Learning in financial markets

Radnaev, Boris (2015) Learning in financial markets. Doctoral thesis, University of London: London Business School.

Abstract

I study the role of learning in asset pricing and corporate finance applications. Firstly, I develop and empirically test a general equilibrium model of asset pricing, financing and investment dynamics in a trade-off economy where heterogeneous firms face unobservable "disaster" risk exposure and engage in rational Bayesian updating. During periods absent disasters [e.g. "The Great Moderation"]: equity premia decrease; credit spreads decrease; expected loss given default increases; and leverage ratios increase, especially amongst firms with high bankruptcy costs. Time since prior disasters is the key model conditioning variable. In response to a disaster, risk premia increase while firms reduce labor, capital and leverage, with response size increasing in time since prior disasters. Disaster responses are more pronounced than in an otherwise equivalent economy featuring observed disaster risk. Empirical tests of novel model predictions are conducted. Consistent with simulated model regressions, in the real-world data leverage and investment are increasing in time-since-prior-recessions, with the effect more pronounced for firms with low recovery ratios. Finally, in a joint work with Andrea Gamba and Christopher Hennessy we develop a positive and normative framework for determining optimal investment and financing policies when the density is not known. The model is tractable yet general. Periodic operating profit shocks are drawn from N possible density functions, with the density itself following a hidden Markov chain. Optimal investment and financing policies are determined by the net worth state, lagged profitability, and state variables describing probability assessments over densities. Beliefs evolve over time based upon Bayesian updating. The model offers a potential rationale for extreme cash hoarding, which can be understood as an optimal response to model uncertainty. We offer empirical diagnostics, showing that learning-type behavior is difficult to distinguish from other potential objective functions if one relies on mean ratios or standard reduced-form regressions.