About a week ago the Financial Times
ran an article about a "new" investment analysis technique that could cut through turbulence in the financial markets: neural network analysis. I thought okay, this isn’t new but maybe the application is innovative. Besides, I liked the metaphor the reporter used, a metal ball dropped in a vat of oil and the ensuing ripples that disturb the oil.
The article is about software developed by a Danish investment firm that turned its back on "linear" models to adopt a neural network approach that continually reclassifies investments in a portfolio and then makes suggestions about which equities to buy and which to sell. The proprietary software chews through a heap of data–prices, price-earnings ratio, and interest rates, for starters, and its performance bench mark is the Russell 1000 index.
The test portfolio used to proof the method was acquired in 2007, just before the ball dropped into the oil. For a time it seemed to hold up but then got caught in the turbulence and its undertow. It has now recovered nicely, ahead of the Russell 1000 in fact, and the asset managers are looking for more investors. This is a sweet success story, especially given the demon turbulence looming over the project and the fact that the assets are apparently owned by the Danish state pension plan.
I understood the use of neural network software to counter nonlinear events like market turbulence, and I understood the continual classification and reclassification. But I was intrigued that nowhere in the article was there a mention of risk, risk analysis, or even risk assessment. Maybe it was there all the time, incorporated in the proprietary software, and maybe it just wasn’t mentioned. Certainly the asset managers who developed the program were aware they were at risk–they were chewing their nails as their fund slid down right beside all the other funds that were dropping in value. But assessing risk doesn’t seem to have been a factor in the firm’s new defense against mayhem in the markets.
So. Is it time to shut down your Monte Carlo software? I don’t think so. . . .