During our research of the hedge fund results, we analysed the 35 hedge fund categories available in the HFR database (www.hedgefund-research.com). The HFR database is one of the largest of its kind and covers approximately 4,150 hedge funds and fund of funds. From this database the HFRI indices are derived. These are often used as the benchmark for hedge funds and fund of funds. The HFRI Fund Weighted Composite consists of more than 1,600 fund of funds. HFR monthly data is available for almost every hedge fund category.
A separate section at the top right of this article contains a summary of the identified hedge fund subcategories. The descriptions of these groups are available on the HFR website.
Our research consisted of three parts. First, we researched data such as returns, standard deviations, draw-downs and Sharpe ratios of the different hedge fund categories. For comparison purposes we added the performance data of the benchmarks of the more traditional asset classes.
Secondly, in order to determine whether the results of hedge funds are too good to be true, we corrected the results of the HFR hedge fund categories using an approach that incorporated the main findings of the Malkiel and Saha study (details below). In the last part of our research we looked at the predictability of hedge fund results.
A study of the HFR database results (see table 1) leads to the single conclusion that the returns are relatively high and the risks are low (quite low standard deviations and low betas). The Sharpe ratios, which provide us with an indicator of risk-adjusted returns, are high and the correlations with the other traditional asset classes are low. Compared with the different subcategories the only drawback (not shown in the table) is the variability of the results, which at times can be considerable.
The maximum returns over three years (Max three years column) indicate the maximum possible three-year holding period return over the sample period (1990-2004). The Min three years column indicates the worst three-year holding period return during the same time frame. Remarkably, one would never have had a period of loss within the larger categories fund of funds and main. Only T-bills and bonds show a similar phenomenon.
Because the standard deviations within these categories (this applies to the hedge fund category fixed income as well) are lower than JP Morgan’s bond benchmark, this data seems to justify the conclusion that hedge funds have comparable risks to bonds, but can generate returns similar to those of equities. This conclusion cannot, however, be generalised. The hedge fund categories emerging markets, sector strategies and equity clearly do have a higher risk profile which, on average, is even higher than that of the equity benchmarks.
Recent research by Burton Malkiel, professor at Princeton University and director of the Vanguard Group and Atanu Saha of the Analysis Group, shows that hedge funds have a number of characteristics which portray them as being too good to be true. Based on the Tass database, which is part of Tremont Capital Management that develops hedge fund indices comparable to HFR, Malkiel and Saha state that the results of hedge funds are being inflated by so-called backfill and survivorship bias.
How does this work? Submitting hedge funds or fund of funds results to institutions like HFR and Tremont is not compulsory. New hedge funds usually register after a certain period of success. However, what is even worse is that these successful results are fed retroactively into the databases of the index institutions (backfilling). The research showed that backfilled returns were 5% higher on an annual basis than the other average returns.
Another well-known phenomenon is that underperforming hedge funds and fund of funds will be discontinued and will disappear from the indices. Therefore indices only contain the satisfactorily-performing (surviving) funds. Underperforming funds will usually cease reporting during the last months before they close. Therefore, their most negative results do not even make it into the databases.
Malkiel and Saha discovered that the most notorious hedge fund in history, Long-Term Capital Management, already stopped reporting before its 92% loss which made it insolvent. Furthermore, the study revealed that the output of the worst performers - that ultimately closed down - was 7.4% lower (on an annualised basis) from the moment they ceased reporting.
Based on the backfill and survivorship bias, Malkiel and Saha concluded that on average the hedge fund indices need a 3.8% downward correction on an annual basis1.
With this correction, the hedge funds results are clearly lower than that of equities. What is more, the researchers found that quite big risks prevailed within certain individual hedge funds. In some periods they showed spectacularly positive returns and in others highly disappointing ones. Their variability is therefore very large – this is something we also experienced in the HFR database. Finally, there seems to be little consistency in these results.
Some critics might suggest that survivorship bias also applies to other indices, such as equities. The most promising or popular shares will be registered and the others will just disappear. This, however, happens much less frequently and on a much smaller scale. Backfill and survivorship bias in hedge fund indices will be much less predominant from February 2006 onwards, when (US) hedge funds will have to report their results to the Security and Exchange Commission (SEC). The assumption is that the SEC will distribute this information.
Assuming that Malkiel and Saha are correct, one would tend to think that all the results, such as in table 1, need to be deflated by 3.8%. But it is likely that the backfill and survivorship bias has a positive correlation to the volatility of the fund category. The more volatile a category is, the higher the odds of the temporary emergence of profitable funds and the demise of underperforming ones. This is why we have made the following correction for hedge fund category ‘X’ in the monthly results in table 1:

s(Rx)
Corr. X = -1 * 3.8/12 * ——————
s(Rhfr)average
in which s(Rx) represents the volatility of the hedge fund category and s(Rhfr) average represents the average volatility of all hedge fund categories. In other words: monthly returns for hedge fund categories with volatility levels equal to the asset class average will be deflated downward by 3.8% (the Malkiel & Saha adjustment). Subclasses with lower risk will be adjusted downward by a smaller amount and hedge fund groups with high volatility will be deflated by a larger percentage.
The results are shown in table 2 and indicate that the superior results attributed to hedge funds as an asset category have disappeared. Hedge funds are ‘just another asset class’. The category Main (fixed income, arbitrage, convertible arbitrage, equity market neutral, global macro and event driven) is the only one which is still performing as it should. The returns are slightly lower than the MSCI World index but the standard deviation is considerably lower. The fund of funds have clearly lost much of their appeal. They achieve lower yields than the
JP Government Bonds index but with a higher standard
deviation.
All other groups show an even more disappointing performance. A closer look at the subcategories (besides the five strategies of the category Main) revealed seven other possibly relevant hedge fund styles. These 12 best subcategories are shown in table 3. We filtered the strategies based on their Sharpe ratios and betas compared to those of traditional equities and bonds.
Therefore, only the category main, (consisting of the most well known hedge fund strategies) is worth consideration. Furthermore, when looking at the factor Min three years, this factor never shows a negative outcome. History therefore suggests that the odds of a positive return are very high. The saying that hedge funds in general offer a positive yield every year is clearly a myth. The only exception is within the category Main.
With regards to return characteristics, hedge funds - at least when looking at the indices - have lost some of their appeal. Hedge funds do however have a benefit that should not be forgotten. Hedge funds generally tend to have a low correlation with other asset categories, which makes them highly suitable in an asset allocation context. Besides having a low correlation, they also are less interest-rate sensitive and have a low beta (market sensitivity).
In order to execute a proper tactical asset allocation strategy2, one requires the expected returns, standard deviations and correlations between the different asset categories. One could of course use the historical data after correction for the findings of Malkiel and Saha. But these have shown a lack of consistency in results. Moreover when implementing tactical asset allocation (correcting the allocation for the short term), it is better to ascertain whether we can predict future returns ourselves by using a quantitative model. That was the third part of our research. Using stepwise regression, we tested all 35 hedge fund subcategories (from x1 to x35) for the predictability of some 20 factors. That is:

R(xi)=f(
q Technical factors (momentum, long, short);
q Monetary factors (inflation, long- and short-term interest rates, euro/dollar rates);
q Macro-economic factors (production and unemployment);
q Equity market factors (S&P 500 returns, dividend returns and average price/earnings ratios);
q Commodities (gold and oil price):
of which R(xi) is the total return of hedge fund category ‘xi’ and ‘f’ means a function of.
We checked, for example, whether the short- and long-term momentum plus the current month’s inflation or last year’s inflation had any effect on next month’s returns. In other words, forecasts based on lagged variables. Building attractive regression models with very acceptable R2 appeared to be possible. The R2 explain the predictability of the model. With more than 50 observations (this applied to all our models) an R2 value of more than .08 already represents a reliable regression (at 95% confidence levels). Our models show values equal to or higher than .20 (see figure 1).
Of course, not every variable or factor performed equally well, but the momentum variables stood out positively. The one-year momentum appeared to be an important variable for explaining next month’s return in all 35 hedge fund models, except in the case of distressed securities. Also short momentum popped up repeatedly. The same applies for the unemployment data, but with a one-year lag.
The short-term interest rate in the previous month played a role as well, regularly causing a negative correlation. High short-term interest rates apparently have a negative impact on the next month’s results. The
12-month lagged short-term interest rate variable showed both a negative correlation with the one-month lagged short-term interest rate and a positive correlation with next month’s price. In other words, a current high interest rate will lead to positive results in the following year. In this way we have found countless interesting links of which momentum is the most important one.
For almost all hedge fund categories the following applies: if a fund has not performed well during the last year, it should be avoided. Figure 1 also shows that the category fund of funds has the lowest level of predictability.
Summary: The asset category hedge funds is believed to offer superior results in comparison to other asset classes. However, as is often the case, there are no ‘ free lunches’ in the financial world. Based upon research by Malkiel and Saha one can conclude that due to backfill and survivorship biases, hedge fund returns are overstated by some 3.8% annually on average. The level of overstating differs within hedge fund categories. The higher the volatility the stronger the bias. After correcting for the Malkiel and Saha findings, it seems that only the hedge fund category Main (plus possibly a few other subcategories within table 3) should be used within an asset allocation framework. Smaller investors - due to size limitations - are often forced to use a fund of funds solution. Unfortunately, after correcting for Malkiel and Saha, this category has lost its appeal.
Fund of funds do have a low correlation with the other traditional asset categories. That is why, from a point of view of risk reduction, a modest part of the portfolio may consist of hedge funds. However the returns should not be overestimated.
Larger institutions can benefit by creating their own fund of funds from the hedge funds of the Main category, possibly supplemented by some subcategories as in table 3. Selecting the right hedge funds and determining the right moment to enter into this market are, however, crucial. Good forecasting models and an excellent knowledge of the market are very important.
1. This 3.8% downward correction is lower than an average of the aforementioned 5% and 7.4% would suggest. This can be explained by the difference in sub sample size between ‘regular’ observations on the one hand and backfilled/survivorship-biased observations on the other.
2. A distinction should be made between strategic asset allocation and tactical asset allocation. The first is the ratio between the preferred long-term asset categories, eg, 60% equity and 40% bonds. Among other things, tactical asset allocation addresses the question of what action to take when market developments change the asset class weights away from the strategic asset allocation weights or how extra returns can be achieved by occasionally deviating from the strategic allocation levels.
Erik van Dijk and Harry Geels are CIO and senior investment manager respectively at Compendeon, based in Bunnik near Utrecht

Topics