Essays on Prediction and Behavior of Financial Markets

Abstract

This cumulative dissertation develops and applies methods to predict and empirically study financial market behavior. It presents three papers examining different research questions on the economic and statistical laws governing financial markets. The first study, Improving Performance of Corporate Rating Prediction Models by Reducing Financial Ratio Heterogeneity, develops a methodology to construct better performing models to predict credit default rates of large corporations across different industries. It was motivated by the fact that our consulting team had difficulties to construct rating models for large corporates, due to limited available data on defaults and heterogeneity in financial ratios across industry groups. Published work did not provide much methodological help. This motivated developing our own methodology to account for industry heterogeneity within the rating model, and thereby achieving a notable improvement in prediction accuracy. The second paper, Exploiting Attention-driven Mispricing: Evidence from Actual-Dollar Trading, develops a systematic trading strategy for U.S. stocks and successfully trades it in a true out-of-sample test with real money. These results not only motivated investors to provide the seed funding to start a quantitative asset management firm, it also posed the question of how these profits could be possible and persistent for a longer period. Given that the widely accepted efficient market hypothesis (Fama, 1970) implies that financial markets eliminate such profit opportunities quickly, this conflicting observation deserved further investigation. Third and finally, the essay High Frequency Trading Intensifies Intraday Extreme Events in Stock Returns investigates whether high frequency trading (HFT) activity exacerbates large intraday price moves in the stock market. The idea of investigating the link between HFT and intraday extreme events was motivated by my intraday market observations from countless hours of automated trading surveillance. Thereby, sudden bursts of activity and volatility – often without any news – were a surprisingly regular phenomenon. At the same time, there is a dichotomy in the literature. On the one hand, several published empirical studies indicate that HFT activity dampens volatility and improves market quality. On the other hand, theoretical models and institutional traders formulate multiple plausible mechanisms by which HFT could cause extreme events in short-term stock returns.