How Software Models Doomed Financial Markets
sciam.com
sciam.com
There is an inherent problem with a lot of financial modeling in that it assumes normal distributions (technically lognormal), so a lot of the models are mean-reverting. But financial markets are anything but statistically normal.
Stop blaming short sellers, the uptick rule, hedge funds etc. If noone wants to buy something, the price will go down. That's it.
"There investment bankers from the largest institutions pleaded successfully with Securities and Exchange Commission (SEC) officials during a short meeting in 2004 to lift a rule specifying debt limits and capital reserves needed for a rainy day."
To me it seems like a lot of the models that the quants came up with were essentially a way to justify the above line of reasoning and to show that dependence on mortgage backed securities was sustainable (when in reality it was a path to a meltdown).. Ofcourse that doesnt mean the quants are to be blamed, when the directive had come from elsewhere...
But quants along with other strats (macro, long/short, etc) all had bad risk management models. You can't blame only program trading when there was so much human error.
Also: quants (meaning algorithmic trading) did not show how MBS were good. That was human error; Moody's and S&P ratings, for example.
The problem is that many investment houses simply used bad quantitative data to evaluate their investments. Whether a computer generates that or a human calculates and writes it down doesn't change things.
I'm a big fan of trying new models with computers. Google does it all the time. However, when Google is doing it, the worst case is that the utility of your search results goes down.
Is this actually true? I thought that a number of models at least correlated price changes at a single point in time, and that agent models were correlated nonlinearly across time as well...
Is he crazy? Of course ALL airplanes will eventually fly solo, and some already do - like the drones we send to kill militants in Pakistan.
And it will be a better world when airplanes, surgery, tollbooths, and many other things are done by computers.
The article's main argument was that some things are too risky without multiple supervision. The extreme and current example is letting financial institutions supervise themselves. How to use computers is just one part of the problem.
Many things can run unattended - but under what circumstances should they? What are the necessary safeguards? Pilots on commerical airplanes don't do much flying, they are onboard to make sure everything is safe and works.
Eventually it's all about acceptable risk. We already know automated trading can be dangerous. But the danger lies in bad human assumptions, bad human models and the widespread idea that somehow no human being remains responsible.
As a complete aside - I worry whenever I hear cheerful talk about "the drones we send to kill militants". That kind of convenient and unaccountable power over life and death in other countries is both morally suspect and open to abuse. Killing by remote is still killing and relying on technological quick-fixes to "solve" terrorism is a dangerous strategy.