August 20, 2007 3:00 am
Dry though the business of investing using computer models may be, it arouses the basest human instincts. People like it when quants lose money.
Human envy of the rich is natural. But quants are not merely rich. They are also incredibly intelligent. Long-Term Capital Management, which melted down nine years ago, famously had two Nobel prizewinners on its boards. Goldman Sachs Asset Management is also staffed by people who are very clever indeed. The advanced mathematics they use puts their models far beyond the comprehension of lay people.
Delight is magnified when the people who lose money are both rich and clever. The title of Roger Lowenstein's epitaph for LTCM, When Genius Failed, sums up the feelings of many. When quants come to grief, the temptation is to spin a story of hubris, of investors who were not as clever as they thought they were. Commentary on the week when one of Goldman's long-short hedge funds managed to lose more than 30 per cent reads very much like the coverage of LTCM a decade ago.
Let us move on from this. What are quants trying to do, how did they lose so much money so quickly, and what does this tell us about quantitative investing going forward?
There are two reasons for using quantitative models. First, quants aim to find market inefficiencies. Both theory and common sense tell us that the only way to beat the market is to find inefficiencies.
Mathematical models can home in on minute inefficiencies and exploit them. They need computer technology to execute swiftly while the inefficiencies exist.
As many people try to perform the same trick, inefficiencies are eliminated, and the mathematics needed to find those that are left gets ever more complicated. But the historic performance of quant funds suggests that out-performance is possible. Long-short funds made nice profits during the correction in February and March of this year.
However, that performance may not be terribly exciting. LTCM used to describe the job of exploiting tiny mispricings as "hoovering up nickels". So leverage is needed to make those returns interesting.
Second, the growing body of research in behavioural finance shows that human judgment when it comes to investment is flawed in predictable ways that lead to predictable mispricings in the market. A quantitative model, that will follow rules set for it by humans without the risk of human judgment subsequently messing things up, is needed to take advantage of those mispricings.
Both of these justifications for quant investing remain intact. So how have the quants just lost so much money? In essence, many quant hedge funds were guided by their models to hold the same positions. When some of them started to lose money thanks to the credit sell-off, the need to meet margin calls on their debt forced them to sell their "good" investments. As so many quants were crowded in, the result was a stampede downwards for the "good" investments. This happened several days in a row.
These were "25 standard deviation" events, according to Goldman - meaning that in a normal bell curve distribution, such a day would happen only once every 100,000 years. Several such days in succession was, therefore, far beyond what the models had anticipated. Leverage amplified the losses.
This pins the central problem that quants must address. It lies in the statistical concept of "fat tails". Returns on financial markets do not show a classic "normal" distribution, popularly known as a "bell curve" because of its shape.
In this distribution, 95 per cent of data points are within two standard deviations of the mean, and 99 per cent are within 3 standard deviations. In financial markets, the curve appears to be more pointed, with longer, flatter tails at either end.
Hence, maybe even more than 95 per cent of returns may be within 2 standard deviations of the mean, but there is a "fat tail" of extreme outliers, such as the 25 standard deviations seen this month. When they do deviate, they can deviate extremely. This effect may be exacerbated by the herding habits of quant funds.
This is not surprising. Over time, markets tend to go up slowly and steadily, and occasionally drop sharply.
It is common sense that quants, charging in where the market is least efficient, and fuelling the mix with leverage, will find their returns amplify these tendencies - very nice profits most of the time, and even sharper falls when the models go wrong.
So, this need not mean the end to quantitative investing. There is a role for the use of mathematical models to identify and exploit market inefficiencies.
There is also a role for a process that avoids persistent behavioural errors made by humans. Plenty of human beings have lost their shirts this month, remember, without the aid of computers.
But it does show that the models must improve. Somehow, the quants must model the "herding" effects when their peers pile into the same trades. They must build in the fact that their own actions will move the market - and may even be more likely to move parts of the market that are relatively inefficient in the first place.
It also demonstrates quants' limits. "Hoovering up nickels" is only worth the effort if the returns can be magnified by leverage. If the regular stream of profits must come with occasional blow-out losses, and there is reason to believe that it must, then the combination of rigid quant strategies with leverage looks unappealing.
From this, it is fair to guess there will be less leveraged active quant funds going forward - and, maybe, that they will be restricted to those, like large investment banks, who have large supplies of capital to draw on when the market springs a surprise.
But, sadly for some, there will still be a role for clever, and well-paid, mathematicians in the world of fund management.
Copyright The Financial Times Limited 2017. You may share using our article tools.
Please don't cut articles from FT.com and redistribute by email or post to the web.