Book Read Free

Positional Option Trading (Wiley Trading)

Page 19

by Euan Sinclair

dynamic trade sizing and the use of stops and profit targets, but

  unless there is real edge somewhere the idea will eventually lose

  money. However, this doesn't make risk control irrelevant. Bad

  risk management can turn a potentially profitable idea into a

  loser. And, because risk management is the only part of the

  trading process that is completely under the control of the trader,

  there is no excuse for not doing it as well as possible.

  In this chapter, I restate why the Kelly criterion should form the

  basis of a trade-sizing scheme and look at two extensions that are

  particularly important when trading options: non-normal trade

  returns and uncertainty about the return distribution.

  The Kelly Criterion

  It is well-known that investing according to the Kelly criterion

  (Kelly, 1956) will theoretically outperform any other sizing

  strategy. No other sizing scheme will produce greater long-term

  growth. This doesn't mean that everyone tries or even wants to

  invest like this. There are three kinds of reasons for this:

  The Kelly criterion maximizes the long-term growth rate of the

  bankroll. But it is completely legitimate to have other goals.

  For example, many traders are more interested in maximizing

  the probability of hitting a goal in a given time (see Browne,

  1999, 2000a, 2000b). No one has a utility function that is as

  simple as the log function that matches the Kelly scheme. This

  is fair.

  Some traders try to deny the math. They don't like the volatile

  return stream and decide that this is a flaw of the system. It

  isn't. Whether or not you like what Kelly says, it is a

  mathematical fact. It's objectively true. It is common for

  people to conflate their dislike of a situation with its truth.

  169

  Examples of this are evolution, climate change, and the Kelly

  criterion.

  Some people create strawman arguments about the

  mathematics, claiming that Kelly only applies to simplified,

  unrealistic situations. This isn't true at all. The mathematics of

  maximizing growth rate are quite general.

  A derivation of the Kelly criterion for both discrete and continuous

  outcomes is given in Sinclair (2013), together with a discussion of the distribution of results we can expect when investing this way.

  The important results can be summarized as follows.

  Good

  Kelly maximizes growth rate.

  The expected time to reach any goal is minimized.

  It is impossible to go bankrupt.

  The strategy depends only on the current bankroll, not the

  specific trade results that led to it.

  It is essentially unbeatable.

  Bad

  The best bets can be uncomfortably large.

  Portfolio volatility and drawdowns are large.

  Because of compounding, it is reasonably common to find that

  an equal number of wins and losses leaves you with a net loss.

  Here I want to look at two slight extensions that are very

  important to traders, particularly option traders. What happens

  when outcomes are highly non-normal? And what happens when

  we are uncertain of outcomes and probabilities?

  These are two aspects of the same general problem. Trading

  success is largely dependent on how robust our ideas are. At best

  our knowledge is uncertain. Probably our knowledge is incomplete

  and only partially correct. In particular, we will be ignorant of the

  true probabilities of rare events. These will be the events that drive

  170

  non-normality, and, because they appear only rarely, they will be

  those that we are most uncertain of.

  To introduce the ideas, we first look at the fairly impractical case

  of discrete trade results.

  Non-normal Discrete Outcomes

  Imagine we have a discrete set of outcomes Wi, each with

  probability pi. We bet a fraction, f, of our bankroll on each

  opportunity. So, the gain factor per trade is

  (9.1)

  Alternatively, the exponential growth per unit bet is

  (9.2)

  To find the value of f that maximizes the exponential growth rate,

  we differentiate with respect to f and set the derivative to zero. If i is greater than 2, this is unwieldly or impossible and we need to

  use numerical methods. A numerical solution of a simple example

  is illustrative.

  Case One

  p1 = 0.55

  p 2 = 0.45

  W1 = 1

  W2 = −1

  (55% chance of winning a dollar and 45% chance of losing a

  dollar)

  This implies f max = 0.1. The dependence of the exponential growth

  rate on f is shown in Figure 9.1.

  171

  FIGURE

  9.1

  Growth

  rate

  as

  a

  function

  of

  f

  (p1=0.55,p2=0.45,W1=1, W2=−1).

  Reconsidered Case

  p1 = 0.55

  p 2 = 0.43

  p 3 = 0.02

  W1 = 1

  W2 = −1

  W3 = −3

  Here the probability, p3, of the extreme event is low enough that

  we could easily misestimate it from historical data. But the

  implications of including this small probability are far from

  negligible. The growth rate is illustrated in Figure 9.2.

  In this case f max is 0.5, half of that in the two-outcome case. And

  growth rate becomes negative for f > 0.1. An event with only a 2%

  chance of occurrence could easily be missed when we estimate

  parameters, and if we erroneously think p3 = 0, we will bet at a

  size that gives a negative growth rate.

  This phenomenon is qualitatively similar when returns are

  continuous. Because this is the more relevant situation for trading,

  172

  that is what we will assume when deriving methods to live with

  these issues.

  FIGURE 9.2 Growth rate as a function of f (P1 = 0.55, P2 = 0.44, P3 = 0.01, W1 = 1, W2 = −1, W3 = −3).

  Non-normal Continuous Outcomes

  We are interested in the case where the outcome of a trade is

  known to have a certain continuous distribution. We bet a

  fraction, f, of our wealth at the start of each period so that

  (9.3)

  where Bn is the random variable giving the result of the n th trade and it has the payoff g( Xn). After a sequence of n trades our bankroll will be

  (9.4)

  Now we take logarithms:

  (9.5)

  173

  so

  (9.6)

  (9.7)

  where Φ( x) is the distribution function that describes the results of

  the trades. If we maximize over the bankroll fraction, f, we find

  that the optimal value is the one that satisfies

  (9.8)

  Applying a Taylor expansion to this equation gives

  (9.9)

  (9.10)

  (9.11)

  This can be further simplified if we note that

  (9.12)

  is the payoff to a unit bet.

  Further

  174

  (9.13)

  (9.14)

  (9.15)

  where

  and

  are the third and fourth raw moments of

  .

  So,
if f is small, we can truncate the series after the first term to get

  (9.16)

  And further, if μ is small, we can further approximate by

  (9.17)

  This is the usual expression for the Kelly ratio of a trade with a

  continuum of outcomes, but it is only an approximation and if we

  are in a situation where skewness is important, a better

  approximation can be obtained if we keep the third term, so that

  equation 9.13 becomes

  (9.18)

  We can solve this equation to get

  (9.19)

  Equation 9.19 only has real solutions if

  (9.20)

  175

  (which is a limitation of our sloppy use of asymptotics).

  Further, it isn't immediately obvious which root is the correct one.

  Also, the case where skewness is zero leads to a singularity. We

  can address these issues by taking the limit as skewness

  approaches zero.

  To do this note that

  (9.21)

  (if b is small relative to a).

  So if

  (9.22)

  we can write

  (9.23)

  And so the negative root of equation 9.19 is approximately

  (9.24)

  which simplifies to

  (9.25)

  So, in order for the limiting case to agree with the Kelly fraction

  when trades are normally distributed (equations 9.16 and 9.17), we need to take the negative root.

  From a practitioner's perspective, the important thing is to note

  that negative skewness decreases the optimal investment fraction

  176

  and positive skewness increases the optimal investment fraction.

  This effect is shown in Figure 9.3.

  Figure 9.4 shows the approximation of equation 9.25.

  FIGURE 9.3 The optimal investment fraction as a function of skewness (return is 0.015, volatility is 0.5).

  FIGURE 9.4 The approximate investment fraction as a function of skewness (return is 0.015, volatility is 0.5).

  Uncertain Parameters

  177

  The value of the optimal sizing fraction will generally need to be

  estimated from empirical data. Because empirical data will always

  have sampling errors and uncertainty, the estimate of the sizing

  parameter will also have a degree of uncertainty attached to it.

  This is well-known by professional gamblers. And to mitigate the

  risk of over-betting, bettors following a Kelly scheme often modify

  the Kelly criterion by investing only a fraction of the optimal

  amount. These schemes are known as “fractional Kelly” sizing. By

  doing this, traders accept that they will be reducing growth but

  will also more drastically reduce variance.

  However, simply scaling the investment fraction doesn't protect

  against a bigger problem: the case in which the investment

  fraction is estimated to be positive, but the true value is negative.

  In this case, investing any positive fraction of the bankroll will be

  over-betting.

  In order to estimate the chances of this happening we need the

  variance of the estimated Kelly criterion ratio (Sinclair, 2014).

  fmax (approximated in equation 9.17) is a statistical estimator and

  has an associated probability distribution.

  First consider the case of normal trade results. Here the central

  limit theorem says that the estimators of the mean,

  , and

  variance,

  asymptotically have the following normal

  distributions, where

  and

  are the population mean and

  variances respectively.

  (9.26)

  (9.27)

  Alternatively, the estimation errors of mean,

  and variance,

  can be approximated by

  (9.28)

  178

  (9.29)

  denoted by f(μ,σ2), the Kelly ratio of equation 9.17. So the estimator is just

  . The estimation errors in the mean and

  variance will lead to estimation errors in f.

  If we define theta to be the column vector of the normal

  distribution's parameters, this has an estimate of

  .

  For IID returns,

  where

  is the variance of the estimation error of

  .

  Denoting the estimator of the Kelly ratio to be

  where f() is

  now a function that estimates the Kelly ratio, we next apply the

  delta method (see, for example, Oehlert, 1992).

  This states that the variance of a function

  is

  (9.30)

  (9.31)

  and

  (9.32)

  so, evaluating equation 9.30 gives the asymptotic variance of our estimate of the Kelly ratio as

  179

  (9.33)

  If the trade returns are not normally distributed, we need to make

  use of the result (Zhang, 2007) that

  (9.34)

  where

  is the third central moment of the population

  distribution. Now equation 9.30 gives

  (9.35)

  (9.36)

  It isn't possible to find the variance of the sizing fraction given by

  equation 9.19, because the variance of the skewness would need to

  be evaluated for the particular distribution the results were drawn

  from. The best we can do in general is to measure the empirical

  skewness, calculate the sizing ratio using equation 9.19, then

  estimate the variance around that value by using equation 9.36.

  We now use an example of real trade results to show the

  importance of including estimation error in trade sizing. The trade

  results are from a proprietary short volatility strategy. It is

  somewhat typical of many such strategies in that it has a positive

  expected value but a large negative kurtosis. The summary

  statistics for these trade results are given in Table 9.1 and the

  distribution of results is shown in Figure 9.5.

  We can rearrange (and slightly modify) equation 9.36 to give an

  explicit expression for the estimated standard deviation of the

  Kelly ratio.

  180

  (9.37)

  where the denominator of n − 1 is due to applying Bessel's

  correction.

  TABLE 9.1 Summary Statistics for the Option Trade

  Sample size

  1000

  Mean

  $0.059

  Standard

  deviation

  $1.137

  Skewness

  ($6.199

  )

  FIGURE 9.5 The distribution of the option trade results.

  Because of the central limit theorem, we know that the

  distribution of f is normal so we can calculate the probability that f is actually below any critical value f*.

  181

  (9.38)

  where Z is the cumulative distribution function of the normal

  distribution with mean of f and a standard deviation calculated

  from equation 9.38.

  Equation 9.17 gives the Kelly ratio as 0.046, but equation 9.37 tells us that the standard deviation of this point estimate is 0.031, so

  our point estimate is only 1.4 standard deviations above zero.

  There is an 7% chance that the true Kelly ratio of the population is

  less than zero.

  Having an expression for the sampling distribution also enables us

  to estimate the chance that we are over-betting so much that ou
r

  growth rate is negative. This case corresponds to the true value of

  f being roughly less than half the estimated value. Equation 9.38

  tells us this is 25%.

  TABLE 9.2 Fractional Schemes Corresponding to Various Probabilities of Over-Betting

  Chance of Over-

  Corresponding

  Kelly Scale

  betting

  Benchmark

  Factor

  0.1

  0.0022

  0.0480

  0.15

  0.0104

  0.2301

  0.2

  0.0169

  0.3748

  TABLE 9.3 Fractional Schemes Corresponding to Various Probabilities of Over-Betting When Setting

  Skewness of the Trading Results to Zero

  Chance of Over-

  Corresponding

  Kelly Scale

  betting

  Benchmark

  Factor

  0.1

  0.0092

  0.2054

  0.15

  0.0161

  0.3574

  0.2

  0.0215

  0.4782

  This leads us to a complimentary way to use the information. We

  can use equation 9.38 to solve for a benchmark given that we want 182

  a certain chance of over-betting. For example, we have just seen

  that using a benchmark of half the measured Kelly fraction (i.e.,

  betting at “half-Kelly”) still implies a 25% chance that we will be

  over-betting. Table 9.2 shows the probabilities of over-betting for various fractional Kelly schemes.

  So, in order to introduce a margin of safety we would need to scale

  the measured Kelly ratio by a considerable amount. This is in line

  with the practice of professional gamblers. Much of this need for

  scaling is due to the presence of negative skewness. If the returns

  were normally distributed, the scaling could be reduced. This is

  shown in Table 9.3.

  Kelly and Drawdown Control

  Even after calculating and allowing for our measurement

  uncertainty, it is likely that investors will find that investing the

  full Kelly fraction leads to results that are unpalatably volatile.

  And the more edge there is, the higher the Kelly ratio will be and

  so the higher the volatility will be. Good trades are the most

  volatile.

  The standard way to mitigate drawdowns is to trade using a

  fraction of the Kelly ratio. In this case, both growth rate and

  volatility will drop. If we trade at a fraction, f, of the Kelly ratio,

 

‹ Prev