The Evolution of Asset Pricing: How Advanced Methodologies Addressed Static Factor Models Issues

Asset pricing models provide expected return and cost of capital estimates, unveil risk-return trade-offs, and offer economic insights that enable market agents to make informed investment choices and proper portfolio allocations. Traditionally, static factor models have been the workhorse for these applications due to their minimal complexity and straightforward implementation. Despite their popularity, static factor models present several assumptions and mathematical limitations, which are not based on economic theory and are derived from OLS regressions, that hinder the empirical performance and explanatory power of those approaches. Such restrictions include the staticity and linearity of parameters, inability to handle and discern among large volumes of data, and fixed functional form. For these reasons, this literature review aims at explaining why it has been necessary and what has been the evolution of asset pricing. Various modelling choices are compared based on their empirical performance and interpretability, highlighting the benefits and disadvantages of each framework. Among time-varying models, Kalman filter methodologies provide the strongest empirical attainments, while IPCA and BMA prove to be the most effective among approaches dealing with the “factor zoo” and the curse of dimensionality. Feedforward neural networks offer the best out-of-sample prediction when accounting for non-linearities and complex variable interactions, while random forests result resilient to noisy datasets. Lastly, this review concisely examines the latest methodological advancements, which are able to tackle multiple issues of static factor models at once. Novel architectures feature and combine advanced neural networks to extract the maximum amount of information from the dataset, achieving strong empirical forecasting performances, while conceding economic interpretability and computational complexity to simpler methodologies.