A view from the tails in equity return distributions: The role of extreme value theory in modeling contagion and evaluating risk

This dissertation focuses on the behavior of equity returns in the tails of the distribution. It studies the impact of tail behavior on the probability of multiple markets simultaneously moving in an extreme direction. It draws on the tools of extreme value theory to characterize differences in tail behavior between markets and regions. Finally it applies extreme value theory to risk management to estimate the risk of extreme equity market movement.; In Chapter I, I measure contagion by the number of joint booms or crashes across a sample of developed and emerging equity markets. I show that the level of co-movement and the asymmetry between crashes and booms cannot be replicated under assumptions of normal or heavy tailed distributions. I model co-movement as a function of changes in related macroeconomic variables. I find that increases in the interest rates typically moderate both booms and crashes while changes in exchange rates exaggerate regional market crashes. The results show co-movement is a promising measure to capture the dynamics of financial contagion.; Chapter II asks whether booms are more or less likely than crashes and whether emerging markets crash more frequently than developed equity markets. We apply Extreme Value Theory (EVT) to construct statistical tests of both questions. We find that negative tails are significantly fatter than positive tails for a subset of markets in both regions. We also document evidence that emerging markets have fatter negative tails than developed markets. Our findings are consistent with prevalent notions that crashes are more frequent in the emerging markets than among developed markets. However our results of asymmetry suggest that the risk of market crashes varies significantly within the region.; In Chapter III, I extend the results of Extreme Value Theory (EVT) to estimate risk. I compare risk estimates based on normality, empirical data and extreme values. I find that assumptions of normality and empirical methods underestimate the risk in the most extreme parts of the return distribution. At more extreme parts of the distribution however, EVT based methods yield higher risk estimates. I find that in the part of the distribution where empirical data are sparse and the magnitude of losses is most severe, EVT is a powerful tool to estimate the level of potential risk.