Risk management

Managed Money

Sortino ratio:
A better measure of risk
By To m Ro l l i n g e r & S c o tt H o f f m a n

The purpose of risk management and subsequent risk measures is to avoid large downside
risk, so a performance metric that punishes excessive positive returns while missing
excessive downside risk is flawed. The Sortino ratio offers a better measure of risk.

M
any traders and investment Sharpe ratio
For more on risk measures,
managers want to measure The Sharpe ratio is a metric that aims go to futuresmag.com/Risk
and compare commod- to measure the desirability of an invest-
ity trading advisors (CTAs) or trading ment by dividing the average period ticularly poor performance metric when
systems. While there are many ways to return in excess of the risk-free rate by comparing positively skewed strategies
measure an investment’s performance, the standard deviation of the return like trend-following to negatively skewed
risk-adjusted returns are one of the generating process. Devised in 1966 by strategies like option selling (see “Bigger
most important measures to consider Stanford Finance Professor William F. winners vs. more winners,” page 42). In
because, given the inherent free leverage Sharpe as a measure of performance for fact, for positively skewed return distri-
of the futures markets, more return can mutual funds, it undoubtedly has some butions, performance actually is achieved
be earned by taking more risk. The most value as a measure of investment “qual- with less risk than the Sharpe ratio sug-
popular measure of risk-adjusted per- ity,” but it has a few limitations also. gests. Conversely, standard deviation
formance is the Sharpe ratio. While the The most glaring flaw is that it does understates risk for negatively skewed
Sharpe ratio definitely is the most widely not distinguish between upside and return distributions, i.e., the strategy
downside volatility (see “Good news, bad actually is more risky than the Sharpe
Positively skewed vs. negatively news,” right). In fact, high outlier returns ratio suggests. Typical long-term, trend-
skewed strategies can have the effect of increasing the value following CTAs, especially those with
Desired target return of the denominator (standard deviation) longer track records, generally have
more than the value of the numerator, Sharpe ratios in the 0.50 – 0.90 range.
used, it is not without its issues and limi- thereby lowering the value of the ratio. However, negatively skewed programs
tations. Because of the way the Sharpe For some positively skewed return distri- (convergent strategies) like option writ-
ratio is calculated, it tends to punish butions such as that of a typical trend-fol- ing will produce high Sharpe ratios, 3.0
upside volatility in a trading program. lowing CTA strategy, the Sharpe ratio can and above, up until a devastating draw-
We believe the Sortino ratio improves be increased by removing the largest posi- down. The Sharpe ratio often misses the
on the Sharpe ratio in a few areas. The tive returns. This is nonsensical because inherent risk of convergent strategies.
purpose of this article is to review the investors generally welcome large positive
Sortino ratio’s definition and present returns. Sortino ratio
how to calculate it properly, because Additionally, to the extent that the In many ways, the Sortino ratio is a bet-
we have seen its calculation often done distribution of returns is non-normal, ter choice, especially when measuring and
incorrectly. the Sharpe ratio falls short. It is a par- comparing the performance of managers

40 FUTURES Februar y 2013 DIGITAL EXCLUSIVE

whose programs exhibit positive skew in Good news, bad news
their return distributions. The Sortino
ratio is a modification of the Sharpe ratio, Large return swings are a sign of volatility and risk, but if a program consistently pro-
using downside deviation rather than stan- duces strong up moves with lower down moves, it should not be punished for those
strong moves in its favor.
dard deviation as the measure of risk — i.e.,
only those returns falling below a user-
Desired
specified target (“Desired Target Return”) Mean target return
or required rate of return are considered Sharpe ratio Sortino ratio
risky (see “Good news, bad news”). penalizes penalizes only
It is interesting to note that even both upside downside
and downside volatility below
Nobel laureate Harry Markowitz, when volatility. target return.
he developed Modern Portfolio Theory
(MPT) in 1959, recognized that because Downside Upside Downside
only downside deviation is relevant to volatility volatility volatility
investors, using it to measure risk would
be more appropriate than using stan- Return Return
dard deviation. However, he used vari-
ance (the square of standard deviation)
in his MPT work because optimizations
using downside deviation were compu- Target Downside Deviation = culation for Standard Deviation has
tationally impractical at the time. no Min() function.
In the early 1980s, Dr. Frank Sortino N
1
had undertaken research to come up
with an improved measure for risk-
N ∑ (Min (0, Xi - T))2 Standard deviation is a measure of dis-
persion of data around its mean, both
i=1
adjusted returns. According to Sortino, above and below. Target downside devia-
it was Brian Rom’s idea at Investment where tion is a measure of dispersion of data
Technologies to call the new measure Xi = ith return below some user-selectable target return
the Sortino ratio. The first reference N = total number of returns with all above target returns treated as
to the ratio was in Financial Executive T = target return underperformance of zero. Big difference.
Magazine (August 1980) and the first
calculation was published in a series of The equation for TDD is very similar Sortino ratio calculation
articles in the Journal of Risk Management to the definition of standard deviation: In this example, we will calculate the
(September 1981). annual Sortino ratio for a hypothetical
Standard Deviation = CTA with positive skew with the follow-
The Sortino ratio, S, is defined as: ing set of annual returns:
N
1
S=
R-T
TDD N ∑ (Xi - u)2 Annual Returns: 2%, 1%, –1%,
i=1 18%, 8%,–2%, 1%, –1%
where Target Return: 0%
• R is the average period return; where
• T is the target or required rate of return Xi = ith return Although in this example we use a tar-
for the investment strategy under con- N = total number of returns get return of 0%, any value may be select-
sideration (originally T was known as u = average of all Xi returns. ed, depending on the application, i.e., a
the minimum acceptable return, or futures trading system developer compar-
MAR. In his more recent work, MAR The differences are: ing different trading systems vs. a pension
is now referred to as the Desired Target 1) In the target downside deviation cal- fund manager with a mandate to achieve
Return). culation, the deviations of Xi from 8% annual returns. Of course using a dif-
• TDD is the target downside deviation. the user selectable target return are ferent target return will result in a differ-
measured, whereas in the Standard ent value for the target downside devia-
The target downside deviation is Deviation calculation, the devia- tion. If you are using the Sortino ratio to
defined as the root-mean-square, or tions of Xi from the average of all Xi compare managers or trading systems,
RMS, of the deviations of the realized is measured. you should be consistent in using the
return’s underperformance from the 2) In the target downside deviation cal- same target return value.
target return where all returns above the culation, all Xi above the target return First, we will calculate the numerator
target return are treated as underperfor- are set to zero, but these zeros still are of the Sortino ratio, the average period
mance of 0. Mathematically: included in the summation. The cal- return minus the target return:

DIGITAL EXCLUSIVE futuresmag.com 41

Managed Money continued

Bigger winners vs. more winners the Sortino ratio, and in particular the
target downside deviation, calculated
The Sharpe ratio assumes a normal distribution and tends to give a false sense of incorrectly. Most often, we see the tar-
security to negatively skewed strategies that tend to produce consistent small positive get downside deviation calculated by
returns but can be punctuated by rare, painful large negative returns.
“throwing away all the positive returns
and taking the standard deviation of
Mean Mean
negative returns.” We hope that by read-
ing this article, you can see how this is
incorrect. Specifically:
In Step 1, the difference with respect
Typical trend-following Typical option-writing
program
to the target level is calculated, unlike
CTA program
the standard deviation calculation
where the difference is calculated with
respect to the mean of all data points. If
every data point equals the mean, then
the standard deviation is zero, no matter
what the mean is. Consider the follow-
Positive skew Negative skew ing return stream: [–10, –10, –10, –10].
The standard deviation is 0; while the
Same Sharpe ratio, much different risk
target downside deviation is 10 (assum-
ing target return is 0).
In Step 3, all above target returns are
Average annual return – Target return 3) T hen, calculate the average of all included in the averaging calculation.
= 3.25% – 0% = 3.25% squared differences determined in The above target returns set to 0% in
Step 2. Notice that we do not “throw Step 1 are not thrown away.
Next, we will calculate the target down- away” the 0% values. The Sortino ratio takes into account
side deviation: both the frequency of below-target
1) F or each data point, calculate Average = (0% + 0% + 0.01% + 0% + returns as well as the magnitude of
the difference between that data 0% + 0.04% + 0% + 0.01%) / 8 = 0.0075% them. Throwing away the zero under-
point and the target level. For data performance data points removes the
points above the target level, set 4) T hen, take the square root of the ratio’s sensitivity to frequency of under-
the difference to 0%. The result of average determined in Step 3. This performance. Consider the following
this step is the underperformance is the target downside deviation underperformance return streams:
data set. used in the denominator of the [0, 0, 0, –10] and [–10, –10, –10, –10].
Sortino ratio. Throwing away the zero underperfor-
min(0, 2% – 0%) = 0% mance data points results in the same
min(0, 1% – 0%) = 0% Target Downside Deviation = target downside deviation for both
min(0, –1% – 0%) = –1% Square root of 0.0075% = 0.866% return streams, but clearly the first
min(0, 18% – 0%) = 0% return stream has much less downside
min(0, 8% – 0%) = 0% Finally, we calculate the Sortino ratio: risk than the second.
min(0, –2% – 0%) = –2% In this article we presented the defini-
min(0, 1% – 0%) = 0% Sortino Ratio = tion of the Sortino ratio and the correct
min(0, –1% – 0% ) = –1% 3.25% / 0.866% = 3.75 way to calculate it. While the Sortino
ratio addresses and corrects some of the
2) Next, calculate the square of each This is a strong score and indicative weaknesses of the Sharpe ratio, we feel
value in the underperformance data of the return stream from which we cal- there is one measure that is even better
set determined in Step 1. culated it. Calculating the Sharpe ratio yet: The Omega Ratio. We look forward
on the same set of returns would have to tackling the Omega Ratio in our
0% ^ 2 = 0% produced a Sharpe ratio (0% RFR) of next article.
0% ^ 2 = 0% 0.52, a mediocre one that indicates more
–1% ^ 2 = 0.01% volatility by penalizing the outsized Tom Rollinger is director of new strategies
0% ^ 2 = 0% positive returns. development for Sunrise Capital Partners.
0% ^ 2 = 0% Previously he was a portfolio manager for
–2% ^ 2 = 0.04% Sortino vs. Sortino quantitative hedge fund legend Edward O.
0% ^ 2 = 0% Often in trading literature and trad- Thorp. Scott Hoffman is the founder of CTA
–1% ^ 2 = 0.01% ing software packages we have seen Red Rock Capital Management.

42 FUTURES Februar y 2013 DIGITAL EXCLUSIVE