ONE of the most alarming things about the crisis in the global financial system is that the warning signs have been out there for some time, yet no one heeded them. Exactly 10 years ago a hedge fund called failed to convince investors that it could repay its debts, thereby bringing the world to the brink of a similar 鈥渓iquidity crisis鈥� to the one we now see. Disaster was averted then only because regulators managed to put together a multibillion-dollar bailout package.
LTCM鈥檚 collapse was particularly notable because its founders had set great store by their use of statistical models designed to keep tabs on the risks inherent in their investments. Its fall should have been a wake-up call to banks and their regulatory supervisors that the models were not working as well as hoped 鈥� in particular that they were ignoring the risks of extreme events and the connections that send such events reverberating around the financial system. Instead, they carried on using them.
Now that disaster has struck again, some financial risk modellers 鈥� the 鈥渜uants鈥� who have wielded so much influence over modern banking 鈥� are saying they know where the gaps in their knowledge are and are promising to fill them (see 鈥淗ow the risk models failed the world鈥檚 banks鈥�). Should we trust them?
Advertisement
Their track record does not inspire confidence. Statistical models have proved almost useless at predicting the killer risks for individual banks, and worse than useless when it comes to risks to the financial system as a whole. The models encouraged bankers to think they were playing a high-stakes card game, when what they were actually doing was more akin to lining up a row of dominoes.
How could so many smart people have got it so wrong? One reason is that their faith in their models鈥� predictive powers led them to ignore what was happening in the real world. Finance offers enormous scope for dissembling: almost any failure can be explained away by a judicious choice of language and data. When investors don鈥檛 behave like the self-interested Homo economicus that economists suppose them to be, they are described as being 鈥渋rrationally exuberant鈥� or blinded by panic. An alternative view 鈥� that investors are reacting logically in the face of uncertainty 鈥� is rarely considered. Similarly, extreme events are described as happening only 鈥渙nce in a century鈥� 鈥� even though there is insufficient data on which to base such an assessment.
鈥淏ankers鈥� faith in their models鈥� predictive powers led them to ignore what was happening in the real world鈥�
The quants鈥� models might successfully predict the movement of markets most of the time, but the bankers who rely on them have failed to realise that the occasions on which the markets deviate from normality are much more important than those when they comply. The events of the past year have driven this home in spectacular fashion: by some estimates, the banking industry has lost more money in the current crisis than it has made in its entire history.
Can modellers do better? There are alternatives to the standard approach: models based on people鈥檚 real-world behaviour (New 杏吧原创, 30 August, p 16) and on 鈥渧irtual agents鈥� (New 杏吧原创, 19 July, p 32) have shown promise, though these are still fringe fields in economics. Most quants, while acknowledging the shortcomings of their models, tend to argue that approximations are necessary, given the difficulty of modelling extreme events, which are in any case rare.
That may be true, but it is dangerous to assume that the approximations are sound. Sometimes even small modelling deficiencies can have huge consequences. , an expert on chance and co-director of the Decision Research Laboratory at the London Business School, argues that all economic activities are subject to extreme events that have complex real-world consequences 鈥� and that such events are inherently unpredictable by any conventional tool. If Taleb is right 鈥� and his argument will be hotly contested 鈥� then we cannot be sure that it is even possible to build risk models capable of making meaningful predictions when they matter most.
Until we are clearer about that, it would be sensible to proceed with caution. Banks should be careful not to assume that they have it right and the rest of the world has it wrong. And regulators 鈥� who have lately allowed themselves to be blinded by science 鈥� should have no qualms about shutting down activities they do not understand. We shouldn鈥檛 need another warning.