Essays In Asset Pricing And Information Economics PDF Download

In this thesis we study two applications of machine learning to estimate models that explains asset prices by harnessing the vast quantity of asset and economic information while also capturing complex structure among sources of risk. First we show how to build a cross-section of asset returns, that is, a small set of basis or test assets that capture complex information contained in a given set of characteristics and span the Stochastic Discount Factor (SDF). We use decision trees to generalize the concept of conventional sorting and introduce a new approach to robustly recover the SDF, which endogenously yields optimal portfolio splits. These low-dimensional investment strategies are well diversified, easily interpretable, and reflect many characteristics at the same time. Empirically, we show that traditional cross-sections of portfolios and their combinations, especially deciles and long-short anomaly factors, present too low a hurdle for model evaluation and serve as the wrong building blocks for the SDF. Constructed from the same pricing signals, our cross-sections have significantly higher (up to a factor of three) out-of-sample Sharpe ratios and pricing errors relative to the leading reduced-form asset pricing models. In the second part of the thesis, I present deep neural networks to estimate an asset pricing model for individual stock returns that takes advantage of the vast amount of conditioning information, while keeping a fully flexible form and accounting for time-variation. The key innovations are to use the fundamental no-arbitrage condition as criterion function to construct the most informative test assets with an adversarial approach and to extract the states of the economy from many macroeconomic time series. Our asset pricing model outperforms out-of-sample all benchmark approaches in terms of Sharpe ratio, explained variation and pricing errors and identifies the key factors that drive asset prices.