Some quantitative investment models performed erratically this summer, leading to losses for investors in some high profile hedge funds. Liam Kennedy asks two quant specialists about this summer's events and the lessons for investors

It's no secret that quant hedge fund managers of various ilks have been recruiting highly qualified PhDs with expertise in fields like behavioural finance in recent years to devise and fine tune their models. And it's also no secret that some of those quant models have fallen foul this year over what have been termed "one-in-10,000-year events".

Nassim Nicholas Taleb - author of The Black Swan: The Impact of the Highly Improbable - debunked such excuses at the recent Hedge Pensions conference in Munich, remarks he also reinforced in October in an article on the op-ed page of the Financial Times. His view, in short, is that we live in a world of ‘fat tails', where unexpected risks are to be expected.

Whatever the explanations and excuses, institutional investors' faith in quant strategies - whether in long-only, 130/30, absolute return or hedge funds - has in some cases been sorely tested.

So what do hedge fund practitioners make of the recent market turmoil, and the failure of quant models to predict the trading patterns in early August that led, in some cases, to heavy losses?

Alexander Raviol, senior risk manager, at Frankfurt-based boutique Lupus Alpha, points out that many quant strategies were pursuing similar strategies. "We believe that there have been many relatively similar strategies in the quant area and that the correlations between many of these quant market neutral funds has increased in the last year, at a time when we see that capital inflows have increased," Raviol says.

The change in factors such as volatility led to a self-perpetuating effect when actual market behaviour did not correspond with the behaviour and conditions that the models were programmed to deal with, where stock prices, for example, changed erratically within a single day's trading. Alongside this, prime brokers increased margin calls and funds, among them some high-profile casualties, were forced to sell liquid assets.

Lupus Alpha undertook an analysis of the volatility of individual factor returns, such as value and growth (see chart on right). Raviol says that in August, as in May 2006 and February/March 2007 there was a clear increase in volatility in the market overall, as well as in the value and growth indices - around 25% in both.

"The annualised volatility of the relative returns moved in a range of between 2.5 and 5% in the three years to August this year and it is interesting to compare the circled areas [see chart below]. Volatility of relative returns did not increase either in May 2006 nor in February/March this year - and was rather independent of the level of market volatility overall. In August this year this volatility increased to almost 10%," says Raviol. "In my view, this is an indication of a different, strongly heightened form of trading activity in value versus growth equities in August this year. This can be interpreted as a sign that many strategies were based on similar factor models at the least."

Thierry Post, head of quantitative strategies at Robeco, agrees with Raviol's general summation of last August's market conditions: "Within equities there has been a lot of crowding in the market place on value momentum strategies [buying short-term winners and selling short-term losers] and the victims have been a victim of their own success. Many parties are focused on the same strategies at the moment, so when there is a liquidity crisis that may originate in a completely different segment of the capital market, in subprime and credit, you may see some kind of domino effect and this was not factored into the quant models.

"Many parties were copying the same strategies over the last couple of years," continues Post, "some of which have been documented in academic literature for more than 20 years, dating back to the early research of de Bondt & Thaler [University of Lausanne], which was in 1985, when they documented the reversal effect [buying long term losers and selling long term winners]. So we were in for something like this."

For his part, Post is keen to draw attention to the fact that these problems were concentrated in certain parts of the market and at certain houses. And he says that Robeco's quant models - which drive funds in the long-only, 130/30 and long-short areas, in equities as well as bonds - survived the crisis well. Its global long-short equity quant fund turned in -2% in August and is above +5% for the year to date. "These quant models actually gave us an early warning sign to stay out of credit and to stay out of equities and we have benefited from that," he says. Downside was limited to -2% in August for Robeco's global long-short quant fund.

Lupus Alpha has several strategies in-house that are based on quant models, one of which contains a component based on factor models. "We also saw the same effect, but not so strongly because leverage to that extent was not part of the strategy," says Raviol.

Post caught international media attention last year for his research into participants in the Dutch game show Deal or No Deal and the lessons for behavioural finance. The high stakes involved are the perfect laboratory, Post argues, for research into how individuals react when they stand to lose or gain such large sums of money.

Post, who started out in econometrics and gradually moved his attention to psychology, has, for example, noted that bond traders take much higher risks in an afternoon's trading session if they lost money in the morning. Gleaning such insights into behavioural finance, he contends, will assist his firm in adapting its quant models to take account of the unusual trading patterns that occurred this August and making them resilient to future shocks.

"Quant models try to repress emotions themselves and they obviously have cognitive ability that far exceeds what any human could possibly do. So there are close links between behavioural finance and quant investment," he says.

Although he points out that other fundamental factors were at work in August - excess liquidity and perhaps monetary policy rather than investor psychology - Post agrees that quant strategies have historically been viewed with suspicion. "They have also sometimes been accused of data mining," Post adds, "searching through large data sets for historical patterns that may not be extrapolated into the future."

So how to navigate the quant minefield? Raviol recommends diversification over a number of quant models. He says investors want to see less ‘black box' and more decorrelation between quantitative strategies.

"I also believe that it continuous development is important with quant strategies. What is dangerous is to develop a model and let it run without continuing to research further," he says. "It is also important to test the strategy not just against the past 5-6 years, which is probably what many have done. Many quant strategies have worked well over that time period and the inherent risks first came to light in August."

Lupus Alpha also attempts to account for unexpected scenarios, says Raviol, "which is not very straightforward because it is not possible to model them, but targeted towards what can happen outside the backtest scenario and not just using recent history."

"One of the ways in which we try to avoid the black box label is by having a continuous debate between researchers and fund managers, and clients are also involved," adds Post. "The second step is to educate clients as to what they can and cannot expect from the model. Expecting that the model could predict the turnaround of a crowded trade, is not realistic. You know that like any good fundamental manager will have his rough times, so any quant model will also have rough times. Clients should choose a strategy and keep to it."

"That would be a third level of behavioural finance, that you avoid that clients, because of shortsightedness or media hype, abandon a profitable strategy for the wrong reasons," adds Eric van Leeuwen, senior vice president, fixed income, at Robeco.

And Post hopes that behavioural finance will continue to provide a useful theoretical basis: "With behavioural finance you have theoretical framework that allows you to have a more focused search and this may help you avoid problems of data mining and to open up the black box. If there is a behavioural finance story behind a model it may be more convincing." Clients will no doubt make their verdict clear.

"Clients are really looking at correlations and making sure that they are not buying identical products," adds Post. "The fact you can explain why it works well helps attract appetite from clients. And the understanding of clients is important because you don't want them to walk away."