Financial markets with data-driven investment decisions

This thesis addresses several topics in finance and consists of two parts. The central theme is formed by economic agents making investment decisions for financial markets using available past and present information. Part I is of a more analytical nature and discusses the influence of ambiguity and transmutability on the investment decisions and argues that limits to rational expectations offer a better understanding of financial market anomalies than limits to arbitrage. It also introduces a modification of the traditional CAPM in which beliefs of economic agents are determined endogenously. Testable predictions of this model are derived. Using these, we test our version of the CAPM, and we also confront it with the traditional multi-factor models. In part II a benchmark multi-asset artificial financial market model is introduced in which economic agents trade in an artificial market, generating prices and return dynamics. In addition, several extensions to this benchmark model are introduced and it is shown that these models can reproduce several well-known stylized facts of asset returns such as heavy tails, volume/volatility correlation and volatility clustering.