Asset allocation based on LSTM and the Black–Litterman model

Abstract

ABSTRACT We propose a novel ranking-based approach combined with long short-term memory (LSTM) networks to generate investor views in the well-known Black – Litterman (BL) model. This approach can effectively distinguish high-quality assets with the potential for future growth. In addition, it discards any information contained in the absolute differences between asset prices to mitigate the negative impact of these estimation errors on the BL model. Our findings suggest that the BL model based on our approach can achieve better out-of-sample performance.