RNN-Based Forecasting for Dynamic Portfolio Rebalancing: Deep Learning Approach to Tactical Asset Allocation

Abstract: Deep learning is becoming increasingly popular in finance and time series prediction, including attempts to predict asset prices and the best weights for a trading portfolio. Portfolio rebalancing is an important concept in finance to ensure an investor’s portfolio reflects the risk level and time horizon they are comfortable with. Classical rebalancing methods, e.g., the equal weight (1/N) approach, are predominantly static in nature and are based on fixed timings or thresholds. These approaches, however, fail to incorporate the time dependent and dynamic nature of financial markets in which asset values are affected by macroeconomic events as well as other stochastic events that change over time.