Anda di halaman 1dari 5

Properties of an ideal risk measure

By: Peter Urbani

Risk like beauty is largely in the eye of the beholder. Although we can probably all agree that risk has something to do with the possibility of loss either in relative- (keeping up with the Joness) or absolute(keeping above the breadline) terms, there are now a multitude of possible risk measures available to confuse you. These range from Standard Deviation, Semivariance, Downside Deviation, Beta, Sharpe Ratio, Sortino Ratio, Treynor Ratio, Jensens measure, M2, LPM, Tracking Error, Information Ratio, Value at Risk, Expected Shortfall, Shortfall Probability, Extreme Tail Loss, Expected Regret, Maximum Drawdown the Kappa function and now the Omega function to name but a few. Given this plethora of options which measure is the best? In order to answer this question we must go back to first principles and ask what exactly it is we want from a risk measure and what the ideal properties of such a measure should be. Only then can we make an informed comparison of different available measures. So what are they? According to the available academic literature a risk measure should have the following properties: 1) It should be Asymmetric 2) Relative to one or more benchmarks 3) Investor-specific 4) Multidimensional 5) Complete in a specific sense 6) Numerically positive 7) Non-linear I shall endeavour to explain some of these concepts in plain English so that you can decide for yourselves which properties you agree or disagree with. Asymmetry of risk deals largely with how we perceive risk. Given two potential investments marked A and B in the following example, most people would intuitively feel that B is riskier than A. This is because although they both have the same mean expected return of 10%, B has twice as much variability as A as denoted by its standard deviation of 10% B also appears to have more periods when its returns are below those of A and also when they are less than zero. The fact that this disquiets us suggests that we are more concerned about the potential downside of an investment than its upside hence our response to risk is asymmetric and so should the ideal risk-measure be. Relative to our budgeted or target rate of return For a sector or index fund the relevant sector or index. Some of these risk benchmarks could also be viewed as performance benchmarks except that falling below them is not simply disappointing but positively undesirable. Since investors have different liability profiles and or objectives and may use different risk benchmarks it is clear that the ideal risk measure needs to be flexible enough to be both investor specific and accommodate multiple benchmarks hence multidimensional. Having justified the first four desirable properties of a risk measure I will address the last three, Completeness, positivity and non-linearity by way of examples of those risk measures which fail to satisfy these requirements. One of the more attractive risk measures is the probability of shortfall. Clearly this is a number we are in general interested in. Unfortunately the probability

It is also clear that we perceive risk as being relative to something in this case A is perceived to be less risky than B but risk can be measured relative to a number of possible benchmarks. These include: In the case of a pension fund the value of the funds future liabilities. For those who abhor losses, relative to zero. For anyone trying to preserve their wealth relative to inflation. Relative to a default no-risk investment of having cash in the bank. Relative to some peer group or benchmark

of shortfall measure is not ideal because it is not complete. If we consider the case of an investor, who is concerned about losing capital relative to an important benchmark and is confronted with two hypothetical investment possibilities, E and F. Both have an expected return of zero relative to the benchmark and both have a probability of shortfall of 50%, but are they equally risky? If an investors only measure or risk is the shortfall probability then he/she will be indifferent between E and F. However we can see that F has a greater potential downside and that everywhere in the shaded area also a greater probability of realising that downside than E. Thus the shortfall probability measure, although interesting, does not address the issue of how severe an event may be. It is thus insufficient and incomplete. Similarly if we now use maximum shortfall as the only measure of risk using example F and G we can see although both have the same probability of shortfall of 50% and the same maximum shortfall of -30% it is not clear which is riskier because the maximum shortfall measure alone says nothing about the size of the typical shortfall. Two investments may have the same worst outcome but one may have many large losses and the other only a few. Information about the end point of the lower tail of a distribution says little about the distribution overall. Moreover, we typically have only a few data points with which to work in the tail making the maximum shortfall measure both numerically-illconditioned and incomplete as a risk measure.

One of the most widely used measures of risk, Standard Deviation or Volatility is not really a measure of risk at all but rather a measure of uncertainty. It is also particularly poorly suited for use as the ideal risk measure for the following reasons. If we consider investments A, C, and D in which both C and D have the same standard deviation as one another (10%) whilst A has a standard deviation of 5%. Using standard deviation as your sole measure of risk you would be indifferent between C and D. But this is clearly wrong since D has an average expected return of -10% versus Cs +10%. Many people object to standard deviation as a risk measure because it gives equal weight to deviations above the mean and deviations below the mean, whereas investors are likely to be more worried about downside deviation than upside deviation.

Another problem with using standard deviation as a risk measure is that it is not sensitive to order. In the below examples you can see that A and C have the same standard deviation.

However C is clearly riskier than A, having lost 38% of its value from its peak to trough during the hypothetical period shown. Markets that look, or feel, volatile often feel that way because of a distinct order of prices or returns: an order that involves choppy movements with frequent reversals. This kind of order dependent volatility is not captured by the technical definition of standard deviation, since standard deviation is not sensitive to order. This point has direct application to hedge fund investing, since many hedge fund managers employ trading strategies whose success or failure will be related not to the volatility of markets but to the path that markets follow

bad as losing half of your money. I dont think so, he says. Its at least 10 times as bad.

Why VaR is not a coherent measure of risk 1) Subadditivity For all random losses X and Y p(X)+p(Y) > p(X+Y) Scenario 1 2 3 4 5 6 7 8 9 10 VaR @ 85% p(X) 0 0 0 0 0 0 0 0 100 0 0 p(Y) 0 0 0 0 0 0 0 0 0 100 p(X+Y) 0 0 0 0 0 0 0 0 100 100

Value at Risk has garnered widespread acceptance in recent years as the new measure of risk. Despite this widespread use it is also not complete in that it is not mathematically coherent. In order to be mathematically coherent a risk measure must satisfy the conditions of: 1.) 2.) 3.) 4.) Subadditivity Monotonicity Positive Homogeneity and; Translation invariance.

0 100 0 + 0 is not > 100

Without going into detail on these, suffice it to say that Value at Risk fails to satisfy both the Subadditivity and Monotonicity conditions. This has two consequences. The first is that the sum of the parts may be less than that of the whole and secondly that the graphing of value at risk as a function of returns, as in the mean / value at risk frontier, may not result in a neat convex function. This makes finding the optimal point difficult using conventional methods. Fortunately there is one measure of risk, closely related to value at risk that is both mathematically coherent and complete. That is the Expected Shortfall measure aka. the conditional Value at Risk or Extreme Tail Loss. What the Expected Shortfall measure provides is a probability weighted average of the expected losses in excess of the value at risk. Hence it is the average of the tail losses conditional on the value at risk being exceeded. As a risk measure, Expected shortfall captures the whole of the downside portion of the relative probability density function and is complete. However, its one remaining failing is that it, like value at risk, is a linear measure of risk. The non-linearity of risk is closely related to investor psychology and utility. More people insure their homes than their pets despite the fact that the possibility of losing a pet is significantly higher than losing a home. This reflects the non-linearity of how we perceive risk. We perceive a low probability of experiencing a large loss as being far worse than a high probability of experiencing a small loss. Frank Sortino says VaR. Its simply a linear measure of risk. It says that losing all your money is twice as

2) Monotonicity If X < Y for each scenario then p(X) < p(Y) Scenario 1 2 3 4 5 6 7 8 9 10 E( r ) SD VaR E( r ) + 2 x SD X Y 1.00 5 2.00 5 3.00 5 4.00 5 5.00 5 5.00 5 4.00 5 3.00 5 2.00 5 1.00 5 3.00 5.00 1.41 0.00 5.83 5.00 5.83 is not < 5

3) Positive Homegeniety For all L > 0 and random losses X p(L X) = L p(X) 4) Translation Invariance For all random losses X and constant a p(X+a ) = p(X)+a

Sortino strongly holds the view that it is downside risk that is most important. According to this view, the most relevant returns are returns below the mean, or below zero, or below some other target or benchmark return. This has lead to a proliferation of measures of downside risk: semi-variance, shortfall probability, the Sortino ratio, etc. Ignoring the specific advantages and disadvantages of each individual candidate to represent the true nature of risk, we would offer two general observations: Frequency vs. Amplitude. The idea of risk as expected pain combines two elements: the likelihood of pain, and the level of pain. The measures described above (with the exception of Expected shortfall) focus on one or the other of these elements, but not both. Semi-variance (and its descendant, the Sortino ratio) focuses on the size of the negative surprises, but ignores the probability of those surprises. Shortfall probability focuses on the likelihood of falling below a target return, but ignores the potential size of the shortfall. If I were forced to pick a single quantitative measure of risk, it would offer the concept of expected return below the target, defined as the sum of the probabilityweighted below-target returns. This measure is essentially the area under the probability curve that lies to the left of the target return level. (Note that this definition is broad enough to cover both normal and non-normal distributions.) Other generalisations of this such as the LPMn measure and the Omega function which captures the full distribution are also available. The final criteria is given as numerical positivity although this is more a desirable than essential requirement. Personally I prefer to show loss percentages such as value at risk in negative terms since this is more intuitive, but it is more common to show them +ve because of the widespread use of quadratic penalties in scientific optimisation. Past performance does not guarantee anything regarding future performance, and past risk does not guarantee anything regarding future risk. This is true even when the historical record is long enough to satisfy normal criteria of statistical significance. The problem is that, just as a performance record is getting long enough to have statistical significance, it may no longer have investment significance. because the people and the organization may have changed. Investors should thus use the full toolbox of available risk measures but not loose sight of the wood for the trees.

References: A unified approach to upside and downside returns Leslie A Balzar (2001) Expected Shortfall, a natural coherent alternative to Value at Risk - Acerbi and Tasche (2001) Coherent measures of risk - Artzner and Delbaen (1999)

Peter Urbani, was most recently Chief Investment Officer for Infiniti Capital, Prior to that he was Head of Quantitative Research for Infiniti, Head of KnowRisk Consulting, Head of Investment Strategy for Fairheads Asset Managers, Head of Portfolio Management for Nexus Securities and Senior Portfolio Manager for Commercial Union Investment Management (CUIM) now part of AVIVA. He can be reached on +64 21 0255 2816 or at peter.urbani@gmail.com

Useful Calculations in Excel




Note Symmetric Measures Only

Normal istri u tion ( ro . ensity)


.

. . . . . . . . .

. r l ct si . . . . . . - . .

- .

- .

l r

ilit

s it

is

l ct

ilit

is - .

r =-(-

l *

-( rtf ll ( )/ i ti

( -

))*

T(

))

ct =-(- * ), , , si T(((( ^ * l = )* T(((

I T( I T(

( I ( -

), , ,T

))*

T(

))

= ((

I T( , I T( , , i ti - )^ I T( ,

, ,

,T

))*( ^ ))* ))/

^ ))I T( ,

,T

))

^ )* , ,

I T( , , )* * )

,T

) (

rtf ll r ilit I T( , t ti l -( rtf ll )*

,T

si =-(- * r =-((

))*

T(

))

I T(

,T

I T(

)*

si = / r t =(( ((

ti l

ti

I T( , , ,T I T( , , ,

)*( ^ )* ^ )*(

^ - * * - )))/

^ ))I T( ,

,T

)/

is - . is .

0% 1 ( #43 0 % )  (' & 1 ! " 

0 % ) (1 2 1 %    $   #  !          

rtf ll i ti l i ti rtf ll r ilit si t ti l r rtf ll si t ti l ti r t

. 7. 7 .9

Anda mungkin juga menyukai