Bayesian Learning in Financial Markets

The dissertation consists of three chapters, each focusing on a different application of learning in financial markets. The first chapter addresses survivorship bias in the global equity markets. Survivorship bias refers to people’s tendency of focusing on successful individuals in inferring global attributes. The application in global equity markets helps attenuate the equity premium puzzle. In a hierarchical Bayesian model, I allow the investors to infer the country-specific crash risk from the cross section of countries. I show the upward bias in the measured equity premium is due to crashes that did not occur in-sample and surprises from updating positively on future valuations. In the second chapter, I focus on the secular decline of interest rate along with other secular trends, such as trends in growth rate and inflation. The motivation of the chapter is to highlight that possible structural shifts in the economy can result in drifting long-run steady-state. I show that detrending the dividend yield with the long-term interest rate leads to a more stationary time series and improves the predictability of future returns. To avoid running into problems with the unit root processes, I model the secular trends as hidden or unspanned in a term structure model. The model captures that the possibility that investors distinguish shocks to the short end and long end of the yield curve, and infer from Fed hiking rates that there is an improvement in long-term economic outlook (the Fed information channel). The last chapter presents an interesting case of memory decay in learning about crash risk. The agents in the model display recency bias in inferring crash risk and forget such occurrences at an exponential rate. Learning leads to endogenous boom-bust cycles and optimal hedging demands in asset allocation problems.