Probability and Risk Management For Sports Betting
Probability and Risk Management For Sports Betting
Probability and Risk Management For Sports Betting
management for
sports betting
Applying principles of portfolio management to real world betting
First Edition
May 2014
Probability and risk management for sports betting 1
Contents
1. Introduction ...................................................................................... 3
2. Exchanges vs. bookies .................................................................... 10
2.1 Who offers the best odds?........................................................... 14
3. Odds and probability ...................................................................... 16
3.1 What are odds? ........................................................................... 16
3.2 What is probability?.................................................................... 19
3.3 Different types of probabilities ................................................... 21
3.4 Some background on probability ................................................ 23
3.5 Forecasting probabilities............................................................. 32
3.6 Elo rating systems ....................................................................... 36
4. Risk management ........................................................................... 38
4.1 When should you bet? ................................................................ 38
4.2 How much should you bet? ........................................................ 39
4.3 What system should you use? ..................................................... 43
4.4 Practical issues with Kelly .......................................................... 47
4.5 Making Kelly work in practice ................................................... 50
5. Capital management ....................................................................... 56
5.1 Cross or wait? ............................................................................. 57
5.2 Liquidity ..................................................................................... 58
5.3 Spread ......................................................................................... 59
5.4 Total amount bet on a market ..................................................... 62
5.5 Maximum exposure to a market ................................................. 62
5.6 Capital required .......................................................................... 65
5.7 Closing out bets .......................................................................... 68
5.8 “Free” bets .................................................................................. 69
6. Bringing it all together .................................................................... 73
6.1 Portfolio management & practical issues ................................... 73
6.2 Risk tolerance ............................................................................. 73
6.3 Choosing your standard bet size ................................................. 74
6.4 Time value of money .................................................................. 78
6.5 Optimising risk/return................................................................. 82
Probability and risk management for sports betting 2
6.6 Allowing for error in your forecasts ........................................... 83
6.7 Correlation of bets ...................................................................... 85
6.8 Taking a profit (or a loss) ........................................................... 87
6.9 Chasing odds and getting whipsawed ......................................... 89
6.10 Software ...................................................................................... 90
6.11 Constant evaluation & improvement .......................................... 91
7. References ...................................................................................... 94
Probability and risk management for sports betting 3
1. Introduction
This book has one aim: it is about how to take your skill in betting
and turn it into a consistent profit by applying the portfolio
management techniques described in this book. Regardless of the
sport, or how you decide on your bets, this book outlines a system for
managing your portfolio of bets by controlling risk and allocating
your capital efficiently across the large range of available betting
options.
The most important requirement to be successful in betting is to
have an edge. The edge can be gained by having information that
others don’t have, or through superior analysis of publicly available
information. To be clear about what an edge is: it is not about
picking winners – most events have a clear favourite with the
bookies and the favouritism is usually justified. An edge comes from
you being a better judge of the probability of future sporting results
than a bookmaker or other punters on a betting exchange. In other
words, determining where the odds are wrong. Your edge is a
necessary step for you to become a successful punter but it takes
more to turn that edge into a consistent profit.
Betting is fundamentally risky and there is no way around the fact
that even where you have an edge and get great odds on an event,
any bet can lose as the events you are betting on are inherently
random.
This book describes some of the tools for taking your edge and
turning into a betting strategy – it does not detail a specific system
for a specific sport. It is about understanding the types and level of
risk you are taking on, and running a portfolio of bets with an eye to
controlling risk while maximising your returns. It uses ideas both
from betting/probability theory and from investment management.
Probability and risk management for sports betting 4
may want to try it live but with a small betting pool. This is a big
money saver as many seemingly good ideas don’t work in practice.
As any experienced punter knows, some good ideas don’t make
money, some even lose money.
Rational betting
This book is written with the rational punter in mind. Unless
you’re a robot you will always have some behavioural biases but the
book has a clear focus on head over heart punters. There are a
number of obvious behavioural biases that some punters fall for such
as “supporting your team” by betting on it, or trying to win back
what you’ve just lost. There are many less obvious biases that
undisciplined punters can fall for.
After winning a bet have you ever said “I should have put more on
that bet?” Then why didn’t you? This is an example of hindsight bias.
With the benefit of hindsight it is obvious that the outsider was going
to win the game and you should have bet your house. The reality is
that before the game it was far from obvious that the outsider was
going to win. That’s why the odds were so long and it’s probably
also why you didn’t put more on the bet. It’s also why you probably
shouldn’t have put more on the bet.
You will find that over time you will have good runs and bad runs,
and this will impact your confidence but this is often just chance
messing with your head. You should never forget the importance of
luck in betting. In the long run a skilful punter will win, but the fact
that luck always plays a part in your results is another reason why
you should not put all of your eggs in one basket. This is no different
to investing – you can tilt the odds in your favour but there will
always be times when you underperform.
Probability and risk management for sports betting 6
Other strategies
To borrow the language of the financial markets, this book is
about “investing” rather than “trading”. Investing is about forming a
view on the “correct” odds and placing your bets accordingly.
Trading strategies typically don’t have a view on the correct odds but
aim to make money through other means such as finding risk-free
opportunities (arbitrage), having a view on how odds will move from
their current level in the short run or making a market.
Probability and risk management for sports betting 8
What to avoid
When someone tells you that buying their betting system will
make you a fortune they are fibbing. If someone has a system that
makes a fortune they will not sell it to anybody – and certainly not a
complete stranger.
Probability and risk management for sports betting 9
50
45
40
35
30
Decimal Odds
25
20
15
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Probability
A B
II. Exhaustive
Two or more events are exhaustive if it is always the case that at
least one of the events occurs. It may be that more than one of the
events occurs. This is shown in this diagram.
Probability and risk management for sports betting
2
5
B A&B
Think of the outcome when you toss two coins. A is the event “at
least one tail” and B is the event “the second coin is a head”. The
diagram here shows the possible outcomes. You can think of the four
possible outcomes to see that the areas are A: HT, TT, B: HH, A & B:
TH. So the events are exhaustive – at least one of the events always
occurs. Note that in this case, Pr(A) + Pr(B) = ¾ + ½ =1¼ > 1.
A wins
B wins Draw
IV. Independence
Two events are independent if the occurrence of one event has no
impact on the probability that the other event will occur. For example,
each week in the premier league there are ten games and the
outcomes of each of these games should be independent of each
other or very close to it. You could contrive an example where
independence is not strictly true – such as where Manchester City
plays Chelsea in an FA Cup game during the week so they each play
a weaker lineup on the prior Saturday against lower ranking teams. If
a team is tired or carrying injuries this is also important to estimating
match probabilities. In practice though, this example would have
little impact on the independence of games.
Probability and risk management for sports betting
2
8
In sports betting, independence is important where you are
managing a portfolio of bets. Independence is an assumption that is
often made subconsciously. Independence between games is usually
a reasonable assumption but it is still important to acknowledge that
the assumption has been made, and to understand under what
circumstances it may not be met and what the impact would be.
There are examples where the assumption of independence is not met
and this has implications for probabilities in a total portfolio of bets.
This is expanded on in Section 6.7.
Independence can also have an impact in the calculation of
probabilities in compound/exotic bets. You need to take this into
consideration when you are calculating the odds for your exotic bet.
V. Expected value
Expected value is a theoretical concept but with an important
practical application. Expected value is very important in deciding
when and how much to bet. One way to think of it is the average
return we would expect if we could run the same event infinitely
many times. The coin example is again useful here. If the quoted
odds are 2 that a coin lands on heads then the expected value is 2 x
Pr(Heads) = 2 x ½ = 1. These odds are called fair odds as they don’t
advantage either side. If the quoted odds are less than 2 then the
expected value is less than 1 so this bet is almost certain to lose
money in the long run. Odds greater than 2 give an expected value
greater than 1 which means there is an edge.
In sports betting, the concept of expected value is harder to define
as each game is played only once so there can only be one
observation of the event. You could think of the expected value as
Probability and risk management for sports betting
2
9
the frequency of wins if we had an infinite number of parallel
universes and could observe the event in each universe.
At fair odds, the expected value is exactly 1.
Probability and risk management for sports betting
3
0
Forecast probabilities
Regardless of what method you use to forecast probabilities, the
sum of your forecast probabilities should be the same as the sum of
the actual probabilities, typically 1. Your forecast of the expected
value for each bet is simply your forecast probability multiplied by
the odds offered. If your forecast of the expected value of a bet is
more than 1, then you should place a bet as your forecast probability
of an event is greater than the probability that the bookie is pricing
into the odds. Conversely, if you forecast the expected value of a bet
to be less than 1 then don’t bet as the odds are too short.
Implied probabilities
The implied probability is what the bookie/market has priced in as
the probability of the event (adjusted for commissions). In the
context of the financial markets, this would be thought of as what
information the market has priced in and incorporates the collective
view of all market participants on the future prospects of returns.
Implied probabilities calculated using odds from a single
bookmaker will always add to more than one (unless they’ve made a
mistake). The sum of the implied probabilities is
1 1 1
1 (3.1)
d1 d 2 dn
The amount that this is above 1 is called the overround. This may
be as little as 5% for some markets, but is often closer to 20%. In
general, the more choices there are in a market, the higher the
overround. Overrounds for betting exchanges are typically, but not
always, lower than for bookmakers, especially for popular markets.
The overround is the bookmaker’s margin and it is a hurdle to
Probability and risk management for sports betting
3
1
making a profit, so generally you want this number to be as small as
possible.
When using a betting exchange there are two implied probabilities
for each bet – at the bet price and at the lay price. The sum of the
implied probabilities that you can bet at if you cross the spread is the
equivalent of odds offered by a bookie and is typically more than 1.
The sum of implied probabilities that you can lay at is typically less
than 1.
You can use the implied probability to decide whether to bet. If
the implied probability is less than your forecast probability then you
should bet as you believe the event is more likely than the
market/bookie does. Conversely, if the implied probability is more
than your forecast probability then you should not bet. In this
situation, it may even be desirable to lay the event. If the implied
probability is close to your forecast probability then you may choose
to sit this one out as there is little benefit in betting.
0 pd,bet pf pd,lay 1
where
pd,bet is the implied probability of the market’s current lay odds,
pd,lay is the implied probability of the market’s current bet odds,
pf is your forecast odds.
Probability and risk management for sports betting
3
2
3.5 Forecasting probabilities
The first question you should ask yourself before you bet is “What
do I know that others don’t?”
This is what is sometimes known in financial markets as your
“alpha source”. It is your strategy, system, ideas, or knowledge that
enable you to make a profit. The system presented here does not
provide an alpha source. It is designed to manage a portfolio of bets
based on your alpha source, whatever it is, as long as it genuinely is
able to forecast probabilities more accurately than other punters in
the market.
You can use any basis at all for forecasting – from a team’s recent
form to astrology. Generating a forecast is easy. Generating useful
(profitable) forecasts is not so easy and there is no method to
guarantee profitable forecasts. Not all of your forecasts will be right
but you need to be able to consistently generate profitable forecasts if
you are going to make money from betting. Picking that the top team
is going to beat the bottom team is probably correct and will help you
in a tipping competition, but it’s also pretty obvious and the bookie’s
odds will already reflect the strong chance that the top team is going
to win. Your forecast needs to help you determine where the market
odds are wrong. Has the bookie over- or under-estimated the
difference between the teams and hence the chance of the strong
team winning?
A forecasting system can be qualitative, quantitative or some
combination of the two. Regardless of how it works, it will use
currently available information to form a view about the future. As
new information comes to light, your forecast should change to
reflect this and the market odds are also likely to change. There are
Probability and risk management for sports betting
3
3
many factors that could be considered in a forecasting model –
different things work for different markets. Two completely different
systems could both work, although probably at different times. There
is no right answer and some seemingly sensible systems don’t work –
possibly because the information being used is already factored in by
the market as everyone else knows about it. Systems don’t have to be
complex. You may want to experiment with a paper betting system
for a while to test if you can make money in a market before you
start betting with real money.
Some factors to consider in constructing your forecast odds:
Current position of each team on the league table
Most recent results for each team (and strength of opponents)
Goal difference (or percentage for AFL)
Attacking/Defensive record
Home ground advantage (or distance travelled)
How well a team travels
Each team’s line-up for this game (consider player ratings)
Any injuries or suspensions
Time since last game and recent workload (fatigue)
The budget for each team
The depth of each team’s player roster
The weather forecast
Several of these measures could be combined into a “team
strength” indicator.
You can use a quantitative system to measure some or all of your
factors, or you can use your judgment. You should understand why
each factor you consider contributes to providing a more useful
probability forecast. You may also want to test your factors on a
Probability and risk management for sports betting
3
4
previous season’s results but there is no substitute for doing a live
run, either with a shadow portfolio or with a small amount of money.
There are many more factors that you could consider and you are
really only limited by your imagination.
As you add more obscure factors it is likely that they will each
have less impact on making an accurate forecast of the probability of
an event and may not be worth the effort. A few key factors will
usually get you pretty close in terms of estimating the probability.
This idea is known as the Principle of Parsimony – keep the model as
simple as is necessary to do the job.
You should aim to make your system robust. A small change in
the inputs should produce a small change in the forecast probability
and not a dramatic change in your suggested bet. If this is not the
case then forecasts are likely to be over-influenced by unimportant
factors or noise, and your forecasts may not be reliable.
In designing a forecasting methodology you are only limited by
your imagination. Qualitative and quantitative forecasters are
typically suspicious of each other but both can work well in the right
hands and can even be combined to produce a hybrid system which
may be more powerful than either system separately. When it comes
to risk management, some basic quant tools are very helpful. To use
quant risk management, your forecasting method doesn’t have to be
quantitative. Even if your forecasts are quantitative, mathematically
sophisticated models are not necessarily better than simple models.
Your insights into the sport are far more important than mathematical
or statistical theory. One quantitative element you do need is a
number as the outcome of your system (at least an approximate
estimate) and that is your forecast probability.
Probability and risk management for sports betting
3
5
Financial market quant analysts have a saying – “garbage in,
garbage out”. In other words, a bad model with poor data is useless
and exactly the same applies in betting as in financial markets. You
need to ensure that your data source is free from errors.
When forecasting probabilities you may want to combine
estimates from more than one system to improve reliability. You will
probably need to combine the team strength scores with other factors
to arrive at your forecast probability. The challenge is that you are
converting an open ended scale to a 0 to 1 scale.
A common way to convert a variable to a 0 to 1 scale is with a
logistic function. It takes an open ended strength scale and
transforms it to the probability scale with a minimum of 0 and a
maximum of 1. Again, this could be a quantitatively or qualitatively
derived strength score but the conversion is quantitative. If you
decide to use a logistic function you’ll need to calibrate it.
The chart below shows a logistic function. There are no values on
the x-axis as this is intended to show the general shape of the
relationship and, as mentioned above, the curve must still be fitted to
your data.
Probability and risk management for sports betting
3
6
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Fixed bet is where you bet a fixed dollar amount on each bet
regardless of the size of your pool, the odds being offered or the
amount by which you think the bet is mispriced (provided you
believe it is mispriced). This is the simplest system but it doesn’t take
into account valuable information about risk and return. The system
has low volatility in the pool size.
Fixed winnings is where you set your stake so that your winnings
will be equal to a fixed amount if you win the bet. This system tends
to bet more on bets that are more certain. An example of how fixed
winnings would work in practice: if you bet $1 at odds of 2 then your
winnings is $1. If you bet $2 at 1.50, or $10 at 1.10 then if you win
the bet your winnings will also be $1.
If C is the amount you want to win when you are successful then
you can calculate the size of your stake, s, as:
C
s . (4.1)
d 1
In effect this is a risk adjustment to your stake size and is a natural
safety mechanism because you end up betting less on long shots
(which usually don’t pay off) and more on favourites. As a result,
this system has less volatile outcomes than the fixed bets or fixed
percentage systems. Note that the bet size doesn’t depend on your
forecast probability or edge. The system has low volatility in the pool
size.
No system is perfect but the Kelly system ticks most of the boxes
with less drawbacks than other systems.
Pros
The Kelly system takes bigger bets where your edge is greater
which helps to maximise returns.
Probability and risk management for sports betting
4
6
The Kelly system takes bigger bets where odds are shorter
(probability is higher) which reduces the volatility of your
pool size.
Kelly leads to the highest expected winnings (not the highest
possible winnings) and it can be shown mathematically that
no other system has a higher expected outcome. This is
“expected” winnings in the mathematical sense of the word
expected – the weighted average of all possible outcomes
multiplied by their probabilities.
Probability and risk management for sports betting
4
7
Cons
The Kelly system can lead to a volatile bank balance.
Anybody who has implemented this formula exactly as
shown above would understand that it can be a pretty exciting
ride. As was mentioned, some punters recommend using a
fractional Kelly system where you bet a fraction of the
amount recommended by the Kelly formula. This will lead to
lower profits but less volatile outcomes. Betting more than
the amount suggested by Kelly is never a good idea – it will
produce a worse outcome with more volatility. In extreme
cases it can wipe you out very quickly.
If the back and lay amounts are close to zero this means the odds
are about fairly priced and you may be best leaving it and looking for
a bet where you have a clearer edge.
If you are betting on an exchange then you have the choice of
crossing and backing at dlay or waiting and offering to back at dbet.
You can calculate the back amounts for both odds values depending
on whether you decide to cross or wait. We will discuss whether you
should cross or wait more in Chapter 5.
Adding to a bet
There are times that you may want to add to an existing bet or lay.
This may happen, for example, where you have placed a range of
bets in a market but wish to increase your exposures further. In
determining the size of your additional bet, it doesn’t matter whether
the forecast or market odds have changed, the only factors that
influence the size of the additional bet are the current forecast and
market odds, and your current risk exposure, r. The Risk
Contribution is defined as r = w – l where w is the amount that we
win if an event occurs and l is the amount we lose if the event
doesn’t occur.
The individual previous bets or expected return don’t influence
the additional bet size other than through the impact they have on
Probability and risk management for sports betting
5
4
your risk exposure. The formulas for back and lay amounts from
above are then adjusted for the amount that you already have at risk.
Back amount = S (d . p f 1) /( d 1) r / d (4.8)
pd p f
S (4.11)
Lay amount = pd
0 pd,bet pf 1
5.2 Liquidity
The liquidity of a market can be defined as the ease with which
you can match a bet of your desired size. In the most popular markets
you can typically bet a large amount with ease. Liquidity can vary
over time and between runners in the same market. It’s not possible
to precisely quantify liquidity but there are some measures that help
to measure it in practice.
Two quick measures of liquidity are: the amount that has been
matched on a market to date; and the amount which is currently
waiting to be matched on the market. While these measures have
value in most circumstances, they have limitations. The amount
matched to date can be less relevant where a market dries up – for
example a “season winner” market may lose interest for punters once
the outcome is nearly certain. In many markets, as events unfold and
an outcome becomes unlikely, odds will move out there will usually
be few backers at odds of greater than 50 or 100. The amount waiting
to be bet may also not be a good indicator in some circumstances as
the liquidity may not be where you are looking to bet, or may be at
very unfavourable odds.
In practice liquidity is variable and unpredictable and must be
monitored closely. The impact of liquidity on your strategy will
depend on the type of strategy that you are trying to implement. For
strategies that involve crossing the spread, more liquid markets are
generally more attractive as the cost of crossing the spread is lower
because the spread is narrow. This is analogous to financial markets
where the greatest cost in buying or selling an investment can be the
Probability and risk management for sports betting
5
9
cost of crossing the spread. It is a cost that can make a strategy that
appears profitable lose money. The downside of more liquid markets
with narrower spreads is that they are typically more efficient and are
harder to make money out of. In the most liquid markets there are
usually a number of professional players and you should be wary of
betting against them – they are professional for a reason. Depending
on your strategy, the markets with good but not great liquidity can
provide the sweet spot for profitability. These
“goldilocks” markets provide a nice balance between liquidity and
efficiency. Obscure, illiquid markets can be difficult to run a
systematic betting approach as you will struggle to match enough
bets but there are ways of managing risks within these markets.
There is a saying in financial markets that liquidity creates
liquidity. The same applies in betting markets - punters will favour
markets that are liquid because that is where they can bet a
reasonable amount without distorting the market odds. This then
attracts more punters to those markets. That is, the liquidity attracts
more liquidity.
5.3 Spread
If liquidity determines how much can be bet, the spread indicates
how much it will cost to bet now by crossing the spread rather than
waiting. A simple measure of the spread for an individual runner in a
market is the ratio of the odds at which you can bet to the odds at
which you can lay. You can measure it for an individual bet as
Spread = d b,i / d l ,i pd ,lay / pd ,bet (5.1)
The Spread must always be greater than 1. If you are using more
than one bookie/exchange then you may find examples where it is
Probability and risk management for sports betting
6
0
less than 1. This means you have found an arbitrage opportunity
where in theory you can make a riskless profit – if the odds don’t
move before you get set on both sides. In practice the spread can be
very close to 1 for popular, liquid markets. The spread measure can
make things appear worse (or better) than they are. A wide spread
doesn’t necessarily mean that crossing the spread will give you bad
odds as it depends where your probability forecast sits relative to the
implied probabilities. The diagram below shows an example where
the spread is wide (about 3) but it is still possible to bet at favourable
odds when crossing the spread and betting at pd,lay as the market odds
are well above the estimate of where they should be. If you are
prepared to wait then the odds could be even more favourable, if you
are matched.
0 pd,bet pd,lay pf
0 pf pd,bet pd,lay
1 / d l ,i p d ,lay,i
Market Spread = i 1
n
i 1
n
(5.2)
1 / d
i 1
b ,i p
i 1
d ,bet ,i
This formula may look a bit intimidating but it is simply the total
of the implied probabilities using current market lay odds (the
overround for a traditional bookie), divided by the total of the
implied probabilities using current market back odds. Note that as
with the spread for a single runner, a large spread doesn’t mean that
you can’t bet at good odds. If you are betting only then the important
factor is the top line of the equation. If this is close to one (or the
number of winners) then this is the same as a bookie having a small
overround. The fact that the bottom line of the ratio may be much
smaller than one is of no concern to you if you are only betting as it
describes the opportunity to lay. Similarly, if you are only laying bets
then it is the bottom line of the equation that is of interest. You
require that the bottom line is close to one (or the number of winners)
and have no interest in the value of the top line.
Market spread is a rough measure of the liquidity of a market: it
shows how tight the market is on average across all of the runners
and roughly how much you will have to sacrifice if you want to cross
the spread to ensure your bet is placed, rather than wait. The
individual spread gives you an idea of how tight the market is for an
individual bet. Neither of these spread formulas measure the amount
available at either side of the spread so they are imperfect measures
of liquidity but useful inputs nonetheless. In practice, markets with a
Probability and risk management for sports betting
6
2
tight spread will typically also have higher amounts available to bet
and vice versa.
Betting on illiquid markets can be frustrating and is often not very
fruitful as many bets go unmatched. If your favourite sport has little
liquidity then you may be better off either using a bookie who offers
odds on the sport, or finding another sport to bet on.
The second step is to calculate the Risk Contribution for each bet.
This is simply the Bet Wins amount minus the Bet Loses amount for
each bet. In this example there is only one winner, as only one of
Probability and risk management for sports betting
6
4
these results can occur. The biggest risk contribution comes from the
largest negative number (or the smallest positive number if all
numbers are positive). In this example it is -$29.25.
To get the total risk exposure for the market, sum the Bet Loses
amounts which gives $5.71 here. Then add this amount to the
smallest (generally the largest negative) risk contribution. This gives
a total of -$23.54. In other words, if everything goes against you, the
most you can lose on this market is $23.54 which occurs if the home
team wins and then you make -$18.00 + -$12.50 + $6.96.
How can this calculation be adapted to the case where there are
multiple “winners”, such as picking who will finish in the Top 4
positions at the end of the AFL season? The algorithm is the same
except that you need to identify the four largest risk contributions
and sum them. In the case of the Top 4 in the AFL, you first put all
18 teams into a table with the Bet Wins and Bet Loses data, then
calculate Risk Contributions. The second step is to star the four
biggest Risk Contributions. The Total at Risk is the sum of the four
starred Risk Contributions minus the sum of all numbers from the
“Bet loses” column. This is described in mathematical terminology
below.
In mathematical terms, consider a market that has n teams and m
winners/place getters. We define wi as the amount that paid by a win
if event i occurs and li as the amount lost if event i doesn’t occur.
Individual Risk Contributions are defined as:
ri wi li (5.3)
Set r(i) to be the ri values sorted from smallest to largest. Then the
total Risk Contribution is:
Probability and risk management for sports betting
6
5
m
R r(i ) (5.4)
i 1
The total capital allocated for the market (maximum loss) is:
C* = R*+L* (5.10)
The amount of capital required to be set aside for unmatched bets
is C – C* which is a positive amount if further capital is required to
be set aside.
Extending the example from Section 5.55.5, we offer to lay an
additional further $15 at odds of 1.40. The Bet Wins value for the
Away team, w(3)*, is now -$24.60. This makes r(3)*, the Risk
Contribution from the Away outcome, -$31.56.
Bet wins Bet loses Risk contribution
Home -18.00 11.25 -29.25
Draw 50.00 -12.50 62.50
Away -24.60 6.96 -31.56 *
Sum 5.71 -31.56
Total at risk -25.85
The maximum total at risk with the additional bet is $25.85, so the
further capital required to offer lay this bet is -23.54 – (-25.85) =
$2.31.
Probability and risk management for sports betting
6
8
Placing additional bets does not necessarily increase the amount
of capital at risk, and the capital required may be zero. This is a “free
bet” in the capital allocation sense and this concept is discussed
further in Section 5.8.
Confidence
Another consideration is your degree of confidence in your
forecast. Is this a market where you have a clear edge? How
confident are you in your forecasts? Do you have past experience
betting on this market? Have you fully researched or are you still
learning about the market?
Where you are more confident of your forecasts, you should bet
larger sums. For markets where you are less confident, or if this is
your first foray into a new market, you may want to start with a small
standard bet size and build up from there.
Liquidity
Exchanges have practical limits on how much you can bet that
depend on how many other punters are also playing in the market.
Some markets have very little liquidity and this limits the amount
you are able to bet, or at least the amount that you are able to bet at
favourable prices. In these markets, it is best to start with small bets
and build up for both risk management and capital management
reasons. The risk management aspect is discussed in the next point
on diversification. The capital management issue is that you will be
placing a large amount of bets on the market which will be waiting
for someone to take the other side of the bet, especially if your
standard bet is large. As this market is illiquid, you are likely to end
up with very few bets matched despite having allocated a lot of
capital to the market while it was live.
Probability and risk management for sports betting
7
6
A lack of liquidity can also be a danger sign, depending on your
strategy, as it will make it harder to place more bets later if you want
to diversify your exposure or close a bet.
Diversification
Diversification is closely linked to liquidity, as your ability to
diversify in a market is impacted by the market’s liquidity. To
maintain diversification in your portfolio, you may choose to build
up your bets in stages. For example, if you are aiming to have a
standard bet of, say, $500 on the winner of the Premier League and
you place bets based on this standard bet size then it’s possible that
you end up with only one $500 bet matched giving you an
undiversified total exposure. Alternatively, you can build up your
standard bet size gradually. For example, you could initially set your
standard bet at $100, then when you have a good spread of bets at
this level and you have a reasonably diversified portfolio, you
increase the standard bet to $200. As your bets are matched and your
portfolio has reasonable diversification, you can repeat this until you
get to your target standard bet size – or until your capital is all
allocated.
Combining these factors is a matter of experimentation and
depends on the types of markets you play, the strategies you use and
how these interact. If you are looking only at individual games you
should get a feel for setting your standard bet after a few weeks. If
you are betting on season bets such as the winner of the Premier
League then it may take longer to get the level right so you may wish
to start with a cautious approach. It is also worth noting that this
approach to sizing season bets soaks up more money as the season
goes on so building up your standard bet size gradually helps to
Probability and risk management for sports betting
7
7
ensure that you have capital available throughout the season. If in
doubt, start below where you think your standard bet should be and
build up to it.
Probability and risk management for sports betting
7
8
Cut your losses, move on
It’s important that your capital is employed where it can earn the
highest return. If one of your current bets is going bad and unlikely
to turn around then the best thing to do is cut your losses, escape with
some of your money intact, and use that capital more productively
elsewhere. When a bet is going against you it is usually better to take
a small but painful loss now rather than the even more painful 90%
(or 100%) loss later on. Learning to cut losers is one of the best
lessons that you can learn in betting and something that good traders
in the financial markets know how to do. Financial markets traders
will often have stop losses – where they automatically sell anything
that drops, say, 10% in price. This limits the loss and stops them
from falling into the trap of justifying to themselves why they should
continue to hold it only to watch it continue to fall. Stop losses can
also make sense in betting markets, depending on the strategies you
are using, and you should consider adapting the idea to work with
your strategies.
where r is your cost of capital (the risk free annual interest rate
that you can earn) and t is the fraction of a year until the bet settles.
Where d.pf < (1 r)t you shouldn’t bet as you would be better
leaving your money in the bank.
Sensitivity analysis
Sensitivity analysis is frequently used in the financial services to
assess the impact if an assumption, estimate or forecast is wrong. In a
sensitivity analysis, you redo your analysis using forecasts for each
parameter that are a small amount higher and a small amount lower,
for example, how does a small change in the level of forecast sales
impact a company’s value?
Sensitivity analysis can be applied in a similar way to sports
betting. What if your probability forecast is out by 1%? Or 5%? If
your team rating is wrong by a small amount then what impact would
it have on your bet size? It may be that you still run with your
original bet size but you will better understand the impact of any
errors in your system. By conducting a sensitivity analysis on each of
your key variables, you may find that some factors are more
important drivers of risk and return, and you may choose to stick to
more robust bets that are less dependent on small variations in your
assumptions.
If you “stress” your probability forecast you will notice that the
impact varies depending on the context. In particular, where the
implied and forecast probabilities are both small, the recommended
bet size (and even whether you should bet) is extremely sensitive to
small changes in the level of these probabilities. Where the
probabilities are large, there is much less sensitivity to changes in the
Probability and risk management for sports betting
8
5
probabilities. The best way to understand the sensitivity to
parameters for the types of bets that you favour is to put the formulas
from Section 4.4 into a spreadsheet and vary each of the numbers,
noting the impact. This analysis may lead you to tone down your bet
sizes where a small forecast error could make a bet much riskier.
6.10 Software
Good software will make your life much easier. Even if you have
a qualitative approach and don’t use software to forecast probabilities,
the risk management system outlined here is far easier to implement
if you use some simple software. The spreadsheet included with this
book covers a basic risk management system for betting on season
results in a 20 team competition and for betting on a soccer game
with three outcomes. The 20 team template can be adapted to any
number of teams, a horse race or any other event.
Probability and risk management for sports betting
9
1
If you want to keep it simple, just adapt the spreadsheet and enter
your forecast probabilities for your chosen market. Alternatively you
can use this spreadsheet as a basis and expand it further to add in
other information that is useful to you. You are really only limited by
your spreadsheet skills. For spreadsheet experts and experienced
programmers, Betfair provides a spreadsheet template that enables
you to pull data directly from the exchange into a spreadsheet via the
API. The API is a big labour and time saver and worth a bit of effort
to integrate into your spreadsheet if you can. Some other bookies and
exchanges may offer similar facilities.
To use the included spreadsheet you first need to set the
parameters. What is your standard bet for each market? What is the
minimum bet allowed by the bookie/exchange (this varies by
currency)? Next you need to enter your probability forecasts for each
runner. The forecasts must add to 1 (or to the number of runners).
You can either enter the forecast odds for each market directly or you
can enter the probabilities for each team finishing in each position
and the spreadsheet will then calculate odds for the separate markets.
The second approach is only practical for those using a quantitative
process.
1. It can highlight what type of bets are making you money and
what type of bets are losing you money;
Probability and risk management for sports betting
9
2
2. It can alert you to changes in the pattern of returns. Markets
are dynamic and if others are catching on to the same idea
that you have then this tells you that it’s time to review your
strategy or cut this market;
3. It can highlight patterns over time, for example, you may be
more successful at different stages of the season.
If you discover that there are types of bets where you make money
and others where you lose money, you can give more weight to your
winning strategies and try to fix your losing strategies.
The first step in understanding your strengths and weaknesses is
that you need to collect the data on all of the bets that you make. At a
minimum, the data you need is the type of bet, the odds, your
forecast probability, back or lay, time of the bet. You may want to
collect other information, such as why you are making the bet, as it’s
always easier to collect information as you go along rather than
trying to reconstruct it after the event. It also keeps you honest and
stops you from justifying bets after the event.
The type of analysis that you conduct will depend on the strategies
you are using. The first things to look at are:
success at different times of the season or times of the
week for individual games;
success in different odds ranges;
success on backs vs lays;