The project will develop a new family of models for identification of tail risks in financial markets from possibly large datasets using deep learning algorithms. Our newly developed methods will allow us to revisit several classical problems in empirical asset pricing. We believe that the results will be of fundamental character and will open number of questions. Specifically, we aim to explore how deep learning and reinforcement learning can help us to understand the behavior of preference makers departing from classical rationality assumptions, especially those looking at quantile preferences and or heterogeneously persistent investment horizons