Sub-Prime and Hedge Fund Time Bombs

Are hedge fund investors sitting on a volcano?

OLIVIER LE MAROIS and RAPHAEL DOUADY, RISKDATA

The sub-prime crisis impacted alternative investments from July 2007 to March 2008, mainly through 'time bomb explosions', ie. a massive drawdown on hedge fund managers, with apparently nothing in their track records prior to June 2007 that could have suggested any potential high risk.

Using a return-based risk model, an investor who, in June, would have only accepted to invest in hedge funds exhibiting 'normal' risk patterns would have slightly overperformed during the crisis compared to an iso-weighted benchmark. This is a result of an effective elimination of extreme risk takers, but the benefit of this elimination is widely offset by the fact that it also eliminates successful risk takers, and completely fails in detecting time bombs.

Using a non-linear factor-based model, such as the one provided by Riskdata, an investor who, in June, would have rejected any potential 'time bomb', detected by comparing past drawdowns with predicted ones (using the factor model), would have overperformed versus the benchmark by 4%. This is due to a significant reduction of time bombs, while keeping successful risk takers in the portfolio.

This demonstrates that pure return-based models, even if sophisticated, are insufficient to support sound risk budgeting. They help reduce the level of risk, but do not reduce the 'hidden' risk neither do they help select the 'good' risk. This can be successfully achieved with an efficient non-linear factor-based model, which is the only approach that can help discriminate between the 'lucky' managers and the 'talented' ones.

Data used for the study

Materials of the study: 3,216 hedge funds and funds of hedge funds, reporting their returns to Hedge Funds Research database; as of 14 April 2008, with a track record covering at least the period December 2004 up to January 2008, downloaded and analysed using FOFIX Active. For each manager, we computed:

Ex Ante ie. using performances as known at the end of June 2007:

  • Statistics of the time series: volatility, max drawdown, skew, kurtosis etc.
  • Expected 99% worst case (worst of the extreme betas), as that could be derived applying the non-linear factor-model of Riskdata on their performances.

Ex Post ie. analysing their track record from July 2007 up to March 2008:

  • Observed max drawdown during the period.
  • Average performance over the period.

Our benchmark portfolio is iso-weighted on these 3,216 funds. The composition of this benchmark is shown in Table 1:

A crisis driven by 'Time bombs' explosion

  • Ex Post, we can classify our benchmark portfolio in three groups as shown in Table 2:
  • 'A' are the funds for which the crisis period was 'business as usual', ie. they did not experience a drawdown higher than twice their volatility prior to the period.
  • 'B' are the ones that experienced very high drawdowns (more than 2.3 times their volatility), but stayed pretty much in line with what they experienced prior to the crises in terms of extreme risk (the June to February drawdown was less than the max drawdown experienced prior to June). In other words, an investor had no reason to be surprised by their behaviour during the crisis.
  • 'C' are the ones that experienced very high drawdowns, not only compared to their volatility (more than 2.3 times) but also compared to prior max drawdowns (more than 2 times past drawdown). In other words, nothing in their track record could have alerted an investor to such a high level of losses.

This classification in Table 3 demonstrates that this crisis has been driven by the 'Time Bombs' explosion: for an investor equally invested across all the funds, they contribute negatively by 2.1%, offsetting the 2% contribution of the good performers.

Unsurprisingly, the highest proportion of time bombs are within credit-related strategies: fixed income, distressed and event-driven. However, one also finds a significant proportion of time bombs among equity-related relative value strategies. On the other side, there are no time bombs within short selling, and only a small proportion among managed futures.

The critical question for any investor is to know if it is possible to detect such time bombs prior to their explosion. In other words, are they simply hiding their risks (like for sub-prime), and in that case this 23% of exploded time bombs are a signal that investors may sit on a volcano, or is the information simply somewhere here, and in that case could an investor using this information correctly stay away from these time bombs?

Return-based analysis doesn't help in staying away from Time Bombs

Let us imagine that you are an investor in June 2007, with US$1 billion to invest in alternative investments. A maximum diversification approach, iso-allocating on all the managers, leads to a loss of US$1 million over the period. Was it possible at that time, using pure quantitative techniques to analyse the returns distribution of the funds, to stay away from the future losers and detect the future winners, resulting in making money?

In order to do this, the natural approach is to stay away from the funds that have an abnormal return distribution as observed in June 2007, ie. any fund that has a high extreme risk compared to its 'business as usual' risk (using of course the actual distribution or sophisticated distribution modelling, rather than a simple normal model).

An investor using this approach would have some reason to be happy using such a return-based approach: investing equally on all funds flagged 'yes' (ie. with ex ante a good distribution), he would have then benefited from a small positive performance (US$4 million) over the period, instead of flat performance if equally invested on all funds as shown in Table 4.

This apparently promising result is obtained by increasing the proportion of 'A' managers, but it is unfortunately offset by the fact that it eliminates most of the extreme risk takers – including the good ones (elimination of the B category), while doubling the proportion of time bombs (increase average loss in C category).

Our investor can also choose to use a 'non-linear factor-model', as the one proposed by Riskdata, to try to detect time bombs. It relies on the following hypothesis: a significant part of the time bombs are simply lucky managers. They are exposed to factors which happen to have low volatility during the period of analysis. A good example is illustrated by Fig.1: this is a hedge fund exposed to the credit spread. In black, its track record prior to June 2007: high returns, low volatility, low extreme risk – a typical good candidate for a return-based selection. In red, its performances after June 2007. In grey, the spread between government and investment grade bonds: it simply happens that the period 2003-June 2007 was exceptional in the life of this spread, compare to before and after… This manager is simply lucky!

We use this simple criterion: eliminate any funds for which predicted extreme risk (using all factor history) is more than twice the observed past max drawdown or 2.3 times the volatility. Results are simply spectacular as shown in Table 5.

Our investor, choosing this approach, would have made a profit of US$40 million over the period, 10 times what can be achieved using the return-based approach. The main reason is that this approach successfully helps to reduce the proportion of time bombs (C category), while increasing the number of 'A' funds, without eliminating good risk takers (B category).

Are the factors selected relevant? The crystal ball test

The ultimate testto ensure that the factors selected and used in the previous model is to look at the performance of an investor who has a crystal ball on the markets: he perfectly anticipates market behaviour between July and March, and therefore selects only the hedge funds for which no losses are predicted by the factor models over the period. In that case, we get a spectacular confirmation of the relevance of the factor selected: an investor who has a crystal ball on the markets, and using the factor model to reject the predicted loser, would over perform by 5.4% the benchmark as shown in Table 6.

Conclusion

Hedge fund investors are not sitting on a volcano, if and only if they use tools which make risk transparent. Risk transparency simply means avoiding nasty surprises.

Simply analysing returns, even in the most sophisticated way, does not help make risk more transparent. A risk system brings true value by revealing and quantifying the so-called hidden risks or time bombs.

In previous studies, Riskdata has highlighted non linearity (ie changes in correlation) and return smoothing as two sources of hidden risks which can be uncovered with appropriate methods. This study demonstrated that the time bomb effect is also a critical source of hidden risk, particularly in a period of low volatility as the one experienced between 2003 and 2007. Time bombs appear in any bubble because people tend to focus on short-term trends while losing memory of what may happen when the markets come back to earth. Factor analysis is the only way to re-introduce long-term memory.

Finally, this study confirms that a good risk system is not a cost, but should be viewed as an investment with a potentially very high ROI. For an investment expressed in hundreds of thousands dollars in June 2007, our investor would have earned back US$40 million nine months later: therefore, who did better in the market?