You are on page 1of 72

Table of Contents

Introduction .............................................................................................................................. 4 I. Using Statistics to Understand Return Characteristics ......................................................... 5 Using Standard Deviation to Predict Possible Return Ranges .............................................. 6 Assessing Skewness and Kurtosis in the Return Distribution ............................................... 8 Predicting Returns using Monte Carlo Simulation............................................................... 10 II. Risk Statistics and Risk-adjusted Statistics ....................................................................... 12 Standard Deviation .............................................................................................................. 12 Sharpe Ratio ........................................................................................................................ 14 Sortino Ratio ........................................................................................................................ 15 Omega Ratio ........................................................................................................................ 16 Drawdown Analysis ............................................................................................................. 21 Calmar Ratio ........................................................................................................................ 23 Sterling Ratio ....................................................................................................................... 23 Comparing Risk Statistics and Risk-adjusted Statistics ...................................................... 23 III. Correlation and Regression Analysis ................................................................................ 24 The Correlation Coefficient (R)............................................................................................ 24 Alpha and Beta..................................................................................................................... 23 The Coefficient of Determination (R2) ................................................................................ 23 Benchmark Ratios ................................................................................................................ 24 IV. Peer Group Analysis........................................................................................................... 26 Top Quartile Performance ................................................................................................... 26 Bottom Quartile Performance .............................................................................................. 27 Manager Search Criteria ...................................................................................................... 27 V. Composite Returns: Portfolio Construction, Optimization, Simulation .............................. 31 Portfolio Construction ......................................................................................................... 31 Optimization ........................................................................................................................ 36 Simulation ............................................................................................................................ 38 VI. Fat Tail Analysis, Risk Budgeting, Factor Analysis & Stress Testing ................................. 40 Fat Tail Analysis ................................................................................................................... 40 VaR (Value at Risk) .......................................................................................................... 40 The Differences between Normal VaR, Modified VaR and Fat-Tailed VaR ................... 40 ETL (Expected Tail Loss) .......................................................................................... 41 ETR (Expected Tail Return) ....................................................................................... 41 STARR Performance .................................................................................................. 41 Rachev Ratio ............................................................................................................. 41 Marginal Contribution to Risk (MCTR) / Marginal Contribution to Expected Tail Loss (MCETL) ............................................................................................................................ 41 Percentage Contribution to Risk (PCTR) / Percentage Contribution to Expected Tail Loss (PCETL) .................................................................................................................... 41
www.evestment.com | 1

Skew .......................................................................................................................... 41 Excess Kurtosis ......................................................................................................... 41 Implied Return .......................................................................................................... 42 Risk Budgeting..................................................................................................................... 43 Factor Analysis & Factor Contribution to Risk ..................................................................... 45 Stress Testing ...................................................................................................................... 47 Conclusion ............................................................................................................................... 48 Appendix I: Key Investment Statistics ................................................................................... 49 I. Absolute Return Measures ............................................................................................... 50 1. Monthly Return (Arithmetic Mean): ............................................................................. 50 2. Average Monthly Gain (Gain Mean): ............................................................................ 50 3. Average Monthly Loss (Loss Mean): ........................................................................... 50 4. Compound Monthly Return (Geometric):.................................................................... 51 II. Absolute Risk-adjusted Return Measures ...................................................................... 52 1. Sharpe Ratio:............................................................................................................... 52 2. Calmar Ratio: A return/risk ratio. ............................................................................... 52 3. Sterling Ratio: ............................................................................................................. 53 4. Sortino Ratio: .............................................................................................................. 53 5. Omega: ........................................................................................................................ 53 III. Absolute Risk Measures ................................................................................................ 54 1. Monthly Standard Deviation: ...................................................................................... 54 2. Gain Standard Deviation: ............................................................................................ 54 3. Loss Standard Deviation: ............................................................................................ 55 4. Downside Deviation: ................................................................................................... 55 5. Semi Deviation: ........................................................................................................... 56 6. Skewness: ................................................................................................................... 56 7. Kurtosis: ...................................................................................................................... 57 8. Maximum Drawdown: ................................................................................................. 58 9. Gain/Loss Ratio:.......................................................................................................... 58 IV. Relative Return Measures .............................................................................................. 59 1. Up Capture Ratio: ........................................................................................................ 59 2. Down Capture Ratio: ................................................................................................... 59 3. Up Number Ratio: ........................................................................................................ 60 4. Down Number Ratio: ................................................................................................... 61 5. Up Percentage Ratio (Proficiency Ratio): ................................................................... 61 6. Down Percentage Ratio (Proficiency Ratio): .............................................................. 62 V. Relative Risk-adjusted Return Measures ....................................................................... 63 1. Annualized Alpha: ....................................................................................................... 63 2. Treynor Ratio: ............................................................................................................. 63 3. Jensen Alpha: .............................................................................................................. 64 4. Information Ratio: ...................................................................................................... 64 VI. Relative Risk Measure ................................................................................................... 65 1. Beta: ............................................................................................................................ 65 VII.Tail Risk Measures ......................................................................................................... 66 1. Value at Risk (Parametric VaR): ................................................................................. 66
www.evestment.com | 2

2. 3. 4. 5. 6. 7.

Modified Value at Risk: ............................................................................................... 66 Expected Tail Loss (ETL): ............................................................................................ 66 Modifed Expected Tail Loss (ETL): .............................................................................. 66 Jarque-Bera: ................................................................................................................ 67 STARR (Stable Tail Adjusted Return Ratio): ............................................................... 67 Rachev Ratio: .............................................................................................................. 67

About eVestment ..................................................................................................................... 68

www.evestment.com | 3

Introduction
The purpose of this Guide is to assist you in gaining a better understanding of how to derive meaningful conclusions from investment statistics. A glossary of the key investment terms used in this Guide is provided in Appendix I at the end of the document. Guides Learning Objectives The Guides learning objectives are as follows: 1. 2. 3. 4. Explain how to use statistics to predict future investment returns. Interpret the different investment risk statistics and risk-adjusted statistics. Explain the concepts of correlation and regression analysis for investment analysis. Describe the key characteristics of peer group analysis and its usefulness in the search process.

www.evestment.com | 4

I. Using Statistics to Understand Return Characteristics


Investment statistics can be used in two ways: To compare the performance histories of multiple investment managers To try to predict a range of future returns for an investment.

A. Comparing Managers Track Records and Time Periods When using statistics to predict an investments return, it is critical to note that the length of the investments track record and the time frame will dramatically affect the calculations. For example, the average annual return of the S&P 500 Index, over the 36 -year period from January 1975 to June 2011, was 11.84%. However, as Figure 1 highlights, if we assess the same data using only 1- or 3-year rolling returns, they range between 61% and -43% on a 1-year rolling basis, and between 33% and 16% on a 3-year rolling basis.

www.evestment.com | 5

Similarly, if we assess a 5-year rolling period, the returns range between -3% and 20%. As Figure 2 illustrates, only when we lengthen the period to 10 years do we begin to see a true reversion to the mean, or a narrowing of the spread of actual returns close to the long-term average or mean return. This means that if an investment has a 20% return one year, per- haps the best prediction for its return the next year is I dont know. A one-year time period doesnt provide sufficient information from whi ch to draw conclusions. Therefore, investors should not rely exclusively on statistics that only cover 1-, 3- or even 5-year periods, since they may not be significant or meaningful over the long term.

B. Understanding Investment Return Characteristics In this section, we review some of the methods and statistics used to predict investment returns, including standard deviation, skewness and kurtosis, and Monte Carlo simulation. Using Standard Deviation to Predict Possible Return Ranges Can we use historical returns to predict future investment returns? As you can see with Figure 2, despite all of our carefully analyzed averages of historical returns, the S&P 500 Index still experienced one of its worst returns ever in 2008.

To help us predict future returns, we can generate a range of probabilities for the expected returns using standard deviation as a mathematical measure of predictability, rather than using historical averages.
www.evestment.com | 6

Standard deviation enables us to generate a probable range of expected returns. To demonstrate this, we can assess the returns of the S&P 500 Index and develop a normal, bell-shaped distribution of returns for the Index. Figure 3 illustrates the distribution of monthly returns for the S&P 500 Index.

From Figure 3, we see that the mean monthly return for the S&P 500 Index is 1.04% for the period January 1975 to June 2011. If we try to predict next months return, based on this information alone, there is a 50% chance that the return will exceed 1.04%, and a 50% chance it will not achieve this return. As Figure 4 illustrates, there is a 75% chance that the next monthly return will be greater than 1.66%, according to the shaded area under the curve. While some might find this information beneficial, there are significant problems with relying too heavily upon standard deviation as a predictive statistic. Perhaps the biggest problem is that very few investments display a normal distribution.

www.evestment.com | 7

Assessing Skewness and Kurtosis in the Return Distribution When returns fall outside of a normal distribution, the distribution exhibits skewness or kurtosis. Skewness is known as the third moment of a return distribution and kurtosis is known as the fourth moment of the return distribution, with the mean and the variance being the first and second moments, respectively. (Variance is a statistic that is closely related to standard deviation; both measure the dispersion of an investments historical returns.) Ideally, investors should consider all four mo ments or characteristics of an investments return distribution. Skewness: Skewness measures the degree of asymmetry of a distribution around its mean. Positive skewness indicates a distribution with an asymmetric tail extending toward more positive values. Negative skewness indicates a distribution with an asymmetric tail extending toward more negative values. Kurtosis: Kurtosis measures the degree to which a distribution is more or less peaked than a normal distribution. Positive kurtosis indicates a relatively peaked distribution. Negative kurtosis indicates a relatively flat distribution. A normal distribution has a kurtosis of 3. Therefore, an investment characterized by high kurtosis will have fat tails (higher frequencies of outcomes) at the extreme negative and positive ends of the distribution curve. A distribution of returns exhibiting high kurtosis tends to overestimate the probability of achieving the mean return.

Figure 5 illustrates both the skewness and kurtosis in the return distribution for the S&P 500 Index from Figure 4. The skewness is negative, which tells us that the returns are negatively biased. Because kurtosis measures the steepness of the curve, we can tell that there is a steep curve by reviewing the kurtosis number. A kurtosis less than zero indicate a relatively flat distribution.

www.evestment.com | 8

Skewness and kurtosis are important because few investment returns are normally distributed. Investors often predict future returns based on standard deviation, but such predictions assume a normal distribution. An investments skewness and kurtosis measure how its distribution differs from a normal distribution and therefore provide an indication of the reliability of predictions based on the standard deviation. As Figure 6 highlights, two investments with very different distribution profiles can have the same mean and standard deviation. Therefore, it is useful to consider other methods for predicting returns.

Source: An Introduction to Omega, Con Keating and William Shadwick, The Finance Development Center, 2002 Table 1 summarizes the key characteristics of a return distribution.

www.evestment.com | 9

Predicting Returns using Monte Carlo Simulation One method that can be used to predict returns is Monte Carlo simulation. Monte Carlo simulation is a method of generating thousands of series representing potential outcomes of possible returns, drawdowns, Sharpe ratios, standard deviations and other investments statistics of a specific investment or portfolio. The simulation calculates the uncertainty of a portfolios returns given its range of potential returns. Software that uses this simulation method can assess the probability of an individual achieving a retirement objective (and/or other financial objectives), given an investment portfolios specific asset allocation. Monte Carlo simulation using a bootstrapping technique allows for both skewness and kurtosis to be preserved. The bootstrapping technique involves resampling the actual data rather than assuming a normal distribution like standard deviation does. Monte Carlo simulations randomly construct a distribution of many possible returns for a portfolio over a specified time horizon. Thousands of possible results are calculated, and a probability profile is constructed for the various statistics. To see how this works, we can look at the stock market crash of 1987. From the period of January 1975 to August 1987, the largest drawdown for the S&P 500 Index was -16.52%, and the average return was 19.45%. Based on these numbers, few investors would have anticipated the crash of October 1987. However, using Monte Carlo simulation, we can see that there was the possibility of a market crash even in August 1987. Figure 7 shows the results of 10,000 Monte Carlo simulations on the S&P 500 Index. Note that the 99th percentile indicates a possibility of a 28.83% drawdown. This percentile indicates that, however remote, there is the possibility of a significant drawdown, one which historical returns and standard deviations do not predict.

www.evestment.com | 10

Figure 8 shows the results of a Monte Carlo simulation that was run as of 6/30/2011. Each bar represents the range of worst potential returns which have a 10% probability of occurring. As can be seen in the chart, from 1975 to 2001 the S&P 500 Index never had a 3 year to 10 year period that fell within the range. However, as of June 2011, the S&P 500 Index had experienced its worst performance in over a 25 year history. This example indicates that although there may be a discrete probability that an event might occur, it does not specify exactly at which time it will occur.

www.evestment.com | 11

II. Risk Statistics and Risk-adjusted Statistics


Many investors approach manager selection and analysis with pre-conceived statistical prejudices based on a misunderstanding of statistics. In many cases, it is because theyve been led to believe that a certain statistic measures something that it does not. Others encounter difficulties trying to use a pre-defined toolkit of investment statistics because theyve been led to believe that those are the right statistics to choose. It is important to remember, however, that investors have different notions of risk. To some, risk is the uncertainty of achieving an expected return. To others, it is not achieving a minimal acceptable return (MAR). Still others define risk as flat-out losing money. To illustrate this point, lets look at how many investors use standard deviation to help them identify strong investments. Standard Deviation Investors sometimes begin a quantitative screening by stating that they want a fund with a low risk. Because of the historical ties between risk and standard deviation in the world of traditional investments, they equate high standard deviation with high risk, and then use standard deviation as a comparative statistic. However, in truth, standard deviation is merely a statistic that measures predictability. A high standard deviation means that the fund is volatile, not that the fund is risky or will lose money, while a low standard deviation means a fund is generally consistent in producing similar returns. A fund can have extremely low standard deviation and lose money consistently, or have high standard deviation and never experience a losing period. For example, without looking at the returns the fund in Figure 9 exhibits a return pattern with overall consistency, which results in a low annualized standard deviation of 3.8%.

www.evestment.com | 12

Is the fund in Figure 9 a good investment? If we assess the same chart with returns plotted on the xaxis, the exact opposite is true. As Figure 10 highlights, this fund, while maintaining a low standard deviation, has a compound annual return of less than 1% (see circled area), and the fund has lost money almost as often as it has generated profits.

Assessing funds based on standard deviation also tends to unjustly penalize funds with high upside volatility. The fund in Figure 11 has a standard deviation of 22.5%, which is generally considered high. However, the monthly returns are skewed to the upside as the result of several months of 15+% returns (see circled area).

www.evestment.com | 13

One of the main differences between traditional return analysis and absolute return analysis is accepting the fact that volatility is good, provided it is on the upside. Indeed, most investors should be less concerned with upside volatility, and consider d ownside deviation as a better measure of a funds ability to achieve its return goal. For this reason, investors should acquaint themselves with downside deviation. Downside deviation introduces the concept of minimum acceptable return (MAR) as a risk factor. If a retirement plan has annual liabilities of 8%, the plans real risk is not earning 8% not whether it has a high or low standard deviation. Downside deviation considers only the returns that fall below the MAR, ignoring upside volatility. As Figure 12 illustrates, if the MAR is set at 8%, downside deviation measures the variation of returns below this value.

So, with standard deviation out of the equation, what statistics can we use to compare funds? While fund returns may seem useful, they do not consider the investments risk. Therefore, investors should always use risk-adjusted statistics such as the Sharpe, Sortino, Sterling or Calmar ratios. Sharpe Ratio The Sharpe ratio is the best-known risk-adjusted statistic. You calculate an investments Sharpe ratio by taking the average period return, subtracting the risk-free rate, and dividing it by the standard deviation for the period.

This calculation generates a number we can use to compare investments. Note that for meaningful comparisons, all comparative investment statistics must be calculated over the same time period.

www.evestment.com | 14

Example: Lets compare two investments, Fund A and Fund B. Fund A has a return of 10% and standard deviation of 8%, while Fund B has a return of 20% and standard deviation of 16%. If the risk-free rate is 4%, Fund A has a Sharpe ratio of 0.75, and Fund B has a Sharpe ratio of 1.0.

Comparing the Sharpe ratios, Fund B would have been the better investment because: If we invested $1,000,000 in Fund A, with a 10% return, we would have $1,100,000 after 1 year. If we invested $500,000 in Fund B, with a 20% return, and $500,000 in a bank account with a 4% return, we would have $1,120,000 after 1 year. This is called de-leveraging.

According to Sharpe, a higher standard deviation is not bad, provided it is accompanied by a proportionally higher return. Note that there is no such thing as a good or bad absolute number for a comparative investment statistic, only its relative relationship to other peers. Also note that in real world investing, investors do not, in fact, de-leverage. Sortino Ratio Since upside volatility will decrease the Sharpe ratio of some investments, the Sortino ratio can be used as an alternative. The Sortino ratio is similar to the Sharpe ratio; however it uses downside deviation instead of standard deviation in the denominator of the formula, as well as substituting a minimum acceptable return for the risk free rate. In other words, the Sortino ratio equals the return minus the MAR, divided by the downside deviation.

Table 2 highlights the difference between the Sharpe and Sortino ratios using the S&P 500 Index and the Barclays Aggregate Bond Index.

We can see from Table 2 that bonds have a somewhat higher Sharpe ratio than stocks (0.58 vs. 0.45). However, if our goal is to achieve a MAR of 10%, the Sortino ratio favors stocks (0.09). For lower MARs, the Sortino ratio favors bonds.

www.evestment.com | 15

Omega Ratio The omega ratio is a relative measure of the likelihood of achieving a given return, such as a minimum acceptable return (MAR) or a target return. The higher the omega value, the greater the probability that a given return will be met or exceeded. Omega represents a ratio of the cumulative probability of an investments outcome above an investors defined retur n level (a threshold level), to the cumulative probability of an investments outcome below an investors threshold level. The omega concept divides expected returns into two parts gains and losses, or returns above the expected rate (the upside) and those below it (the downside). Therefore, in simple terms, consider omega as the ratio of upside returns (good) relative to downside returns (bad). Omega Ratio - The Omega Ratio is a measure of performance that doesnt assume a normal distribution of returns.

Where r is the threshold return, and F is cumulative density function of returns. There are several ways to estimate the risk of not achieving a given return, but most of them assume that returns are normally distributed. However, as stated above, investment returns are not normally distributed, as they tend to be skewed or fat-tailed (i.e., there are more extreme returns than implied by the theoretical normal distribution). The omega calculations are important as they use the actual return distribution rather than a theoretical normal distribution. Thus the omega ratio and its components more accurately reflect the historical experience of the investment being measured. Since omega considers all information available from an investments historical return data, it can be used to rank potential investments in a manner specific to the investors threshold level. However, the omega decisions are not static for at least two reasons: 1. As return information is updated, the probability distribution will change and omega must be updated. 2. As an investors threshold level changes, the rankings among comparative investments may change. Therefore, omega allows investors to visualize the trade-off between risk and return at different threshold levels for various investment choices. Note that when the threshold is set to the mean of the distribution, the omega ratio is equal to 1.

www.evestment.com | 16

Figure 13a and 13b highlights the omega ratios for two different threshold levels (a 0% return and a 10% return) for the S&P 500 Index and a range of funds (F1 to F9). Therefore, if an investors minimum acceptable return (MAR) is 10% rather than 0%, the omega rankings among the investments will change. For example, for a threshold return above 0%, F1 has the highest omega ratio followed by F2 and F3, but for a threshold return above 10%, F8 has the highest omega ratio followed by F9.

www.evestment.com | 17

A key question to consider is how the Omega Ratio compares when ranking Omega Ratio relative to more commonly used Statistics. Table 3 summarizes the return data for Fund A and a mix of asset class benchmarks. The table is sorted on Sharpe Ratio, and the Omega Ratio has a 1% monthly return threshold. When looking at the rankings, Bonds appear to be a good choice when looking at Sharpe Ratio, however the Bonds have the lowest Omega Ratio at the 1% monthly return threshold. Fund A provides the highest Sharpe and Omega ratios and has the smallest drawdown.

www.evestment.com | 18

Because the Sharpe ratio is calculated from return data that has been averaged or annualized, the resulting ranking of the investments do not include higher levels of information specific to the shape of the distribution of the underlying return data. Therefore, it is reasonable to conclude that the observed differences in rankings are due to the higher levels of information contained in the Omega calculations. In effect, Omega as a risk-adjusted measure provides investors with additional information to better understand the risk/reward characteristics encapsulated within an investments historic returns. Figure 14 illustrates the omega ratios for five different investments (Funds A-E) for a 12-month holding period. The two selected thresholds are a minimum acceptable return (MAR) of 2% (dashed purple line) and a target return of 8% (solid purple line). If a 2% downside is an investors primary concern, then Figure 14 shows that there is considerable difference between the funds, as Fund A (the red line) has a much better chance of exceeding the downside (i.e. it has a higher omega ratio at 2%). However, there is less difference between the funds as far as earning a target return of 8%. This means that choosing between the funds should be based more on the downside risk than on the expected return. The 8% target is close to the crossover point of all the funds. The Omega ratio is a useful investment tool because it can be used in a compact way to show how different investment options relate to a target return and to a MAR.

www.evestment.com | 19

Figure 14 used a 12-month holding period, which is appropriate for short-term expectations. However, Figure 15 uses an investment horizon of 5 years (60-month holding period), and shows that downside risk is less of a consideration for this period. The main decision is the target return. Once again, it is essential to consider the specific time period when analyzing investment returns.

www.evestment.com | 20

Drawdown Analysis Drawdown analysis can be an excellent way to screen investments. A Maximum Drawdown is the maximum amount of loss from an equity high through the drawdown and back to the point the equity high is reached again. There could be many drawdowns over a given date range and will be listed starting with the maximum drawdown. Figure 16 illustrates the maximum drawdown over the period 122009 to 07-2011, although it is not the only drawdown. Maximum drawdowns being analyzed on numerous investments should be calculated over the same date range.

www.evestment.com | 21

In addition, it is important to remember that drawdowns are relative to return. The S&P 500 Index in Figure 17 exhibits drawdowns that would give most investors pause. However, if you chart the losses and time underwater versus the returns of the fund (Figure 18), some investors might find the fund worthy of additional consideration.

There are numerous reasons for a drawdown, including market stress, giving back part of unrealized profits after a large increase in equity, or just poor trading. From a quantitative perspective, however, it is important to analyze the reasons that caused a particular drawdown, and not exclude a fund based on just absolute numbers.

www.evestment.com | 22

The Calmar and Sterling ratios provide additional comparative information for a risk-adjusted assessment of drawdown analysis. Calmar Ratio: The Calmar ratio is the annualized return for the last 3 years divided by the maximum drawdown during these years.

Sterling Ratio: The Sterling ratio is the annualized return for the last 3 years divided by the average of the maximum drawdown (in absolute terms) in each of the preceding 3 years, less an arbitrary 10%. An extra 10% is subtracted from the drawdown as one assumes that all maximum drawdowns will be exceeded.

Comparing Risk Statistics and Risk-adjusted Statistics So, lets put all of our investment risk statistics and risk-adjusted statistics to work. Figure 19 summarizes the statistics of a fund, Fund C, versus the S&P 500 and Barclays Aggregate Bond Indices.

www.evestment.com | 23

III. Correlation and Regression Analysis


We have all heard investment managers discuss their low correlations, high alpha, and low beta, but what is the real meaning of these terms? All of them relate to how different investments react relative to one another. The correlation coefficient (R) measures the extent of linear association of one or more funds or indices. Alpha is a measure of value added by an investment relative to the market (i.e. an index) or to another investment. Beta is a measure of the volatility of an investment relative to the market (i.e. an index) or to another investment. The coefficient of determination (R2) measures how well the regression line fits the data. The R 2 is the only measure that attempts to be predictive. The regression line is a graph of the mathematical relationship between two variables.

In Figure 18, the regression line is the line of best fit drawn through a scatter plot, and represents a linear relationship between two investments. The Correlation Coefficient (R) The correlation coefficient can range from -1 to +1. As Figure 20 illustrates, a correlation value near +1 indicates a high positive correlation between two investments. If one investment has positive returns during a period, it is highly likely that the other investments returns will also be positive.

www.evestment.com | 24

As Figure 21 illustrates, if two investments have a correlation of -1, the investments are negatively correlated, so if one investment has a positive return one month, it is likely that the other investment will have a negative return for that month.

Finally, as Figure 22 illustrates, if an investment has a correlation approximately equal to 0, the investments are not correlated to one another, and move independently. If one investment is up, the other could be either up or down. If one investment is down, the other could be either up or down.

www.evestment.com | 22

Alpha and Beta Alpha and beta are also related to the regression line. As Figure 23 illustrates, alpha is the Y intercept of the regression line. In other words, if the benchmark returned 0%, the Y intercept would tell us what the investment could be expected to return. In Figure 23, the alpha is 2.23%. That means if the benchmark returned 0%, we would expect our investment to return 2.23%. Beta is the slope of the line and measures the volatility of a particular investment relative to the market as a whole. (Note: The market can be defined as any index or investment.) Beta describes an investments sensitivity to broad market movements. For example, in equities, the stock market (the independent variable) has a beta of 1.0. An investment with a beta of 0.5 will tend to participate in broad market moves, but only half as much as the overall market.

The Coefficient of Determination (R2) It is important to consider the coefficient of determination (R2) when evaluating a funds alpha or beta. The R2 measures how well the regression line actually fits the data. A high R2 value means that the regression line closely fits the data, as in the example in Figure 23 (R 2 = 0.9373). However, if R2 is low, the regression line does not fit the data, and any calculations based on regression line analysis, including alpha and beta, become less meaningful as the R2 value drops.

www.evestment.com | 23

As Figure 24 illustrates, the returns have little relation to the regression line, and R 2 is essentially 0. Drawing conclusions about a funds alpha or beta based on this particular regression analysis would yield no meaningful results.

Benchmark Ratios Benchmark ratios are also useful in evaluating investments. These ratios provide information in a single number about a funds performance relative to a benchmark. Obviously, the selection of an appropriate benchmark is critical. Active Premium: The simplest benchmark ratio is the active premium, which takes the funds annualized return, and subtracts the benchmarks annualized return to yield the funds gain/loss that is over/under the benchmark (i.e., the excess return). Positive active premium is good, while negative active premium is generally bad. In Table 4, the active premium of -1.31% tells us that, overall, the fund underperformed the index.

www.evestment.com | 24

Up and down capture ratios are also helpful when evaluating investments. Up Capture Ratio: The up capture ratio is a measure of a funds cumulative return when the benchmark was up, divided by the benchmarks cumulative return when the be nchmark was up. The greater the value, the better. Down Capture Ratio: The down capture ratio is a measure of a funds cumulative return when the benchmark was down, divided by the benchmarks cumulative retu rn when the benchmark was down. The smaller the value, the better. Up and Down Number Ratios: Up and down number values measure the percentage of time that an investment moves in the same direction as the markets. In Table 4, approximately 92% of the time the fund moves up and down with the benchmark. U nfortunately, this statistic doesnt yield much information about the funds relative outperformance, so we combine these numbers with the proficiency ratios (up/down market percentage ratios) to glean more information about the fund. Up Market Percentage Ratio: The up market percentage ratio is a measure of the number of periods that an investment outperformed the benchmark when the benchmark was up, divided by the number of periods that the benchmark was up. The larger the ratio, the better. In Table 4, the fund outperformed the index 54.5% of the time when the index was up. Down Market Percentage Ratio: The down market percentage ratio is a measure of the number of periods that an investment outperformed the benchmark when the benchmark was down, divided by the number of periods the benchmark was down. The larger the ratio, the better. In Table 4, the fund outperforms the benchmark only 30% of the time when the benchmark was down. This would not be considered strong performance for a hedge fund manager.

www.evestment.com | 25

IV. Peer Group Analysis


Peer group analysis is the final piece of the basic investment quantitative toolkit. It allows investors to see how a particular fund ranks over various periods compared to funds using the similar investment strategies. Top Quartile Performance The fund in Figure 25 ranks in the top quartile of its peers over all of the trailing periods measured. This means that, out of all of the funds that manage money in the same investment strategy, the selected manager ranks in the top 25% of his/her peer group. If we assume an investment universe of 100 funds, this manager would rank within the top 25 funds in terms of recent returns for that investment strategy. You can perform peer group analysis on numerous statistics, including compound annual return, drawdown, Sortino ratio, Sharpe ratio, percent profitable months, etc. Managers who consistently rank high among their peers in a number of statistical categories are generally considered better managers.

www.evestment.com | 26

Bottom Quartile Performance Figure 26 shows the same fund measured against its peers based on calendar years instead of recent periods. The fund ranks in the bottom quartile on more than one occasion. This means that, out of all of the funds that manage money in the same investment strategy, the selected manager ranks in the bottom 25% of his/her peer group. If we assume an invest- ment universe of 100 funds, this manager would rank within the bottom 25 funds in terms of calendar year returns for that investment strategy. Although the fund looked good based on trailing periods (Figure 25), Figure 26 illustrates that there had been no consistency in top performance over the years.

Peer group analysis is important for a number of other reasons. Perhaps its best use is in screening data to find new investment managers. All too often, investors let arbitrary wish lists govern their screens. Seeking a manager with a 3-year track record, no losing months, an annualized return greater than 15%, and a Sharpe ratio greater than 2 does not guarantee that such a manager exists, or if they do exist, that they will be open to new investments. To select the best funds from a quantitative standpoint, it is best to think in terms of percentiles rather than absolutes, and herein lies the value of peer group analysis. Therefore, your goal should be to find the best fund manager comparatively, rather than search for an investment fantasy. Manager Search Criteria Assume you are searching for a hedge fund of funds (FOF) and require a MAR of 10%. You can complete the following 5-step process: Step 1 - The first step is to narrow the investment universe to include only FOFs. Starting with a large hedge fund universe comprised of all strategies we searched for FOFs, narrowing our hedge fund universe from approximately 1,700 funds to about 250 funds. Step 2 Select the Statistics for Screening: Using your new knowledge of statistics, select realistic
www.evestment.com | 27

statistics to use in your screens. For example, if youve developed an investment mandate that dictates you need high returns and are less concerned about drawdowns, you might use the compound annual return, rolling returns, and Sortino ratio (MAR 10%) rather than drawdown statistics. If losses are your paramount concern, you might consider the Sortino (MAR 0%) ratio, drawdown, Sterling and Calmar ratios. Step 3 Determine the Investment Period: To ensure you compare apples to apples, screen for funds with similar track records. Because the markets experienced a significant stress point in 1998, search for funds that started in or before January 1998. This track record narrows our sample to about 75 funds. Step 4 Select Funds that Meet the MAR: Next, we search for all funds with an annualized return greater than 10% (our MAR). These actions reduce our universe to 36 funds. Step 5 Rank the Funds by Percentiles and Assess Sharpe Ratios, Maximum Drawdowns, Sortino Ratios and Correlations: Rather than search all 36 funds for the wish list criteria, we first rank the funds by percentile to determine reasonable search characteristics. Since the required MAR is 10%, we do not rank on annualized return, so we first assess the Sharpe ratio, as the Sharpe measures risk vs. reward and a manager with a strong risk-reward profile is part of our mandate.

Next, we screen for maximum drawdown, because our particular investment mandate dictates that we want to limit drawdowns, even if it means limiting some upside potential. In Figure 28, we see that to be in the top quartile for maximum drawdown, funds must have lost less than -5.50% during their maximum drawdown.

www.evestment.com | 28

Then, we focus on the Sortino ratio to ensure we give downside deviation its due without sacrificing any upside that might be generated by a high standard deviation with good returns. As Figure 29 illustrates, to be in the top quartile for the Sortino ratio (MAR of 10%), funds must have a Sortino ratio greater than a 0.69.

www.evestment.com | 29

We then search our FOFs universe for those funds with a Sharpe ratio greater than 1.28, less than a 5.50% maximum drawdown, and a Sortino ratio greater than 0.69. In other words, we wish to find FOFs with top quartile performance in all criteria categories. As a result, we find 4 funds out of 36 that are quantitatively the cream of the crop, given our criteria and compared with their peers. Another reason that investors consider FOFs is to diversify their long-only portfolios. So a final step in our quantitative search might be to find FOFs with a negative correlation to the S&P. When applying this criterion, we reduce our list of 4 to 2. This same 5-step search process can be completed on virtually any universe of funds, including hedge funds, separately managed accounts, commodity trading advisors (CTAs) and mutual funds. Perhaps the most important part of the search process is to first establish reasonable search parameters. Otherwise, you risk setting the bar too high for any fund to hurdle, and backing down from there later. Also, recognize that peer group analysis is still important after an investment is made, since it provides one way to determine if the fund continues to offer the best investment option within a given investment strategy, or if other funds provide better options.

www.evestment.com | 30

V. Composite Returns: Portfolio Construction, Optimization, Simulation


An investment in a single fund has both market risk and manager risk. By constructing a portfolio of funds the manager risk can be minimized. Optimization allocates among the managers in the portfolio to maximize the return for the investors risk tolerance or minimize the risk for the investors desired return. Using a Monte Carlo simulation, the investor can predict the portfolios future returns or at least the likelihood of future returns. Portfolio Construction Whether a portfolio is constructed to provide exposure to the market or targeted exposure to a particular region or strategy investing in multiple funds will reduce the manager-specific risk. Figure 30 demonstrates the benefits of diversification using two asset classes, US Equities (S&P 500 Index) and Commodities (Barclays CTA Index), for the period January 1, 1995 through December 31, 1999.

US Equities and Commodities are not perfectly correlated. The lack of perfect correlation means that their gains and losses occur at different times. As a result, its less risky to be invested in a portfolio with exposure to both assets classes. How much exposure the investor should have to each asset class will be covered in the Optimization section. This same principle holds true when investing in individual funds that are not perfectly correlated. The more funds in the portfolio the more the risk created by an individual manager is reduced. The risk that cannot be eliminated by diversification is the systematic risk or common sources of risk to all managers in the market, strategy, region and sector. At some point, there is a limit to the diversification benefit provided by adding an additional fund. There is a cost, both in terms of portfolio performance and financially, to invest in an additional fund that may exceed the diversification benefit.
www.evestment.com | 31

To measure the benefits of a portfolio of funds a composite return series is created. The portfolios performance consists of the underlying funds performance, the allocations or weights to the underlying funds in the portfolio and the rebalancing schedule. For one month the calculation of the portfolio performance is:

Important factors in the calculation of portfolio performance are: Allocation (weight) Is the percentage of capital assigned to funds in the portfolio. Capital can be equally allocated across all funds in the portfolio, mandated by an investment policy or decided using optimization software to determine an ideal allocation. Leverage The investor can increase their position in an underlying fund by using leverage. For example, if the position is levered 2 to 1, we are taking a dollar of investor capital and borrowing another dollar to invest 2 dollars in the underlying fund, effectively doubling the position in the fund. Rebalancing When investing in private investments, the rebalancing schedule is typically never rebalanced due to the illiquid nature of the investments. Only a monthly rebalancing schedule will maintain the same starting allocations each month. The other rebalancing choices will allow the percent allocations to the underlying funds to change with the NAVs of the underlying funds until the portfolio is rebalanced. The common rebalancing schedules include: Never allocations are applied at the start date and the assets are allowed to grow. Never rebalance might be used with a buy and hold strategy, when invested in illiquid assets or while investing in funds with lockups which will not allow redemptions. Annual every 12 months the original allocations are applied. This rebalancing frequency might be used where the portfolio has target allocations that are maintained each year. Semi Annual every 6 months the portfolio returns to the original allocations. d. Quarterly every 3 months the portfolio returns to the original allocations. Monthly at the beginning of each month the portfolio resets to the original allocations. Monthly rebalancing is most commonly used when constructing a blended benchmark, e.g. 60 percent equities and 40 percent bonds.

www.evestment.com | 32

Manual any allocation can be given to any funds at any time in conjunction with other rebalancing schedules. Manual rebalancing represents a subscription in a new fund, additional subscription in a fund currently held in the portfolio and/or a partial or total redemption of an existing fund in the portfolio.

www.evestment.com | 33

When the portfolio is rebalanced, in effect, for the funds that did well a portion of their gains are being sold and these gains are being invested in the funds that had losses to bring the portfolio back to the original allocations. Figure 31 is an example of quarterly rebalancing. The portfolio starts January 1, 2011 with a capital balance of $10,000,000 equally allocated among four funds or $2,500,000 invested in each fund. On March 31, 2011 the funds allocations are no longer equal. Fund A = $2,643,555.25 Fund B = $2,601,364.11 Fund C = $2,468,360.71 Fund D = $2,554,751.27 Summing the four positions gives a portfolio value $10,268,031.34. For an equally allocated portfolio starting April 1, 2011, each funds position value should be $2,567,007.84. Portions of Fund A and Bs gains are being sold and invested in Fund C and D in order to bring the portfolio back to its original equal allocations. At the end of each subsequent quarter the portfolio is once again rebalanced.

www.evestment.com | 34

Lack of performance When constructing a pro forma portfolio some funds may have inception dates after the desired inception date of the portfolio, thus lacking several months of data. The lack of performance can be addressed by reallocating to other investments in the portfolio or by using backfill. Reallocating to other investments will proportionality redistribute the funds allocation to funds with performance until the fund does have performance. Backfilling adds returns from another fund or index to the fund with the later inception date. The following common backfill approaches are: Zero performance equivalent to the percent of capital invested in the fund not earning a return for the months where the fund lacks returns. Fixed Rate equivalent to the percent of capital invested in the fund without performance earning the interest in a savings account. Proxy Benchmark for the fund without performance, the performance of the selected benchmark would be used until the fund in the portfolio has performance. For example, in Figure 32 the portfolio has a start date of January 2004 and Fund C has inception date of April 2004. For the first quarter of 2004 Fund C is been backfilled with the performance of the HFRI Index, the proxy benchmark in this example.

www.evestment.com | 35

Optimization In an effort to improve the portfolios performance there are different methodologies to assist the investor in determining the optimal allocations to funds in the portfolio. Some of the most commonly used methods are linear optimization, Markowitz Efficient Frontier and Black-Litterman. Linear optimization provides options in addition to the traditional tasks of maximizing return and minimizing volatility. For example, linear optimization can be used to allocate among the funds to minimize the drawdown or downside risk of the portfolio. Markowitz Efficient Frontier is an intuitive and easy to use model that is best fit when the funds historically have had a normal distribution and the funds are expected to perform similarly in the future. However, linear optimization and Markowitz Efficient Frontier are models that rely on historical returns. When a funds future performance is believed to be different than the funds past performance t he Black-Litterman optimization may be a good choice due to the relative and absolute views used to enter expected performance. All of these optimization methodologies allow for minimum and maximum allocation constraints that keep allocations within investment policy requirements. Linear Optimizer as the name suggests, this optimization methodology is focused on a single performance or risk measurement to solve for the allocations to the underlying funds that maximizes the performance measurement or minimizes the risk measurement. Solvers calculate the selected statistics for the portfolio within the constraints set for the individual underlying assets. Then the allocations are changed and the selected statistic is calculated. Through multiple iterations an optimizer solves for the top portfolios for the selected criteria. Markowitz Efficient Frontier a mean variance optimization methodology that for a selected level of risk determines the highest return or for a target return determines the lowest risk by allocating among the underlying funds. Inputs needed to create optimal portfolios are the expected returns, variances for all assets and covariance between all of the underlying funds in the portfolio. The inputs are calculated from the funds historical returns. As such the start and end dates used for the optimization are important factors in the outcome of the optimization. Depending on the investment horizon for the portfolio it may be appropriate to use a subset of the available data to determine the optimal allocations. Black-Litterman like Markowitz Efficient Frontier, is a mean variance optimization methodology but allows the input of relative and/or absolute return assumptions. Market Capitalization of the asset classes in the portfolio is used to attempt to address over allocating to a single asset class. The Black-Litterman approach was originally used for global equities, bonds and currencies, where the market capitalization can fairly easily be determined. However, when using the approach with other asset classes the market capitalization might not be possible to ascertain. One suggestion for using Black-Litterman with a portfolio of hedge funds would be to use the funds AUM as a proxy for the market capitalization.

www.evestment.com | 36

Views allow assumptions about future expected performance to be used in lieu of the funds historical performance to determine the optimal allocation. The views can take two forms; relative and absolute. The absolute view is a forecast of a funds future performance. Figure 33 A relative view is a statement that a fund or group of funds will outperform another fund or group of funds. Figure 34

www.evestment.com | 37

Simulation Predicting Returns using Monte Carlo Simulation One method that can be used to predict returns is Monte Carlo simulation. Monte Carlo simulation is a method of generating thousands of series representing potential outcomes of possible returns, drawdowns, Sharpe ratios, standard deviations and other investments statistics of a specific investment or portfolio. The simulation calculates the uncertainty of a portfolios returns given its range of potential returns. Software that uses this simulation method can assess the probability of an individual achieving a retirement objective (and/or other financial objectives), given an investment portfolios specific asset allocation. Monte Carlo simulation using a bootstrapping technique allows for both skewness and kurtosis to be preserved. The bootstrapping technique involves resampling the actual data rather than assuming a normal distribution like standard deviation does. Monte Carlo simulations randomly construct a distribution of many possible returns for a portfolio over a specified time horizon. Thousands of possible results are calculated, and a probability profile is constructed for the various statistics. To see how this works, we can look at the stock market crash of 1987. From the period of January 1975 to August 1987, the largest drawdown for the S&P 500 Index was 16.52%, and the average return was 19.45%. Based on these numbers, few investors would have anticipated the crash of October 1987. However, using Monte Carlo simulation, we can see that there was the possibility of a market crash even in August 1987. Figure 35 shows the results of 10,000 Monte Carlo simulations on the S&P 500 Index, the "portfolio" in this example. Note that the 99th percentile indicates a possibility of a 28.83% drawdown. This percentile indicates that, however remote, there is the possibility of a significant drawdown, one which historical returns and standard deviations do not predict.

www.evestment.com | 38

Figure 36 shows the results of a Monte Carlo simulation that was run as of 6/30/2001. Each bar represents the range of worst potential returns which have a 10% probability of occurring. As can be seen in the chart, from 1975 to 2001 the S&P never had a 3 year to 10 year period that fell within the range. However, as of June 2004 the S&P had experienced its worst performance in over a 25 year history. This example indicates that although there may be a discrete probability that an event might occur, it does not specify exactly at which time it will occur.

www.evestment.com | 39

VI. Fat Tail Analysis, Risk Budgeting, Factor Analysis & Stress Testing
Fat Tail Analysis VaR (Value at Risk) - The highest possible loss over a certain period of time at a given confidence level. A 99% Value at Risk is interpreted as the level at which the losses of an asset or a portfolio will not be exceeded with a probability of 99% (i.e. in 99 out of 100 cases the analyzed asset or portfolio return will be above the estimated VaR value). Calculating VaR can be done using the historical fund data or parametrically fitting the data to a distribution and simulating the risk variables of this fund to create potential outcomes that did not occur in the past. The traditional way of calculating VaR parametrically utilizes the Normal distribution which only accounts for 2 moments, mean and standard deviation. It also assumes that the return distribution of risk variables is normally distributed despite ample empirical evidence against this assumption. Utilizing Normal VaR underestimates downside risk and can overestimate upside potential because the tails of these fund risk drivers are completely ignored. In addition, if the right tail is longer, the normal distribution may underestimate the upside potential and you run the risk of leaving money on the table. The Fat Tailed distribution helps you avoid this by accurately capturing both the left and right tails. To solve this problem, Fat -tailed VaR utilizes the Skewed Students distribution which accounts for higher moments including skewness and degrees of freedom (tail fatness) allowing for a more accurate picture of tail activity and asymmetrical behavior. The Differences between Normal VaR, Modified VaR and Fat-Tailed VaR - In commercial products, VaR is widely used in combination with the normal distribution. In order to deal with the shortcomings of Normal VaR (as described above), Modified Value-at-Risk (mVaR) is often used. The calculation of mVaR is not tied to a distributional assumption. It can be viewed as a non-parametric estimator of the empirical VaR, employing the first four moments computed from the observed returns (Mean, Standard Deviation, Skewness, and Kurtosis). Some in the industry consider mVaR to be a Fat-tailed VaR. This statement is only true to the extent that it distinguishes mVaR from the Normal VaR methodology based on the normal distribution. However, mVaR is a non- parametric estimator and while it has some of its own merits, it is not much different than any other non-parametric estimator and has many deficiencies including: It becomes less reliable for probabilities close to 0 or 1. That is, the deeper we go into the left or the right tail, the worse the approximation gets. In effect, VaR at high confidence levels get more inaccurate. Basically, VaR at 99% cannot be reliably calculated. It works well only for non-normal distributions which are close to the normal distribution and not for those which deviate significantly. As a result, it will not work well for distributions with high degree of skewness or fat tails.

Therefore associating mVaR with the term Fat-tailed VaR is inaccurate. Fat-tailed VaR is VaR based on a non-normal distributional assumption for risk factor returns. Recognizing that returns are fat-tailed and skewed one needs to explicitly assume that an appropriate non-normal distribution is necessary to model these properties. In this report, we use the Skewed Students t distribution to compute a true and more accurate fat-tailed VaR number.

www.evestment.com | 40

ETL (Expected Tail Loss) The average expected loss beyond VaR is also known as Conditional Value at Risk (CVaR) or Average Value at Risk (AVaR). It can be interpreted as the expected shortfall assuming VaR as a benchmark. ETL does not have the deficiencies of VaR as it is a true downside risk measure which can recognize diversification opportunities and has good optimality properties. ETL is a sub additive measure and therefore can be used to aggregate or decompose risk at the portfolio or strategy levels. This is why ETL is the risk measure used as the basis for risk budgeting. ETR (Expected Tail Return) ETR uses the same calculation as ETL, but refers to the positive side of the return distribution. STARR Performance Similar to the Sharpe Ratio which is a standard deviation-based performance measure, but STARR (stable tail-adjusted return ratio) uses the ETL in the denominator as a risk measure. STARR can be seen as a more effective indicator of risk-adjusted performance because it penalizes only for downside risk, while the standard deviation does not distinguish between upside and downside risk. Rachev Ratio The ratio of the ETR to the ETL. This ratio demonstrates the funds upside potential as measured by expected return in the right tail and expected loss in the left. This measure is similar to the Omega Ratio but is data adaptive because the confidence level is user defined (how far out into the tails to go). Marginal Contribution to Risk (MCTR) / Marginal Contribution to Expected Tail Loss (MCETL) These statistics show how much additional risk would be added to the portfolio if an additional 1% were to be invested in that specific manager. If the measure is positive for a portfolio fund, increasing the allocation by 1% to that fund would increase the portfolio's risk. If the measure is negative, increasing the allocation by 1% to that fund would decrease the portfolio's risk. Therefore, a negative MCTR/MCETL is a preferred characteristic for an investment. MCTR uses Standard Deviation as the risk measure; MC ETL uses Expected Tail Loss. Percentage Contribution to Risk (PCTR) / Percentage Contribution to Expected Tail Loss (PCETL) These statistics give information about the fraction of total risk contributed by each of the funds in the portfolio. They are computed by multiplying the weight of the investment by their marginal contribution stat (defined above) and dividing the total by the appropriate risk measure. PCTR uses Standard Deviation as the risk measure; PC ETL uses Expected Tail Loss. Percentage contributions to risk sum up to 100%. Skew This statistic represents the shape of the funds distribution around the mean. A negative skewness measure indicates that the distribution is skewed to the left - (i.e. compared to the right tail, the left one is elongated.) The opposite conclusion is drawn in the case of positive values. Excess Kurtosis The measure of peak and tail fatness in the distribution. The tails of a probability distribution contain additional information about the extreme values a random variable can take. The value is the excess over the standard normal distribution, for which the kurtosis is 3. Positive values signal more weight around the mean and fatter tails relative to the normal distribution.

www.evestment.com | 41

Implied Return - This represents the return a fund must deliver in order to justify the amount of risk it contributes to the overall portfolio. In economic terms, the implied return can be seen as the hurdle rate of the fund given its risk profile. Implied return can either use volatility or ETL as its risk measure. When using ETL Implied Return is equal to the STARR optimal portfolio (tail risk-adjusted return optimal) multiplied by each funds marginal contribution to ETL

www.evestment.com | 42

Risk Budgeting Risk budgeting involves looking at individual fund risk and return contributions and then reallocating to maximize portfolio performance. Risk budgeting is a powerful technique because it accounts not only for individual fund performance but also for interaction effects within the portfolio stemming from the dependency structure of the funds. Implied returns are the result of reverse engineering an optimal portfolio. The portfolio is optimal in the mean-ETL sense, which signifies that the portfolio is using implied returns based on tail loss. Implied returns represent the return a fund must deliver in order to justify the amount of risk it contributes to the overall portfolio. In economic terms, the implied return can be seen as the hurdle rate of the fund given its risk profile. In the maximum STARR portfolio, implied returns and mean returns are equal. It follows that whenever there is a discrepancy between mean or expected returns and implied returns, there is room for improvement. The reallocation rule is: allocate to those funds for which mean return exceeds implied return; and decrease allocations to funds for which implied return exceeds mean return. Following these guidelines will improve the risk adjusted performance of your portfolio. Risk budgeting is a useful tool because it allows you to incorporate your knowledge of a funds liquidity and capacity in your portfolio thereby allowing you to make realistic allocation choices that fit into your current investment policy. These are just some considerations why risk budgeting may be more useful compared to the pure out of the box optimization approach. We usually recommend the use of optimization as a guideline and then reallocating the portfolio based on a well-constructed risk budgeting process. There are two ways to quickly select the funds with the best risk budgeting potential. In the data table there is a Difference column that shows the difference between Mean Return and Implied Return. The higher the number, the more this fund justifies its risk and can be considered a good candidate for increased allocation. Funds with negative differences may be considered redemption candidates. In the table below, Fund_16 would need to have a monthly mean return of 5.58 in order to justify its tail risk. However it only returns 3.47 per month (difference of -2.11), meaning that this fund may be considered for redemption or decreased allocation. On the other hand, Fund_78 only needs to have a monthly mean of 0.5 to justify its tail risk. However since it has a mean of 1.4 (difference of 0.9) you may consider an increased allocation to this investment.

www.evestment.com | 43

The second way to view this is by using the chart below.

This chart provides a visual depiction of the mean return versus implied return concept. The diagonal line represents the STARR optimal portfolio (best tail risk-adjusted return ratio for the entire portfolio). Points above this line (i.e. point 14/ Fund_7) represent funds that have higher mean returns than implied returns meaning that their return to the portfolio more than justify the tail risk that they add. These funds provide you with good allocation opportunities. Points below the line are underperforming relative to the tail risk they are contributing.

www.evestment.com | 44

Factor Analysis & Factor Contribution to Risk The purpose of factor analysis is to uncover the relationship between one dependent variable (typically a fund) and one or more explanatory variables (typically market or style indices) by using a statistical estimation method. The Stepwise regression method is a selection process that utilizes the Akaike Information Criterion (AIC for short) in order to select the factors that best explain the variance of each fund in your portfolio. Most stepwise regression methods in standard software packages either deploy a forward or a backward looking selection process. Forward selection: Involves starting with no variables in the model, trying out the variables one by one and including them if they are 'statistically significant'. Backward elimination: Involves starting with all candidate variables, testing them one by one for statistical significance, and deleting any that are not significant.

Its important to note that your system should utilize both, testing at each stage for variables to be included or excluded for the final model. The output of the calculation is always a set of coefficients which describes the linear dependency between the dependent and the explanatory variables.

www.evestment.com | 45

The R statistic measures the overall goodness of fit of the model. It shows you how much of the total variance of the fund can be accounted for by the factors. R can take values between 0 and 1. R value of 1 means that you have perfect fit of the model. There are no set rules for the range of R that would be indicative of a good model. In theory, hedge funds are supposed to be uncorrelated to the market in general, so finding a good model by using market factors should be impossible. However, this theory rarely holds since for most hedge funds, you will be able to find a set of factors that adequately captures the fund profile. In general, an R above 0.7 is considered a good fit. An R of 0.4 - 0.7 is an adequate fit and an R below 0.4 is a poor one. The magnitude of the R statistics varies across hedge fund strategies. An R of 0.38 may be considered a poor fit for a directional strategy like long/short equity. But it can be a very reasonable fit for a relative strategy like fixed income arbitrage. Factor contribution to risk gives you insight into the breakdown of risk by its systematic and specific parts. It allows you to see the exposure of the portfolio to market factors and the contribution of those factors to the risk of the portfolio. The risk measure for this section is standard deviation. In other words, the provided information is about factor contribution to portfolio volatility. This section provides an analysis of the total risks of the portfolio broken down into two components: Systematic Risk the risk attributable to those factors to which the portfolio has exposure Specific Risk the risk which cannot be attributed to market factors.

The included chart graphically shows this breakdown in percent of systematic and specific risks of the portfolio as well as the breakdown of the systematic portion of risk across all factors used in the model. The percentage of specific risk is represented by the blue bar on the chart. The red bars represent the percentage contribution to portfolio risk coming from each of the factors which form the systematic portion of risk. You can also view this in data format.

www.evestment.com | 46

Stress Testing Portfolio value stress tests allow you to examine the effect of various scenarios on the value of your portfolio. You can use stress tests to investigate how the value of your portfolio would change under a hypothetical scenario or under a historical crisis that actually occurred in the past. For example, a single stress test will look at the performance of various indices during a specific period of time (i.e. during the market drawdown caused by the Lehman collapse of 2008). Based on the fund to factor relationship (beta) and the allocation (weight) of the fund in the portfolio, the stress test will determine how your portfolio would perform if this scenario were to occur again. A stress test report typically consists of several scenarios.

We then break down the stress tests at the fund level to display the top ten funds based on the greatest negative impact.

www.evestment.com | 47

Conclusion
All investors who are required to select and monitor investment managers should develop a basic understanding of investment statistics. These tools will help make manager selection and monitoring easier and more productive. Of course, you can never neglect qualitative analysis, but you can and should use all of the quantitative tools at your disposal to narrow your list of investment candidates before you embark on the arduous process of due diligence. In addition, the quantitative tools will provide you with good insight that you can use in your qualitative interviews with managers, and when monitoring your investments.

www.evestment.com | 48

Appendix I: Key Investment Statistics


We have categorized the key investment statistics according to absolute or relative return measures, riskadjusted return measures and risk measures. The explanations for each statistic are provided below.

www.evestment.com | 49

I. Absolute Return Measures 1. Monthly Return (Arithmetic Mean): A simple average return (arithmetic mean) that is calculated by summing the returns for each period, and dividing the total by the number of periods. The simple average return does not consider the compounding effect of returns.

2. Average Monthly Gain (Gain Mean): A simple average (arithmetic mean) of the periods with a gain. It is calculated by summing the returns for gain periods (returns 0), and dividing the total by the number of gain periods.

3. Average Monthly Loss (Loss Mean): A simple average (arithmetic mean) of the periods with a loss. It is calculated by summing the returns for loss periods (returns < 0), and dividing the total by the number of loss periods.

www.evestment.com | 50

4. Compound Monthly Return (Geometric): The monthly average return that assumes the same return every period that results in the equivalent compound growth rate from the actual return data. The geometric mean is the monthly average return that, if applied each period, would produce a final dollar amount equivalent to the actual final value- added monthly index (VAMI) for the funds return stream. (The VAMI reflects the growth of a hypothetical $1,000 in a given investment over time, with the index equal to $1,000 at inception.)

www.evestment.com | 51

II. Absolute Risk-adjusted Return Measures 1. Sharpe Ratio: A measure of a funds return relative to its risk. The return (numerator) is defined as the funds incremental average return over the risk -free rate. The risk (denominator) is defined as the standard deviation of the funds returns.

2. Calmar Ratio: A return/risk ratio. The return (numerator) is defined as the compound annualized return over the last 3 years, and the risk (denominator) is defined as the maximum drawdown (in absolute terms) over the last 3 years. (If there is not 3 years of data, the available data is used.) Calmar Ratio = Compound Annualized ROR ABS (Maximum Drawdown)

www.evestment.com | 52

3. Sterling Ratio: A return/risk ratio. The return (numerator) is defined as the compound annualized return over the last 3 years, and the risk (denominator) is defined as the average yearly maximum drawdown over the last 3 years, less an arbitrary 10%. To calculate the average yearly drawdown, the latest 3-year returns (36 months) are divided into 3 separate 12-month periods, and the maximum drawdown is calculated for each. These 3 drawdowns are then averaged to produce the average yearly maximum drawdown for the 3-year period. (If there are not 3 years of data, the available data is used.)

4. Sortino Ratio: A return/risk ratio. The return (numerator) is defined as the incremental compound average period return over a minimum acceptable return (MAR), and the risk (denominator) is defined as the downside deviation below the MAR.

5. Omega: A relative measure of the likelihood of achieving a given return. It represents a ratio of the cumulative probability of an investments outcome above an investors defined return level (the threshold level), divided by the cumulative probability of an investments outcome below an investors threshold level. Omega considers all information readily available from the investments historical return data. The higher the omega value, the greater the probability that the given return will be met or exceeded.

www.evestment.com | 53

III. Absolute Risk Measures 1. Monthly Standard Deviation: Measures the degree of variation of a funds returns around the funds mean (average) return for a 1-month period. The higher the volatility of the returns, the higher the standard deviation. The standard deviation is used as a measure of investment risk.

2. Gain Standard Deviation: Measures the funds average (mean) return only for the periods with a gain, and then measures the variation of only the winning periods around this gain mean. This statistic is similar to standard deviation, but only measures the volatility of upside performance.

www.evestment.com | 54

3. Loss Standard Deviation: Measures the funds average (mean) return only for the periods with a loss, and then measures the variation of only the losing periods around this loss mean. This statistic is similar to standard deviation, but only measures the volatility of downside performance.

4. Downside Deviation: This measure is similar to the loss standard deviation except the downside deviation considers only returns that fall below a defined minimum acceptable return (MAR) rather than the arithmetic mean. For example, if the MAR is 6%, the downside deviation would measure the variation of each period that falls below 6%. (The loss standard deviation, on the other hand, would take only losing periods, calculate an average return for the losing periods, and then measure the variation between each losing return and the losing return average).

www.evestment.com | 55

5. Semi Deviation: a measure of volatility in returns below the mean. It's similar to standard deviation, but it only looks at periods where the investment return was less than average return.

6. Skewness: This measure characterizes the degree of asymmetry of a distribution around its mean. Positive skewness indicates a distribution with an asymmetric tail extending toward more positive values. Negative skewness indicates a distribution with an asymmetric tail extending toward more negative values.

Note: If there are fewer than three data points, or the sample standard deviation is zero, Skewness returns the N/A error value.

www.evestment.com | 56

7. Kurtosis: This measure characterizes the relative peakedness or flatness of a distribution compared with the normal distribution. Positive kurtosis indicates a relatively peaked distribution. Negative kurtosis indicates a relatively flat distribution.

Note: If there are fewer than four data points, or if the standard deviation of the sample equals zero, Kurtosis returns the N/A error value.

www.evestment.com | 57

8. Maximum Drawdown: Measures the loss in any losing period during a funds investment record. It is defined as the percent retrenchment from a funds peak value to the funds valley value. The drawdown is in effect from the time the funds retrenchment begins until a new fund high is reached. The maximum drawdown encompasses both the period from the funds peak to the funds valley (length), and the time from the funds valley to a new fund high (recovery). It measures the largest percentage drawdown that has occurred in any funds data record. 9. Gain/Loss Ratio: Measures a funds average gain in a gain period divided by the funds average loss in a losing period. Periods can be monthly or quarterly depending on the data frequency. Gain/Loss Ratio = ABS (Average Gain in Gain Period Average Loss in Loss Period)

www.evestment.com | 58

IV. Relative Return Measures 1. Up Capture Ratio: Measures a funds compound return when the funds benchma rk return increased, divided by the benchmarks compound return when the benchmark return increased. The higher the value, the better.

2. Down Capture Ratio: Measures the funds compound return when the benchmark was down divided by the benchmarks compound return when the benchmark was down. The smaller the value, the better.

www.evestment.com | 59

3. Up Number Ratio: Measures the number of periods that a funds return increased, when the benchmark return increased, divided by the number of periods that the benchmark increased. The larger the ratio, the better.

www.evestment.com | 60

4. Down Number Ratio: Measures the number of periods that a fund was down when the benchmark was down, divided by the number of periods that the benchmark was down. The smaller the ratio, the better.

5. Up Percentage Ratio (Proficiency Ratio): Measures the number of periods that a fund outperformed the benchmark when the benchmark increased, divided by the number of periods that the benchmark return increased. The larger the ratio, the better. This is a proficiency ratio.

www.evestment.com | 61

6. Down Percentage Ratio (Proficiency Ratio): Measures the number of periods that a fund outperformed the benchmark when the benchmark was down, divided by the number of periods that the benchmark was down. The larger the ratio, the better. This is also a proficiency ratio.

www.evestment.com | 62

V. Relative Risk-adjusted Return Measures 1. Annualized Alpha: Measures the funds value added relative to a be nchmark. It is the Y intercept of the regression line.

*See Beta calculation on page 65.

2. Treynor Ratio: This measure is similar to the Sharpe ratio, but it uses beta as the volatility measure rather than standard deviation. The return (numerator) is defined as the incremental average return of a fund over the risk-free rate. The risk (denominator) is defined as a funds beta relative to a benchmark. The larger the ratio, the better.

*See Beta calculation on page 65.

www.evestment.com | 63

3. Jensen Alpha: Measures the extent to which a fund has added value relative to a benchmark. The Jensen Alpha is equal to a funds average return in excess of the risk-free rate, minus the beta times the benchmarks average return in excess of the risk-free rate.

*See Beta calculation on page 65 4. Information Ratio: Measures the funds active premium divided by the funds tracking error. This measure relates the degree to which a fund has beaten the benchmark to the consistency by which the fund has beaten the benchmark. Information Ratio = Active Premium Tracking Error Active Premium = Investments annualized return - Benchmarks annualized return

www.evestment.com | 64

VI. Relative Risk Measure 1. Beta: Represents the slope of the regression line. Beta measures a funds risk relative to the market as a whole (i.e. the market can be any index or investment). Beta describes the funds sensitivity to broad market movements. For example, for equities, the stock market is the independent variable and has a beta of 1. A fund with a beta of 0.5 will participate in broad market moves, but only half as much as the market overall.

www.evestment.com | 65

VII. Tail Risk Measures 1. Value at Risk (Parametric VaR): The Value at Risk is the maximum loss that can be expected within a specified holding period with a specified confidence level. In the API the Value at Risk is expressed as a percentage loss. The VaR is returned as a positive percentage even though it represents a loss. The VaR calculation assumes a normal distribution of returns.

2. Modified Value at Risk: The Modified Value at Risk is calculated in the same manner as Value at Risk but doesnt assume a normal distribution of returns. In contrast it corrects the Value at Risk using the calculated skewness and kurtosis of the distribution of returns.

3. Expected Tail Loss (ETL): The Expected Tail Loss is the average of returns that exceed VaR. Also known as CVaR

4. Modified Expected Tail Loss (ETL): The calculation of then Modified Expected Tail Loss is identical to the Expected Tail Loss with the exception that it uses the Modified Value at Risk.

www.evestment.com | 66

5. Jarque-Bera: The Jarque-Bera test is a measure of the normality of a distribution of returns. It uses the calculated skewness and kurtosis of the distribution of returns.

6. STARR (Stable Tail Adjusted Return Ratio): Evaluation of risk adjusted performance in an alternative to the Sharpe Ratio way, but STARR takes into account the major drawback of the standard deviation as a risk measure, which penalizes not only for upside but for downside potential as well and employs the ETL of the asset returns for the performance adjustment. It is defined as:

7. Rachev Ratio: Rachev Ratio (R-ratio): Reward-to-risk measure and defined as the ratio between the ETL of the opposite of the excess return at a given confidence level 1- and the ETL of the excess return at another confidence level 1- . That is:

Rachev Ratio is the Expected Tail Return (ETR) divided by Expected tail loss (ETL). ETL is the average of the returns that exceed your VaR number in the left tail, the ETR is the average of the 5% of returns in the right tail at the 95% confidence level.

www.evestment.com | 67

About eVestment
eVestment provides a flexible suite of easy-to-use, cloud-based solutions to help global investors and their consultants select investment managers, enable asset managers to successfully market their funds worldwide and assist clients to identify and capitalize on global investment trends. With the largest, most comprehensive global database of traditional and alternative strategies, delivered through leading-edge technology and backed by fantastic client service, eVestment helps its clients be more strategic, efficient and informed. eVestment recently acquired PerTrac, a leading provider of robust, hedge fund analysis software and workflow solutions, and Fundspire, an innovative, cloud-based technology provider of hedge fund analytics.

www.evestment.com | 68