After the 2008-09 financial crisis, policymakers around the world focused on enacting improvements that would make the emergence of a financial crisis less likely (ex ante) and recovery from one more rapid (ex post). This Article identifies a gap in both the academic literature and the current financial regulatory framework in exploring how to limit the damage—to other firms, and to the financial system—when a crisis is ongoing. Policymakers cannot predict the origins of every future crisis, just as firefighters cannot predict the origins of every future fire. Once one begins, how can they keep the damage from spreading?
The academic theory on financial crisis “firefighting” divides into two main camps. The “capital view” claims that runs on financial institutions are fundamentally rational, and that investors care mainly about solvency. Under this view, the best way to fight runs is to raise capital requirements ahead of time, to multiples of current levels. The “contagion view” claims instead that the lack of liquid assets both defines and causes bank runs; an institution’s access to cash (and instruments like it) determines whether and when investors will withdraw funding. Under this view, the best way to fight runs is for governments to lend banks money—freely, at high rates, and against good collateral—and to promise to do so well before a crisis starts.
In this Article—the first to directly address this question empirically—we show that neither view fits the most catastrophic financial shock of the last ninety years: the 2008 Lehman Brothers bankruptcy. In some cases, banks with more capital and liquidity were actually more exposed, not less, to the market panic following Lehman’s collapse. By contrast, we show that simple market correlation was a powerful predictor of exposure to the Lehman run. We also show that market valuations of large banks are more highly correlated today than they were in September 2008, creating a potential unaddressed conduit for an unexpected shock to metastasize into a contagious run.