The AIG Risk Models Failed, Or Did They?

An recent article in the WSJ discusses the AIG risk models developed by Gary Gordton (note, Gorton, not Gordon as originally posted), a professor at the Yale School of Management. The headline of the article boldly states "Behind AIG's Fall, Risk Models Failed to Pass Real-World Test." Yet, did the models really fail? Gordton's models were developed to gauge the risk of AIG's credit default swaps, but according to the article, "... AIG didn't anticipate how market forces and contract terms not weighted by the models would turn the swaps, over the short term, into huge financial liabilities." The quote is interesting in that it highlights what may be at the heart of AIG's problems. As a result of its ignorance on whether the short-term collateral risk needed to be considered, or its belief that such risk was not something to be worried about, AIG made a decision to not have Gordton assess these threats - even stating later that it knew his models did not consider such risk. So this begs the question once again. Did the models really fail (as approached by Gordton and approved by AIG), or was it more of a lack of understanding of the very products they were modeling? I know some will ask what's the difference - in the end the models were incomplete - but the distinction is significant.

In hindsight, it is easy to point fingers and wonder exactly what risk AIG was even trying to manage. But the real problem here seems to be less about one particular modeler getting it wrong, or developing incomplete models, and more about management ignoring to consider some risk while putting faith in the very same models that were not designed to give the level of confidence or enterprise-wide coverage that is being used to engender confidence. Even the WSJ article (in the body of the story) mentions how "Mr. Gordton's models harnessed mounds of historical data to focus on the likelihood of default, and his work may indeed prove accurate on that front. But as AIG was aware, his models didn't attempt to measure the risk of future collateral calls or write-downs, which have devastated AIG's finances." Of course, this did not keep AIG from trading as though it did, and therein lies the problem. The failure here is less about modeling, or even risk management, and more about corporate management and decision making. Yet, the perception that the problem is with modeling is widespread. Even Warren Buffett is quoted as saying "All I can say is, beware of geeks .... bearing formulas." But what is the alternative? Shall we abandon all risk modeling and simply use our gut instincts? Should we just take risk off the table completely? I don't believe so. While better risk management models should continue to be developed, maybe a little humility is a good place to start. Understanding a company's limitations is key to uncovering its strengths and protecting against its weaknesses.