When Quant Models are Complicated

Paul Wilmott, a mathematician by training who publishes a lot on derivatives mathematics, notes the following:

On a one-to-one basis many people working in banks will complain to me about the models they have to implement. They will complain about instability of the Heston volatility model for example. I will explain to them why it is unstable, why they shouldn't be using it, what they can do that's better and they will respond along the lines of "I agree, but I don't have any choice in the matter." Senior quants are clearly insisting on implementations that those on the front line know are unworkable.

And a large number of people complain to me in private about what I have started calling the 'Measure Theory Police.' These 'Police' write papers filled with jargon, taking 30 pages to do what proper mathematicians could do in four pages. They won’t listen to commonsense unless it starts with 'Theorem,' contains a 'Proof,' and ends with a 'QED.' I'll write in detail about the Measure Theory Police at a later date, but in the meantime will all those people complaining to me about them please speak up...you are preaching to the converted, go spread the word!

In my experience as a risk manager any model that most people don't understand is not used for its stated purpose. Thus, most Asset and Liability committees that make decisions about interest rate risk--what duration to take--use simple, arbitrary (eg, 100 basis point) parallel shifts in interest rates, perhaps a twist. A bank might have a quant working on a Heath-Jarrow-Morton type model, usually put out there as a distraction for outsiders such as rating agency representatives, equity analysts, or regulators, lame attempts impress these outsiders as if such high-tech methods are part of the institutional alpha. Creators of complex models that their colleagues do not understand should know they are being patronized, used as tools, and should find a job where either their complex models are appreciated or simplify their models so that their work is actually used.

Thus, the situation described by Paul here seems to me to be a case of some quants who work for someone who is either too stupid to realize they are props, or don't care and simply enjoy their role for what it is. 'Theorems' and 'proofs' may impress referees at academic journals, but unless you work for Bell Labs, this means you are working on things real decision makers do not use directly. Thus, trying to convince such people that they should not use a Heston model, but rather a simpler model, like GARCH, or even an exponentially weighted volatility, is all well and fine, but given their revealed preference for complex, unstable models, there is probably no appetite for using anything coming out of this group anyway, rather, merely to obfuscate. To dumb things down so that such work will be useful might violate the unspoken understanding in the organization where the quants are supposed to create really complicated models merely to impress obtuse outsiders. One should not mention or discuss the 'usefulness problem' directly with anyone, however, just as one does not raise their hand in the United Way drive to say 'if the company is just trying to buy corporate PR with our donations, shouldn't our salaries be adjusted for this?'. To prosper in such an environment one must understand the game being played.

The only times I have seen complicated models used is when both the quant and his boss understand what he is doing. If only the modeler understands his model, and even he thinks it is unstable, you can be sure his work is a pretext. These are great jobs in only one sense--you can work on your next job for most of the day. They are dead ends, because there is an infinite supply of super educated quants who are clueless as to practicalities and office politics. They are also obvious candidates for layoffs in tough economic times.