is an excellent framework to investigate the effect
We test whether the chosen threshold variable is meaningful by of an FSI shock under three different dimensions, the degree of interde-
employing Bai and Perron's (1998, 2003) tests for each equation in the pendence between the two regimes, positive versus negative and large
VAR system because this test is designed for the univariate regressions. versus small, financial shocks. Formally, GIR are defined as:
If the threshold values γ were known, the conventional F-test for the
null hypothesis A2 = Φ2(L) = 0 would give reliable results. However, ac-
cording to Hansen (1996), standard inference cannot be applied as the GIRy ðk; Ωt−1 ; ut Þ ¼ E½Y tþk jΩt−1 ; ut −E½Y tþk jΩt−1  ð4Þ

Fig. 2. CISS and recession dates. This figure plots the Composite Indicator of Systemic Stress (CISS, in red) within the OECD based recession indicator for the Euro area (green).
74 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81

where Yt+k is the vector that contains the responses of variables at ho- tighter credit standards that were caused by the positive FSI shock
rizon k, Ωt−1 is the information set available up to the time of the shock t (Apostolakis & Papadopoulos, 2015). Similarly, the policy rate declines
and ut is the vector of errors from a standard linear VAR model. This for- in the first two quarters, but afterwards it shows a positive response.
mulation indicates that GIR are defined as the difference between the The immediate decline in the rates might indicate that the FED reacts
forecasted path of variables with and without a shock to the variable by putting an obstacle in the decline of the economy as a result of the
of interest. It also implies that the responses depend on the initial con- negative effects of the FSI shock. On the other hand, this monetary loos-
ditions and that no restriction is imposed in the symmetry of the shocks. ening may result in boosting economic growth and inflation. The posi-
GIR are simulated for various sizes and signs of the shock. The simu- tive response of the policy rate which is observed after the first two
lation is performed by assuming that at the time of the shock the model quarters implies that policymakers may increase the rate in the mid-
is either in the high or in the low regime. The algorithm used for the run so that they moderate the boost in economy. In assessing the trans-
simulations is as follows. First (i), a sequence of initial values of the ac- mission of the financial shock from the US to the Eurozone and judging
tual and contiguous lagged values of the endogenous variables Y is cho- from the performance of the industrial production index, we find a sig-
sen, corresponding to a particular history Ωt−1 falling under one of the nificant negative impact on the Eurozone's growth. On the other hand,
two regimes. Then (ii), a random sample of shocks et+k is drawn based while inflation increases immediately after the shock, it very quickly
on the variance-covariance matrix of the residuals of the estimated VAR takes negative values. This means that a US worsening in financial con-
model. For each sequence of shocks, (iii) a path of the variables of inter- ditions would push headline inflation in the Eurozone lower. Given that
est is simulated using the estimated coefficients for both regimes as well Eurozone exhibits currently very low rates of inflation, a fresh US finan-
as the shock process for k + 1 periods. Hence the model is allowed to cial stress shock might push the euro area into deflation. Except the
change regimes over the forecast horizon. The resulting sequence is de- negative effect in prices and economic activity, financial stability in
noted by Yt+k |et+k ,Ωt−1. In the next step (iv), conditional in the same the Eurozone is equally affected. In particular, increases of the systemic
series of random shocks, an extra one standard deviation shock ut is stress indicator imply that financial conditions in the region worsen.
added at time t. The resulting sequence provides another estimate for This finding highlights the importance of stress co-movements which
Yt+k |ut ,et+k ,Ωt−1. Steps (ii) to (iv) are repeated K times so that we av- may occur between US and Eurozone during crucial stressful events;
erage out the shocks. We set the number of simulation M to be 500 as in due to the amplified interconnection of their financial markets especial-
Balke (2000). Steps (i) to (iv) are repeated N times to obtain an average ly during times of increased uncertainty.
over the respective regime history. Finally, the GIF is the difference be- A very interesting finding is the effect of the FSI shock on the capital
tween the two simulated forecasts. Finally, we follow Zheng (2013) flows of the Eurozone. In particular, while capital flows in the US do not
and Schmidt (2013) in computing confidence bands by bootstrapping seem to be affected significantly, a detrimental financial conditions
the E-TVAR residuals. shock in the US would temporarily harm the Eurozone, since it would
suffer big net capital outflows in the next quarter. This finding is consis-
5. Discussion of the results tent with Lane (2013) who finds that, after the extraordinary boom-
bust cycles in both gross flows and net flows since 2003, the recent crisis
5.1. Transmission of financial stress in a non-regime framework lead euro area member countries exhibit a reversal in capital flows. In
the same spirit, Fratzscher (2012) find that euro area countries recorded
We first report the responses for a standard linear VAR. Fig. 3 depicts the strongest net outflows, particularly in bonds, throughout the crisis.
the results of a positive (detrimental) FSI shock to the US economy and Overall, we interpret these results as first important evidence that a det-
the euro area. rimental US financial shock leads to a worsening of economic and finan-
As expected, economic activity and prices decline gradually up to cial conditions both domestically and internationally.
four quarters ahead. This is in accordance with Hakkio and Keeton Standard VAR frameworks are not able to capture the fact that the
(2009) who show that during financial stress events the cost of financ- relationship between macroeconomic and financial variables may be
ing new debt is increased which in turn lowers economic activity. The subject to non-linearities. This is particularly evident during times of
decline in prices is due to persistent deflationary pressures. These defla- high financial stress and during financial crises. During those times,
tionary pressures are the result a strong negative demand shock due to a the transmission of financial shocks is different during high stress pe-
combination of increased uncertainty and financing costs as well as riods compared with normal times. This means that financial shock

Fig. 3. Impulse responses-standard VAR model. This figure shows the impulse responses to a positive (detrimental) FSI shock to the US economy. All responses are related to one standard
deviation shock. Black line represents the median while the dashed lines represent 95% confidence interval.
 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81 75

Table 2 One can see that there are threshold effects in eight out of ten equa-
Non-linearity tests. tions. This means that FSI is an appropriate threshold variable and that it
This table depicts a Sup-F test (Bai & Perron, 1998, 2003a, 2003b) for non-linearity effects
in each equation of the E-TVAR model. Threshold variable is FSI. Under the null hypothesis
is meaningful to separate into two regimes that are being characterized
there is no structural break. The critical values of Bai-Perron test reported at 1% signifi- by statistically different dynamics. We estimate the threshold endoge-
cance level. nously by implementing a grid search over the possible values of the
switching variable. We follow Andrews (1993) by setting the trimming
Equation F-stat Critical value
factor to 15%. We identify two regimes which correspond roughly to re-
fsi 23.34 8.02
gimes characterized by a slowdown of the economy (high regime or fi-
ciss 7.96 7.04
usflows 7.01 6.72 nancial market stress period) and by regimes characterized by
euroflows 7.74 7.04 economic growth (low regime or normal periods). Accordingly, regime
uscpi 2.46 7.04 one is the regime in which FSI index is low while regime two indicates
uscpi 9.07 8.02 the state of the economy in which FSI is high.
usip 27.37 8.02
eurip 4.1 7.04
As a preliminary exercise, we check the threshold behavior to verify
ffr 8.13 6.72 whether movements in the variables of our system are synchronized
eurorate 0.66 7.04 with, especially, high regimes. Fig. 4 plots the data of our chosen vari-
ables together with the two distinguishable regimes. The threshold is
estimated to be 0.29, which splits the sample into low and high financial
may have a limited role in normal times, but have significant effects in stress periods that represent 13 and 87% of all observations respectively.
episodes of high financial stress. In addition, different shocks in the fi- Shaded regions correspond to the second regime of high financial stress.
nancial conditions will affect various economic measures dispropor- Such division seems to be well in line with the fact that in our sample
tionately. Therefore, we expect that financial shocks might differ with the duration of expansions is higher than the duration of recessions.
regards to the size, direction and differences in initial conditions. The very low number of observations in the financial stress regime is a
common characteristic in most of threshold VAR studies which separate
5.2. Transmission of financial stress under different states of the economy between high and low uncertainty (stress) periods (see for instance Van
Robays, 2012 who examines the impact of oil price shocks and Afonso et
Our E-TVAR model allows us to take into account non-linearities in al., 2011 who examine the impact of fiscal policy, on different states of
the form of threshold effects. FSI is used as a threshold variable to distin- the economy).
guish between two different states of the economy, “bad” times when The financial stress regime is evident after the onset of the 2007 fi-
there is increased financial stress and “good” times characterized by nor- nancial crisis, either in greater magnitude (wide shaded areas, from
mal financial conditions. The Bai and Perron (1998, 2003a, 2003b) Sup-F the mid-of 2008 to September 2009 and from October 2011 up to Au-
test is used to test non-linearity for each equation of the E-TVAR system. gust 2012) or to a smaller magnitude (short shaded areas, mid-of
The null hypothesis is that there is no structural break against the alter- 2013 and first quarter of 2015). It is worth noting that the shaded area
native that there is one structural break. Table 2 displays the results. which corresponds to the 2008–09 period is associated with all the

Fig. 4. Time series -FSI threshold. This figure plots the data of all variables within the two different financial stress regimes. Shaded regions correspond to the second regime, i.e. that of high
financial stress.
76 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81

Fig. 5. Switching probabilities from regime 1 to regime 2. This figure displays the probability of regime-switching, from regime one (low financial stress) to regime two (high financial
stress), in response to all types of shock. The empirical probability of regime-switching is derived from the number of times the FSI index crosses its threshold value and it is plotted
against the forecast horizon (in months which are displayed on the x-axis) at which the regime-switching occurs.

variables in the US and euro area. This means that the large drop in the Next, we compute standard, linear impulse responses (IR, not to be
real activity, inflation and interest rates in both regions during this peri- confused with non- linear GIR examined later) by assuming that the
od, coincide with the financial stress regime. Similarly, besides the triv- economy stays within the respective regime which was in place when
ial case in which high levels of FSI are associated with the second the shock initially hit, i.e. shocks cannot cause regime changes in the
regime, high values of the composite indicator of systemic stress index forecasting horizon. The reason for this analysis is to investigate wheth-
in the euro area consistently fall within the wide shaded areas. er a shock to the system generates movements in the threshold variable,

Fig. 6. Switching probabilities from regime 2 to regime 1. This figure displays the probability of regime-switching, from regime two (high financial stress) to regime one (low financial
stress), in response to all types of shock. The empirical probability of regime-switching is derived from the number of times the FSI index crosses its threshold value and it is plotted
against the forecast horizon (in months which are displayed on the x-axis) at which the regime-switching occurs.
 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81 77

Fig. 7. Regime one - Large positive shock. This figure shows the generalized impulse responses of a large positive (detrimental) US financial shock when the economy lies in the first (low
financial stress) regime. All responses are related to one standard deviation shock. Blue line represents the median while the dashed lines represent 95% confidence interval.

and what type of reduced-form innovation might lead to regime- but the probability is very low. The transition between regimes is more
switching. We consider four different cases which represent different intense from stress periods to normal periods (Fig. 6). For the US explic-
size and directions of the shock. Accordingly, we examine the effects itly, the results indicate that a negative financial stress shock in the US, a
of a relative small (0.5 standard deviation) and a fairly large (1.5 stan- boost in economic activity, a monetary loosening (interest rate de-
dard deviation) shock as well as the effects of a positive and a negative crease) and an increase in inflation rate might push the US economy
shock. Figs. 5, 6 display the probability of regime-switches for the in more normal states. In particular, the latter finding might be consis-
threshold variable. Following a shock to each of the variables, IR of the tent with the FED's aim to increase inflation to deal with anemic eco-
threshold variable is calculated for each data point in the initial regime. nomic growth. On the other hand, in the Eurozone, a negative interest
The probability of regime-switching is derived from the number of rate shock is the only shock that might lead to a regime switch from a
times the switching variable crosses its threshold and is plotted against financial stress period to a more stable environment. This finding
the forecast horizon at which the regime-switching occurs. The choice might be consistent with ECB's current monetary policy stance to keep
of the ten months ahead forecasts is long enough to describe any signif- interest rates very low, even in negative levels, to boost inflation in
icant movements between the regimes after the shock hits the the euro area.
economy. Overall, together the results from Figs. 5, 6 show evidence in favor of
Fig. 5 shows the probability of regime switching from the normal pe- a strong channel through which US financial stress shocks can affect
riod (regime one) to the stress period (regime two) and Fig. 6 depicts economic and financial activity and cause regime switches either from
the probability of regime switching from the stress period to the normal stress periods to normal ones or from normal ones to stress periods.
one. Fig. 5 shows that a detrimental shock in financial conditions might This empirical finding is a first evidence of non-linearity in financial
lead the US economy from normal periods to stressed periods with a stress transmission which is dependent on different states of the econ-
probability close to 40%. Shocks in economic activity, inflation and inter- omy. To get the complete picture about the dynamics of the model,
est rates might also push the economy to a more unstable environment, we estimate GIR for various signs and sizes of the shock. Figs. 7 to 11

Fig. 8. Regime two - Large positive shock. This figure shows the generalized impulse responses of a large positive (detrimental) US financial shock when the economy lies in the second
(high financial stress) regime. All responses are related to one standard deviation shock. Dotted line represents the median while the dashed lines represent 95% confidence interval.
78 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81

Fig. 9. Regime two - Small positive shock. This figure shows the generalized impulse responses of a small, positive (detrimental), US financial shock when the economy lies in the second
(high financial stress) regime. All responses are related to one standard deviation shock. Dotted line represents the median while the dashed lines represent 95% confidence interval.

report the median response (along with the 95% confidence bands) of a a disproportionately larger impact on the economy compared with
financial stress shock, both for a high and for a low financial stress re- small detrimental shocks. This second question is relevant to the discus-
gime and for different type of shocks. In order to make the results com- sion about whether economic fluctuations are best described by an ac-
parable across the different types of shocks, GIR were all scaled to cumulation of small shocks or they are due to large shocks. Third,
correspond to equally sized positive financial stress shocks. In particular, given that the economy is in a financial stress period, we analyze possi-
the responses from negative shocks were multiplied by −1. In addition, ble asymmetries on the transmission of the shock. The reason behind
responses from a small, 0.5 standard deviation shock were multiplied by this is to uncover whether US negative financial shocks have a greater
2 while the responses from a large, 1.5 standard deviation shock were impact than positive shocks.
divided by 1.5 to achieve comparable GIR corresponding to 1 standard The first hypothesis is examined by looking at Figs. 7 and 8, which
deviation shocks. show the effect of a positive US financial shock when the economy lies
We analyze three hypotheses in the form of policy questions. First, in the first and second regime correspondingly.
we examine in which of the two regimes, does a positive (or detrimen- The sign of the responses is common in both regimes and it remains
tal) large shock in the US financial conditions has the largest impact almost unchanged compared to the standard VAR, i.e. negative effect in
both domestically and internationally? We would expect that, in “bad” the macroeconomic conditions in both regions, positive response of FFR,
states of the economy, during which the financial system is not operat- obvious effects towards capital outflows in the Eurozone and a deterio-
ing normally, the effect of adverse financial shocks would be amplified. ration of the financial stress index in the Eurozone, particularly in the
Second, given that the US economy lies within the financial stress re- short term. Next we disentangle the asymmetric transmission of the
gime, we investigate whether large detrimental financial shocks have shock between the regimes. Fig. 7 shows that the responses of all

Fig. 10. Regime two - Large negative shock. This figure shows the generalized impulse responses of a large, negative (beneficial), US financial shock when the economy lies in the second
(high financial stress) regime. For comparison reasons, responses are inverted to depict a positive one standard deviation shock. Dotted line represents the median while the dashed lines
represent 95% confidence interval.
 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81 79

Fig. 11. Regime two - Small negative shock. This figure shows the generalized impulse responses of a small, negative (beneficial) US financial shock when the economy lies in the second
(high financial stress) regime. For comparison reasons, responses are inverted to depict a positive one standard deviation shock. Dotted line represents the median while the dashed lines
represent 95% confidence interval.

variables are insignificant when the economy lies in the good regime. We examine the second policy question by looking at the Figs. 8, 9;
On the contrary, as Fig. 8 denotes, a positive FSI shock has a more pro- which depict the responses of various variables to a positive large and
nounced effect on the basic US macroeconomic variables. In particular, small shock in the second regime.
CPI is significantly affected in the first seven months after the shock We observe that, when a US detrimental financial shock hits the
while the reduction of IP is only slightly insignificant for the first five economy during periods of financial uncertainty, there are significant
months after the shock. In addition, the magnitude of the responses differences in the magnitude of the effect between the responses of
for IP, CPI and FFR - even if the response of the latter is insignificant- is large and small shocks, particularly for inflation and interest rates. In
much bigger when a positive shock hits the economy in the high finan- particular, the response of FFR to a positive shock (Fig. 8) is insignificant
cial stress regime (Fig. 8) compared to the normal state (Fig. 7). Al- while the response of CPI is significant in the first four months after the
though we should interpret the results with caution given the shock. On the other hand, the responses of the same variables under a
insignificance of some responses, there is an indication that financial small shock (Fig. 9) are not only significant but also bigger in magnitude
factors tend to be episodic in nature. The stronger response of the IP compared to the large shock. The same pattern arises when we look at
and CPI in the bad regime, could be attributed to the financial accelera- the international context. In particular, judging from the magnitude of
tor (Bernanke & Gertler, 1989; Bernanke, Gertler, & Gilchrist, 1999), the responses in Fig. 9 compared to Fig. 8, small US financial stability
stating that adverse shocks to the economy may be amplified by wors- shocks have a greater impact on inflation and interest rates than large
ening financial market conditions due to credit market distortions. The shocks. We interpret the results as indication of more powerful effects,
stronger response of the core macroeconomic variables and the mone- which might operate via small financial stress shocks rather large
tary policy tool in the distressed regime compared to the normal regime ones. In particular, these results are related to the relative debate in
is an important finding that has been overlooked by previous studies macroeconomics on whether economic fluctuations are due to the accu-
(Cardarelli et al., 2011; Gilchrist et al., 2009). These studies considered mulation of small shocks or due to infrequent large shocks (see for ex-
no regime models and therefore they were unable to uncover the het- ample Bernanke, Gertler, & Gilchrist, 1996 and more recently Gonzalo
erogeneous transmission of a financial stress shock dependent on differ- & Martinez, 2006). In our context, a positive small financial stress
ent states of the economy. shock can be viewed as a decline in liquidity which impairs the mecha-
In the international context, one can notice that impulse re- nism of exchange by raising the cost at which households and firms are
sponses in the normal regime are all insignificant. On the contrary, able to obtain liquidity services from the financial markets.
in the stress regime, economic activity and interest rates (from Our findings indicate that such a small shock can lead to large infla-
fourth to eighth month) are significantly different from zero. In addi- tion changes; which is in line with Jaccard (2013) finding, according to
tion, their impact is more exaggerated and prolonged compared to which, small liquidity shocks are able to create large fluctuations in in-
the normal regime. In particular, economic activity is reduced flation rates. Overall, this result highlights the benefits of constructing
while the interest rate after an initial insignificant increase, it very GIR within our E-TVAR model and points out the limitations of using lin-
quickly starts to decline. This drop in the short term rate might be ear VAR models in examining the financial stress transmission mecha-
interpreted in the context of a loosening monetary policy that ECB nism. In particular, in contrast with the existing literature which
might have to follow in order to deal with the negative effects ap- assumed no differentiation on the size of the financial stress shock,
peared, due to the transmission of financial stress episodes generat- our assumption of dissimilar effects on the economy dependent on the
ed in the US. Overall, these results indicate that financial stress size of the shock, allow us to bring into light the power of small financial
disturbances in the Eurozone are negligible during calm periods stress shocks to create large inflationary pressures.
characterized by decreased uncertainty. In contrast, during turmoil Next, we investigate our third policy question by examining dif-
periods, the reduction in the eurozone economic activity and the ferences in the responses between large (and small), positive and
negative interest rate effect provide an indication that the financial negative shocks assuming that the economy is in the financial stress
accelerator amplifies and propagates US shocks in the Eurozone. regime.
80 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81

By looking at large, positive and negative shocks (Figs. 8, 10 corre- Second, are macroeconomic fluctuations best described by the accumu-
spondingly), we find no noticeable change in the magnitude of the ef- lation of small financial stress shocks or by infrequent large shocks?
fect in any of the macroeconomic and financial variables for both Third, given that the US economy is in a high financial stress regime,
regions (note that the negative shock has been inverted for compari- what are the effects on the economies of negative versus positive finan-
son). However, for some of the variables, the transmission of small US cial stress shocks?
financial shocks varies significantly between positive and negative Accordingly, the lessons to be learned from our analysis are the fol-
shocks (Figs. 9, 11 correspondingly). This last finding highlights for lowing. Especially for the US, a loosening monetary policy, an increase
once again the benefits of using the dynamic structure of our E-TVAR in economic growth, a beneficial shock in financial conditions and an in-
model instead of standard linear VARs, which do not account for the crease in inflation rate would help the US economy to move out of in-
asymmetric effects due to the sign of the financial stress shock. In partic- tense financial stress periods. In addition, the effect of a detrimental
ular, when a positive shock is considered (Fig. 9), the responses of CPI US financial shock is more pronounced when the economy lies in a
and interest rates for both the US and the Eurozone are most of the high financial stress period rather than to a normal period. This result
time significantly different from zero. On the contrary, Fig. 11 shows could be attributed to the financial accelerator under which adverse
that, in response to a negative shock, most of the impulse responses shocks to the economy may be amplified by worsening financial market
are insignificant. Therefore, in a state of high financial stress, the effect conditions. More importantly, during those turmoil periods, the finan-
of a financial-increasing shock has larger negative effects in the econo- cial accelerator amplifies and propagates the US financial stress shocks
my compared with the beneficial effects generated by a financial-stress to the Eurozone by reducing Eurozone's economic activity.
reducing shock. In addition, we find that financial stress–increasing shocks have a
bigger effect than financial stress-reducing shocks. This finding implies
5.3. Robustness checks that if policymakers worry about the impact of financial stress episodes
in the economy, they will have to respond more strongly to positive fi-
We estimate several alternative specifications in order to check the nancial stress shocks. Another particularly interesting finding from a
robustness of our results. First, we estimate the E-TVAR model by policymaking perspective is that, part of the fluctuations observed in
changing the ordering of the variables. Accordingly, we order the FED economic activity in both regions, are due to the accumulation of
monetary policy rate first in the vector of endogenous variables. This small financial stress shocks. This implies that small, unimportant dis-
is because, in the post-crisis period, policy rate is unlikely to respond turbances, which can be viewed in isolation, could generate data that
to any of macroeconomic or/and financial change contemporaneously, mimic the behavior of the macroeconomic series. In particular, our find-
so it might be considered as the most exogenous variable. In addition, ings show that large fluctuations in inflation can arise from what appear
we examine another specification in which the US monetary policy in- to be relatively small financial stress shocks, such as adverse liquidity
strument is ordered only after output and inflation. This ordering fol- shocks. This result should be taken into account by policymakers
lows a more traditional ordering according to which monetary policy when they are investigating what might be affecting inflation prices.
responds contemporaneously only to output and inflation (Taylor,
1993). Second, we estimate the model by measuring financial stability
Acknowledgements
in the US, using an alternative financial stress index for the US. Accord-
ingly, we measure financial stability by using the Kansas City stress
We would like to thank two anonymous referees for their insightful
index (KCFSI). The KCFSI combines eleven variables that provide a
comments and suggestions.
range of economic signals of financial stress. These explanatory vari-
ables fall into two categories: average yield spreads and measures
References
based on the actual or expected behavior of asset prices. A positive
value indicates that financial stress is above the long-run average, Alves, I. (2005). Sectoral fragility: Factors and dynamics, in “Investigating the relationship be-
while a negative value signifies that financial stress is below the long- tween the financial and real economy”. 22. (pp. 450–480). Bank for International Set-
tlements, 450–480.
run average. Third, we change estimate the model with two lags and
Afonso, A., Baxa, J., & Slavik, M. (2011). Fiscal developments and financial stress: A threshold
fourth, we adopt IP as a different threshold variable. The main conclu- VAR analysis. (ECB Working Paper Series No. 1319. ECB, Frankfurt am Main).
sions of this analysis are robust to all cases considered above. Andrews, D. W. K. (1993). Tests for parameter instability and structural change with un-
known change point. Econometrica, 61(4), 821–856.
Apostolakis, G., & Papadopoulos, A. P. (2015). Financial stress spillovers across the bank-
6. Conclusion ing, securities and foreign exchange markets. Journal of Financial Stability, 19, 1–21.
Apostolakis, G., & Papadopoulos, A. P. (2014). Financial stress spillovers in advanced econ-
During periods of economic downturns, the quality of the assets of omies. Journal of International Financial Markets Institutions and Money, 32, 128–149.
Bai, J., & Perron, P. (1998). Estimating and testing linear models with multiple structural
financial institutions deteriorates mainly due to the increase in the changes. Econometrica, 66(1), 47–78.
number of non-performing loans and because economic sentiment in fi- Bai, J., & Perron, P. (2003a). Computation and analysis of multiple structural change
nancial markets reduces the value of other financial assets. Thus, the ef- models. Journal of Applied Econometrics, 18(1), 1–22.
Bai, J., & Perron, P. (2003b). Critical values for multiple structural change tests. The
fect of financial stress shocks on financial sector and the macroeconomy Econometrics Journal, 6(1), 72–78.
might be different from what is usually observed in normal times. Balakrishnan, R., Danninger, S., Elekdag, S., & Tytell, I. (2009). The transmission of financial
This study examines how a deterioration in US financial conditions stress from advanced to emerging economies. World economic outlook (April 2009).
Washington, DC: International Monetary Fund.
affects the US economy, whether the Eurozone is affected by a US Balcilar, M., Thompson, K., Gupta, R., & van Eyden, R. (2016). Testing the Asymmetric Ef-
shock in financial stability and if there is an asymmetric transmission fects of Financial Conditions in South Africa: A Nonlinear Vector Autoregression Ap-
of the shock according to different states of the economy (i.e. normal proach. Journal of International Financial Markets, Institutions and Money, 43, 30–43.
Balke, N. S. (2000). Credit and economic activity: Credit regimes and nonlinear propaga-
and financial stress periods). We address these issues by deriving an tion of shocks. The Review of Economics and Statistics, 82(2), 344–349.
empirical threshold VAR model with exogenous variables (E-TVAR), Bernanke, B., & Gertler, M. (1989). Agency costs, net worth, and business fluctuations.
which is applied in the international context and which accounts for American Economic Review, 79(1), 14–31.
Bernanke, B. S., Gertler, M., & Gilchrist, S. (1996). The financial accelerator and the flight to
the presence of different regimes of financial stress periods.
quality. The Review of Economics and Statistics, 78(1), 1–15.
Within this framework, we investigate and discuss some issues of Bernanke, B. S., Gertler, M., & Gilchrist, S. (1999). The financial accelerator in a quantita-
particular interest for macro-financial and monetary policy purposes, tive business cycle framework. Handbook of Macroeconomics, 1, 1341–1393.
in the form of three policy questions. First, we examine in which state Bernanke, B. S., Boivin, J., & Eliasz, P. (2005). Measuring the effects of monetary policy: A
factor-augmented vector autoregressive (FAVAR) approach. The Quarterly Journal of
of the economy, does a detrimental in shock in the US financial condi- Economics, 120(1), 387–422.
tions have the largest impact both domestically and internationally? Bloom, N. (2009). The impact of uncertainty shocks. Econometrica, 77, 623–685.
 A. Evgenidis, A. Tsagkanos / International Review of Financial Analysis 51 (2017) 69–81 81

Boeckx, J., Dossche, M., & Peersman, G. (2014). Effectiveness and transmission of the ECB's Hubrich, K., & Tetlow, R. J. (2015). Financial stress and economic dynamics: The transmis-
balance sheet policies. CESifo Working Paper No. 4907. sion of crises. Journal of Monetary Economics, 70, 100–115.
Borio, C., & Lowe, P. (2002). Asset prices, financial and monetary stability: Exploring the Hollo, D., Kremer, M., Lo, M., & Duca, M. (2012). CISS – A composite indicator of systemic
nexus. Bank for International Settlements (Working Paper No. 114). stress in the financial system. European Central Bank working paper series no. 1426.
Calza, A., & Sousa, L. (2006). Output and inflation responses to credit shocks: Are there Illing, M., & Liu, Y. (2006). Measuring financial stress in a developed country: An applica-
threshold effects in the Euro area? Studies in nonlinear dynamics & econometrics. tion to Canada. Journal of Financial Stability, 2, 243–265.
Vol. 10. (pp. 1–21) (May). Jaccard, I. (2013). Liquidity constraints, risk premia, and the macroeconomic effects of liquid-
Cardarelli, R., Elekdag, S., & Lall, S. (2011). Financial stress and economic contractions. ity shocks, working paper series 1525. European Central Bank.
Journal of Financial Stability, 7, 78–97. Kaminsky, G., & Reinhart, C. M. (1999). The Twin Crises: The Causes of Banking and Bal-
Davig, T., & Hakkio, C. (2010). What is the effect of financial stress on economic activity? ance-of-Payments Problems. American Economic Review, 89(3), 473–500.
Kansas City fed economic review second quarter (pp. 35–62). Kim, S. (2001). International transmission of U.S. monetary policy shocks: Evidence from
De Graeve, F., Kick, T., & Koetter, M. (2008). Monetary policy and financial (in)stabil- VAR's. Journal of Monetary Economics, 48(2), 339–372.
ity: An integrated micro-macro approach. Journal of Financial Stability, 4, Klomp, J., & De Haan, J. (2009). Central bank independence and financial instability.
205–231. Journal of Financial Stability, 5, 321–338.
European Commission (2009). Economic Crisis in Europe: Causes, Consequences and Re- Koop, G., Pesaran, M. H., & Potter, S. M. (1996). Impulse response analysis in nonlinear
sponses. European Economy, 7. multivariate models. Journal of Econometrics, 74, 119–147.
Fink, F., & Schüler, Y. S. (2015). The transmission of US systemic financial stress: Evidence Lane, P. R. (2013). Capital flows in the Euro area. (CEPR discussion papers 9493).
for emerging market economies. Journal of International Money and Finance, 55(C), Mallick, S. K., & Sousa, R. M. (2013). The real effects of financial stress in the Eurozone.
6–26. International Review of Financial Analysis, 30, 1–17.
Fratzscher, M. (2012). Capital flows, push versus pull factors and the global financial cri- Mittnik, S., & Semmler, W. (2013). The real consequences of financial stress. Journal of
sis. Journal of International Economics, 88(2), 341–356. Economic Dynamics and Control, 37(8), 1479–1499.
Gadanecz, B., & Jayaram, K. (2009). Measures of financial stability. Proceedings of the IFC Moser, T. (2003). What is international financial contagion? International Finance, 6,
Conference on “Measuring financial innovation and its impact”. 31. (pp. 365–380). 157–178.
Bank for International Settlements. Moshirian, F. (2008). Globalization, growth and institutions. Journal of Banking & Finance,
Gertler, M., Gilchrist, S., & Natalucci, F. M. (2007). External constraints on monetary policy 32, 472–479.
and the financial accelerator. Journal of Money, Credit, and Banking, 39(2–3), 295–330. Park, C. -Y., & Mercado, R. V. (2014). Determinants of financial stress in emerging market
Gilchrist, S., Yankov, V., & Zakrajsek, E. (2009). Credit market shocks and economic fluc- economies. Journal of Banking & Finance, 45, 199–224.
tuations: Evidence from corporate bond and stock markets. Journal of Monetary Peersman, G., & Smets, F. (2001). The monetary transmission mechanism in the Euro area:
Economics, 56, 471–493. More evidence from VAR analysis, Working paper series 91. European Central Bank.
Gonzalo, J., & Martinez, O. (2006). Large shocks vs. small shocks. (Or does size matter? Schinasi, G. J. (2006). Safeguarding financial stability: Theory and practice. Washington, DC:
May be so). Journal of Econometrics, 135(1–2), 311–347. International Monetary Fund.
Haas, R. D., & Horen, N. V. (2013). Running for the exit? International bank lending during Schmidt, J. (2013). Country risk premia, endogenous collateral constraints and non-linear-
a financial crisis. Review of Financial Studies, 26, 244–285. ities: A threshold VAR approach. (Working Paper) Geneva: Institute of International
Hakkio, C., & Keeton, W. (2009). Financial stress: What is it, how can it be measured, and and Development Studies (IHEID).
why does it matter? Federal Reserve Bank of Kansas City. Taylor, J. B. (1993). Discretion versus policy rules in practice. Public Policy, 39, 195–214.
Hanschel, E., & Monnin, P. (2005). Measuring and forecasting stress in the banking sector: Van Robays, I. (2012). Macroeconomic uncertainty and the impact of oil shocks. (Working
Evidence from Switzerland. BIS papers chapters. In Bank for International paper series 1479) European Central Bank.
Settlements (Ed.), Investigating the relationship between the financial and real econo- Zheng, J. (2013). Effects of US monetary policy shocks during financial crises - A threshold
my. 22. (pp. 431–449). vector autoregression approach. (CAMA Working Papers 2013-64) Centre for Applied
Hansen, B. E. (1996). Inference when a nuisance parameter is not identified under the Macroeconomic Analysis, Crawford School of Public Policy, The Australian National
null hypothesis. Econometrica, 64(2), 413–430. University.