Brave new market world
Barron's this weekend explains the world has changed. It means quant models need to be rethought.
What is being predicted as rare events by the models are daily occurences.
The world has changed, the models are virgin once more. Hacker quants are going to be in demand for a long time. A quant will build a model and test its assumptions with existing data. But where no historical data exists, no validation has been made, or no model covers it. It is a new configuration.
What is being predicted as rare events by the models are daily occurences.
All the melodramatic talk of the lowest-ever consumer-confidence reading, the greatest increase in the money supply and the highest sustained market-volatility levels in the annals of markets is spreading numbness.
The clustering of so many once-rare "90% days" -- when 90% of stocks and trading volume move in one direction -- has muted the commentary on them. When the Dow industrials jumped 6.7% Thursday after erasing steep losses, it was their largest one-day jump since...Oct. 28. So maybe it's telling that the morning after this huge gain (and 11% intraday reversal), the market story made neither page A1 of The Wall Street Journal nor the front page of the New York Times business section.
The world has changed, the models are virgin once more. Hacker quants are going to be in demand for a long time. A quant will build a model and test its assumptions with existing data. But where no historical data exists, no validation has been made, or no model covers it. It is a new configuration.
Comments
Taleb a quant if there ever was one. But if you think about it the quants exist because of Taleb's swans. IF there is no swans then your models converge and you capture your context and your model prints money. You are done, you can fire your quant since he is done writing your model.
But then something like subprime mess happens (although it was exactly a black swan) the context changes and the models have got to be rewritten from scratch.
You sure you want to fire you quant then ?
A lot of my friends from the Polytechnique Paris are in the derivatives industry. A good friend of mine is head of R&D worldwide.
They are good people. They are not evil. They write models that don't pretend to be right or wrong, they are just models.
Models, like learning networks, are trained with historical data. But retraining the models is hard and there is no data, so they have to go back to root causes as opposed to just extracting steady state signal from historical patterns of the market.
There are new patterns in the markets, they have never been seen before, their probability in the old models was infinitesimal and now they appear every other day. You don't want to blame your geek for that, au contraire! you want your geek quickly figuring out what the hell is going on. Yes, write a new model, retrain a new network, fudge some factors in software, replace the whole hardware farm for that.
Good point you've made that many geeks just work on models, knowing and not trying to hide that there is much that they do not know and the resulting very high level of uncertainty in these models. If someone elects to gamble that the models are predicting market movements, it is not really the fault of such honest geeks. I am remain skeptical that these models will be sufficiently dependable predictors to be useful as investment tools in our lifetime, and possibly never, due to the possibility, if not probability, that there is a certain amount of randomness that is fundamental to our universe. It may be a noble academic pursuit by geeks to attempt to model the behavior of markets, but it is my opinion that they are tilting at windmills and their talents could be better used in more likely achievable pursuits. Therefore, I would not wish to pay them to model markets in the belief, or hope, that the models will predict the movements of the market with sufficient certainty to make money over a long period without substantial risk of a model breakdown and a blow up. The models are too unreliable to justify their employment, and will remain so even after they tune them for recent events.
As someone said the difference between a horrid investor and a great one? the bad one is right 4 out of 10, the good one is right 6 out of 10. You don't have to be right all the time. The black swan theory, that a model will never get it right 100%, doesn't invalidate the usefulness. It is an opinion however.
Even though I do not personally invest in hedge funds, i just don't like the fees, I do want to believe in the predictive power of models.
I view finance and economics closer to 'medicine' than 'mathematics'. Mathematics are a support tool, a mighty one, but not all knowing.
A great deal of math, modeling, number-crunching, calculation and thinking goes into weather predictions. There's a ton of CPU power behind it. Lots of geek code too.
But nobody thinks the predictions for next summer's weather is going to be any good at all. Whether it's going to rain tomorrow or not -- well not an exact science but still gives you a decent prediction with some probability.
When it comes to weather people already have the right expectations.