Black–Litterman portfolio optimization based on GARCH–EVT–Copula and LSTM models

Abstract

In constructing diversified portfolios, the investors might be interested in incorporating some quantifiable views or opinions. The Black–Litterman model is a useful approach to integrate investors’ views into the Markowitz allocation model. In this paper we utilize a deep learning model to estimate the investors’s views and use GARCH–EVT–Copula to model the dependence structure between stock market returns in a large portfolio. The findings show that the Black–Litterman model for portfolio optimization based on GARCH–EVT–Copula and LSTM (Long Short Term Memory) models gives better performances as compared with the traditional max-Sharpe and the original Black–Litterman portfolio problems.