Hyperparameter Optimization for Black-Litterman Model via Genetic Algorithms

Abstract

In modern portfolio theory, mathematical models that depend on the return and the risk of portfolio, are commonly used in portfolio and asset management. These models can be considered most useful due to their customizability. Consequently, hyperparameter optimization objective is a priority for modifying and customizing these models. In this study, the Black- Litterman model and its parameters are explored for the aim of further enhancing the performance of this model in portfolio optimization task. Next, the effect of the Black- Litterman model’s hyperparameters on producing optimal portfolio distribution is examined by applying a hyperparameter optimization technique based on genetic algorithm. Using this technique, optimal values of historical data range, return range, benchmark distribution and risk aversion factor are produced and utilized in the BlackLitterman model to generate a diversified portfolio. To evaluate the performance of the suggested optimization method, real-world data of five popular indexes in the Turkish stock market is utilized to showcase the benefit of presented technique on real world investment strategies such as, maximizing profit amount, maximizing return per risk and estimation accuracy.