Online Portfolio Selection with Long-Short Term Forecasting

This work considers an online portfolio selection problem with reward and risk criteria. We use short-term historical data to forecast the reward term, reflecting the current market trend. We use conditional value-at-risk estimated by long-term historical data to measure the investment risk implied in the market. We reformulate the online portfolio selection model with long-short term forecasting as a linear programming problem. Numerical experiments in various data sets examine the superior out-of-sample performance of the proposed model.