The necessary post-mortem analyses for the financial system's collapse have begun in earnest. While some have taken to talking about Wall Street's "stupidity," this is an unsatisfying and ultimately counterproductive approach. Clearly, the financial crisis had numerous major and contributing causes (the true "perfect storm"), but failures of models and analytics were key factors.
Rather than think of "models" as a singular entity, it's useful to separate them into "consumer behavior" and "position management" models. The first would include credit performance models, which predicted losses that were low by several orders of magnitude. The failure to correctly project credit-related losses led directly to a cascading series of problems, including the creation of entire classes of securities (such as CDOs backed by subprime subordinate bonds) that were fundamentally flawed.
A major factor in the poor performance of credit models was their failure to anticipate the interaction between weak underwriting standards and the nationwide weakness in home prices. However, the fundamental problem with these models is the difficult nature of understanding and forecasting human activities and choices. Modelers are forced to deduce the reasons underlying consumer behavior based on limited data, which often leads to erroneous or incomplete conclusions. In addition, behavior changes over time, based on factors such as economic circumstances and cocktail party chatter.
Compounding the problems caused by the industry's gross underestimation of losses was the way the faulty projections were incorporated into risk management systems and models. These types of models have failed spectacularly in the past, most notably in the LTCM near-disaster in 1998. Cottage industries have been built around the problem of so-called fat tail events, in which seemingly unlikely events and correlations have a much higher likelihood of occurring than predicted by statistical models.
While much blame for the financial crisis has been placed on these systems, the financial system's gargantuan losses were not primarily due to bad modeling but bad management. Over time, the risk management cultures of both individual firms and the financial services industry broke down completely. These failures are represented by the widespread use of "value-at-risk" (VAR) models to manage risks. These types of systems attempt to boil down a position into a single dollar value that represents how much a firm can make and lose in a day, based on statistical probabilities of potential daily changes in the market values of assets and hedging instruments.
In retrospect, many management decisions of the past few years were made without the slightest nod to common sense. Irrespective of what any VAR model might say, huge and highly leveraged positions in low-quality mortgage assets (made to shaky borrowers with minimal equity) are enormously risky. Adding in the trillions of dollars in off-balance sheet assets controlled by the major firms makes the shocking near-collapse of the financial system, in retrospect, seem almost inevitable.
Equally telling were the failures in risk management processes. As an egregious example, The Wall Street Journal reported last year that, during the boom in CDO issuance in 2004-2006, Merrill Lynch's traders were encouraged by their superiors to withhold information from risk management. If true, this represents a fundamental institutional and cultural weakness that no amount of mathematical brilliance can overcome.
The ubiquitous nature of VAR models was symptomatic of the abdication of risk management responsibility to quants. All models, including those measuring VAR, are simply tools. Effective risk management represents the use of information to allow managers to monitor and limit exposure to a variety of factors that include model error. In this light, the financial crisis was not caused by "stupidity" but instead resulted from fundamental management and organizational failures.
Bill Berliner is a mortgage and capital markets consultant based in Southern California. His web site is www.berlinerconsulting.net
(c) 2009 Asset Securitization Report and SourceMedia, Inc. All Rights Reserved.