Quantitative investing, a widely-used strategy that employs complex computer models to make rapid trading decisions, has taken a blow to its reputation in the fall-out from the credit squeeze. The models failed to cope with sudden and unusual market conditions and banks took heavy losses.
Riccardo Rebonato, global head of market risk and of quantitative research and analysis at the Royal Bank of Scotland, warns of the dangers of taking human decision making out of risk management in his new book, Plight of the Fortune Tellers.
He argues that the current risk management rests on conceptually shaky foundations and that we should restore genuine decision making to our financial planning, using a proper understanding of probability, experimental psychology and decision theory.
So is quantitative investing overused as a strategy? Does it increase risk by burying decision-making too deeply and thereby threatening market stability? Or will improved models solve past problems and lead to efficient and profitable trading?
Mr Rebonato is answering your questions now in a live Q&A. Please refresh this page for the latest answers.
As we have seen in evolution of various financial tools that ultimately they become easily available to retail investors and day traders. Do you see that happening with quants’ black boxes?
Aurangzeb Bozdar, London
Riccardo Rebonato: Quant black boxes are unlikely to be made directly available to retail investors or day traders. In many cases they rely on exploiting minute price differences, much smaller than the bid-offer spreads available to non-professionals. In other cases, their strategies require being simultaneously long and short, which is well-nigh impossible for retail investors. Not for Aunt Agatha (yet...)
Is it ever possible for a computer programme to be able to monitor levels in fx, commodity and equity charts simultaneously, to look at evolving chart patterns as well as momentum and sentiment the way a real-life human trader (potentially!) can?
Suvra Das, London
Riccardo Rebonato: You ask if a computer programme can monitor fx, equities, commodities etc more efficiently than a human being. As far as monitoring is concerned, it certainly can. It can also beat humans hands down in analyzing data on the basis of a set of pre-determined rules (the rules, by the way, may have been ’discovered’ by the programme itself, not necessarily by a human being: this is how neural networks work).
Where a human being can outperform a computer programme is in recognizing quickly that the world has changed and that yesterday’s rules no longer apply
Do the intellectual tools currently used to assess risk take adequate account of political risk? Do you regard this as a problem? If so, do you think it is a soluble problem? Or is there some inherent difficulty in marrying the kind of necessarily rather imprecise thinking used by political analysts with the kinds of thinking used by people who attempt to model risk by supposedly rigorous mathematical methods?
David Habakkuk, London
Riccardo Rebonato: You raise a very good point. Tetlock has written a book on expert political judgement, and found that experts were by and large found wanting in their predictive power (”hedgehogs” more so than ”foxes”). Still, we can tackle this kind of problems within a robust mathematical and analytical framework: in doing so we move from probabilities as frequencies (as in coin tossing) to probabilities as degree of belief (basically the odds you would advertise if you had to enter a bet). I believe that this second way of looking at probability would be particularly fruitful in risk management, and should be applied more widely.
How can human behaviour be represented by mathematics? Is there any way to understand human intuition in the form of mathematical models?
Amit Kumar Singh, Bangalore
Riccardo Rebonato: Human behaviour may not be representable by mathematical model, but can certainly be modelled as such. The relevant question is the accuracy of the modelling given the application at hand. Modelling human beings as rational utility maximizer, for instance, may not be a perfect, or even a good, representation of their behaviour, but it can help us predict and explain many (not all) interesting human phenomena.
As for human intuition, the way we learn from experience is one of the most distinctive human features. A branch of probability (Bayesian theory) deals exactly with this and, in my opinion, does so very successfully.
Are not these errors traceable to analysts assuming the simplest and the most mathematically tractable distribution for deviations from the mean (the Gaussian normal curve) instead of a long-tailed curves like the inverse square distribution?
Alun Wyn-jones, New York
Riccardo Rebonato: Nobody really still clings to the idea that returns are normally distributed: risk managers are not flat-earthers. However, it is not enough to say that something is not normally distributed: determining and estimating the non-normal tail behaviour is difficult, especially when, by definition, rare events are, well, rare. Another problem is that the pairwise type of dependence (correlation) that can be easily estimated during normal market conditions does not apply during periods of market turmoil. This is well understood at a theoretical level, but, again, calibrating the models to scarce data is very difficult.
In retrospect, doesn’t it appear that the flaw in the quantitative analysis methodology was that it fails to take into account that if everybody is doing something, that an imbalance existed that could turned into a panic by an exogenous event. Like everyone happily enjoying a movie and a fire breaks out... So, a smarter system would have included this in the equation and concluded it was dangerous so exit the the game...or even better, bet against the game.
Riccardo Rebonato: You raise an interesting point, but the matter is a bit subtler. ’Crowded trades’, as they are called, have occurred in the past. If they affect prices in any way (and I am sure they do), their effect should be visible in past price histories. Probably, the effect of a past ”rush for the door” in a crowded trade would manifest itself as a fat-tail event, which is there for everyone to see. So, a smart enough model could, in principle, take this into account.
As for exiting early, let’s not forget that crowded trades can remain crowded, and get more and more so, for a very long time: I seem to remember that Chairman Greenspan’s ’irrational exuberance’ warning was in the mid-1990s. An early exit can be every bit as painful as a late departure.