Eric Falkenstein has posted an exceedingly well-written response to Felix Salmon's recent Wired article, "Recipe for Disaster: The Formula that Killed Wall Street" over on SeekingAlpha. Falkenstein nicely captures the interaction between the "quants" and senior management in financial institutions when he writes,
"The decision makers are rich, powerful, kind of smart, do not feel embarrassed by their lack of knowledge in obscure technical trivia, and surely are not intimidated by it."
In my experience, the best quants are generally quite keen to identify and debate the risks and assumptions in their models, but too rarely encounter non-quant managers who show the patience to comprehend the implications. If you're rich, powerful and kind of smart, it's more comforting to say: "I don't pretend to understand all this greek, but I've hired the smartest guys to do the math" than to say "I've tried to understand it, and frankly it's over my head."
Felix Salmon's Wired article is indeed worth reading, although the headline is a tad melodramatic. In Salmon's view, the adoption of Gaussian copula techniques to model the risk characteristics of mortgage-backed securities mortally wounded Wall Street.
Over at the New York Times, Joe Nocera seems to think it was a Value-at-Risk spreadsheet that killed Wall Street in last month's "Risk Mismanagement" article.
Perhaps Wired magazine will sponsor a debate between Mr. Salmon and Mr. Nocera about whether it was a VaR model or a Gaussian copula model that buried Wall Street. Nassim Taleb might be willing to moderate.
To Mr. Salmon's credit, his article does discuss what is -- in my opinion -- the single biggest source of failure in these risk management models. Namely, the use of CDS price data as a proxy for otherwise hard-to-track correlations among many discrete and illiquid securities. It was the availability of a real-time price series that apparently made the Gaussian copula function "tractable", but the use of CDS price data appears to have led to models that vastly and tragically oversimplified the real world relationships they were meant to simulate. Moreover, as Salmon points out, the limited history of CDS prices meant that the historical data was largely drawn from a period of benign economic data. Finally, the CDS market ultimately became a speculator's market with the notional value of default insurance dwarfing the underling credits supposedly being insured, which undoubtedly raised the noise-to-signal ratio on CDS price movements and correlations.
Salmon's and Nocera's articles are the most prominent examples of a growing "The model made me do it" set of explanations for Wall Street's current predicament. Unfortunately, the journalistic imperative for a catchy headline and strongly themed story tends to gloss over the more complex human, institutional and managerial failings that are more appropriately to blame.
If you've read this far, and you're truly interested in the subject, I highly recommend reading UBS's confessional Shareholder Report on UBS Writedowns (pdf). This remarkable report from April 2008 describes the causes of UBS's losses related to US residential mortgages, which at that time were a mere $18.7 billion.
What you will find in this report is a board-approved "hurry up" strategy to rectify lagging league table performance in global fixed-income markets, which led to an aggressive market entry into the RMBS market as it was peaking. You will find that internal capital charges were not routinely adjusted for the true risks of proprietary positions, leading to "carry trades" and correspondingly high inventories of ultimately risky securities. You will find that "warehoused" securities held-for-sale were not hedged at all. You will find that dubious AAA-rated mortgage backed securities were hedged based on the five-year default histories of the small handful of remaining AAA-rated corporates during period of strong economic growth. And you will find that compensation policies rewarded traders for current year profits, even if the positions they held proved toxic down the road. What you don't find in this document is "Oops, the models broke" types of excuses.
For anyone interested in the subject, I also recommend Suna Reyent's article on SeekingAlpha as well.