A0942
Title: Portfolio allocation using echo state networks
Authors: Michael Grebe - The University of Manchester (United Kingdom) [presenting]
Ekaterina Kazak - University of Manchetser (United Kingdom)
Abstract: Portfolio optimization has been extensively investigated since the 1990s and found widespread application in finance and economics. The state-of-the-art Echo State Network is applied to forecast portfolio weights and improve portfolio allocation. Echo State Networks present a novel approach to the estimation method of Recurrent Neural Networks, reducing computational effort and overcoming commonly faced challenges of Recurrent Neural Networks. The portfolio optimization problem is studied for different portfolio sizes using a dataset on the S\&P100 index from the beginning of 2014 to the end of 2019. A dynamic hyperparameter optimization approach is employed and the empirical results show that the Echo State Network outperformed the commonly used weight estimation approaches based on dynamic conditional variance models.
