The Risk Mirage at Goldman
VaR, used for decades on the Street, supposedly reveals the maximum amount an investment house could lose (to a statistical degree of confidence) on its trading portfolios in a specified period. And the peak in Goldman's average daily VaR—$245 million in the second quarter, almost double that of a year earlier—prompted accusations that the well-connected firm had dangerously ratcheted up risk-taking in the midst of a crisis.
There's no reason to think this is true. Whatever your opinion of Goldman's fortunes and market forays in this recession, the fact is that a VaR-based analysis of any firm's riskiness is useless. VaR lies. Big time. As a predictor of risk, it's an impostor. It should be consigned to the dustbin. Firms should stop reporting it. Analysts and regulators should stop using it.
Some, including regulators who base capital reserve requirements on this metric, may call VaR a "measure of market risk" and "predictor of future losses." But it is neither of those things. Its forecast of how much an investor can lose from a trading position is entirely calculated from historical data. It's a mathematical tool that simply reflects what happened to a portfolio of assets during a certain past period. (The person supplying the data to the model can essentially select any dates.)
PAST ISN'T PROLOGUEThe VaR metric has little to do with how a portfolio will fare in the future—and that includes tomorrow. When it comes to the market, the past is definitely not prologue. A current calm may not yield future placidity. Turbulence is not inevitably followed by further chaos.
VaR models also tend to plug in weird assumptions that typically deliver unrealistically low risk numbers: the assumption, for instance, that markets follow a normal probability distribution, thus ruling out extreme events. Or that diversification in the portfolio will offset risk exposure (because a group of assets happened to move in different directions previously).
The result? Well, let's look at the years and months—even the days—leading up to the credit crisis. VaR was very low across all banks generally up to mid-2007, and even well into 2008. VaR looked in the rearview mirror and saw no trouble, thus concluding that a rosy future was inevitable.
As we now know, however, in the real world, banks—even those with relatively small VaR numbers—were running huge risks, despite what the metric reported. Indeed, at least two of the worst-hit investment banks during the mayhem, Merrill Lynch and Bear Stearns, had significantly lower VaRs than Goldman did before the troubles kicked in. Once the crisis hit, there was a plethora of so-called VaR exceptions, days when (surprise!) trading losses exceeded the theoretical losses churned out from the models. For Swiss giant UBS (UBS) in 2007 and 2008, those added up to almost 20 times more loss days than what VaR would have predicted.
So what should we read from Goldman's currently high VaR figures? Not much. VaR naturally increases following a period of broad unrest, because the model's inputs reflect more of that turbulence and less of the prior tranquility. So a pumped-up VaR doesn't automatically point to wild-eyed gambling. The opposite could even be true—Goldman could be trying to trim risk while its VaR rises.
What should replace VaR? We should reintroduce the spirit behind the way we calculated risk before VaR took over on trading floors and in the offices of regulators in the early 1990s. That means using intuition and experience-honed common sense. It means accepting the principle that toxic assets should be considered riskier than liquid assets—and that fancy math and past performance can be deceiving predictors that often deliver a lethal dose of leverage. We need rules and risk committees that limit a bank's "bad" leverage by requiring much more capital to cushion, say, a subprime collateralized debt obligation than to offset Treasuries. It's time to give up analytics so that real risk can be revealed.
Jack and Suzy Welch are off this week.