Asset Allocation Based on Lstm and Black-Litterman Model

We propose a novel approach to generating the expected returns of investor’s views. Specifically, we use the LSTM to train a classification model to forecast asset price trends. Next, according to the prediction results, we construct investor’s views in the Black-Litterman (BL) model. This approach can efficiently reduce the sensitivity of the BL model to investor’s views, making the BL model not concentrate portfolio weights on a few assets. Our findings suggest that based on the investor’s views derived by LSTM, the portfolios of the BL model outperform mean-variance, equally weighted, and market portfolios