Tactical Asset Allocation Using Deep Reinforcement Learning and Latent Macroeconomic Conditions

Tauseef Kazi¹, Ken Abbott², and Joe Wayne Byers^3*

Abstract

Traditional asset allocation techniques fail to adapt to sudden and severe economic downturns and lead to a loss of opportunities for investors. This paper seeks to address this problem by automating the Tactical Asset Allocation (TAA) framework that allocates the asset weights based on the latent macroeconomic conditions and market regimes. This project employs Deep Reinforcement Learning (DRL) to allocate funds to 36 Exchange Traded funds (ETFs) representing multiple geographies and asset classes. This study demonstrates that a macro, temporal, and spatially aware transformer agent guided by the LSTM-based macroeconomic regime shift encoder can outperform the traditional benchmarks, while taking the transaction cost into consideration.

Keywords

Corresponding Author

Joe Wayne Byers, WorldQuant University, Oklahoma State University, USA.

Citation

Kazi, T., Abbott, K., Byers, J. W. (2025). Tactical Asset Allocation Using Deep Reinforcement Learning and Latent Macroeconomic Conditions. Econ Dev Glob Mark, 1(1), 01-32.