| This dissertation consists of three chapters. The first chapter employs a novel statistical technique, the dynamic principal component method, to estimate dynamic factors from a panel of monthly stock returns. I find that stock returns' responses to the small number of factors account for most of the profits of momentum strategies; momentum profits due to the stock specific components of returns are insignificant or even negative. I further show that the statistical factors can be dynamically related to a few macro-economic variables. These findings support a rational explanation for the momentum phenomenon.;The second chapter, co-authored with Pierluigi Balduzzi, explores whether investor heterogeneity may help explain certain empirical asset pricing puzzles. We derive a new aggregate pricing kernel that accounts for consumption heterogeneity and is robust in the presence of measurement errors on individual consumption. When tested using household consumption data from the Consumer Expenditure Survey, the new kernel can reconcile the observed equity premium with reasonable values of relative risk aversion. The estimated risk aversion also falls as we select households with higher financial wealth. In some cases the heterogeneity-adjusted kernel is able to explain the cross sectional variation of risk premiums on stocks and bonds. Finally, portfolios mimicking consumption heterogeneity command significant negative risk premiums and Sharpe ratios, indicating that heterogeneity is associated with priced risks.;The third chapter, co-authored with Eric Jacquier, examines several issues in evaluating the performance of technical trading rules. We document the ability of trading rules in predicting mean, variance and higher moments of returns in both equity and currency markets. Yet despite the prevailing evidence on return predictability, trading rules' profitability is still elusive. At realistic horizons of five and ten years, trading rules can not consistently beat a buy-and-hold strategy. Further, we use the genetic algorithm as a tool to quantify the effect of data-snooping in evaluating trading rule performance. At five and ten year horizons data-snooping creates so large biases in performance statistics that no true economic profitability can be detected. |
