Latent factor models and returns on financial assets.

Summary Given the empirical failure of observable risk factors in explaining financial returns, the aim of this thesis is to use latent factor models and recent econometric developments to improve the understanding of asset risk. First, we describe latent factor models applied to finance and the main estimation methods. We also present how the use of financial and econometric theories allows us to link statistical factors with economic and financial variables in order to facilitate their interpretation. Second, we use latent factor models in cross-sections to estimate and interpret the risk profile of hedge funds and stocks. The methodology is consistent with the statistical properties of large samples and the dynamic nature of systematic risk. In a third step, we model a market where prices and volumes are sensitive to intraday liquidity shocks. We propose a structural mixture model of latent two-factor distributions to capture the impact of information shocks and liquidity frictions. This model allows us to construct a static liquidity measure specific to each stock. Second, we extend our structural model to account for the dynamic properties of liquidity risk. In particular, we distinguish two liquidity problems: liquidity frictions occurring at an intraday frequency and illiquidity events that persistently deteriorate market quality. Finally, we use statistical time series modeling to construct dynamic liquidity measures.