Dynamic Asset Allocation Strategies Based on Reinforcement Learning

DOI

10.2991/978-94-6239-801-6_43How to use a DOI?

Keywords

Dynamic Asset Allocation Strategies; Reinforcement Learning; Risk Control

Abstract

Inherent non-stationary, non-linear, and sudden risk shock characteristics of financial markets often cause decision biases and model failures for traditional dynamic asset allocation models, which depend on steady-state returns. Reinforcement learning, because it can do adaptive sequential decision making, has become a key tool to deal with such problems. This paper addresses the main shortcomings of traditional models by constructing a reinforcement learning dynamic asset allocation model based on the PPO algorithm that incorporates multiple market features and transaction costs. Empirical testing was carried out with the daily data of domestic asset classes between 2010 and 2023. From the empirical results, the study observes that this strategy achieves an annualized return of 14.82% and a maximum drawdown of 11.68%, demonstrating strong profit and risk-control performance. Strategy Sharpe Ratio and Calmar Ratio are 1.23 and 1.27, respectively, much better than the two traditional benchmarks of equal weight and mean variance. It shows a good adaptability under different market situations, and multi-dimensional robustness tests also support the validity of the conclusion. This research improves the theoretical framework of dynamic asset allocation and gives asset management market players a smart allocation instrument that is both dependable and useful.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Chengyue Shi
PY  - 2026
DA  - 2026/10/23
TI  - Dynamic Asset Allocation Strategies Based on Reinforcement Learning
BT  - Proceedings of the 2026 2nd International Conference on Financial Innovation and Marketing Management (FIMM 2026)
PB  - Atlantis Press
SP  - 398
EP  - 406
SN  - 2352-5428
UR  - https://doi.org/10.2991/978-94-6239-801-6_43
DO  - 10.2991/978-94-6239-801-6_43
ID  - Shi2026
ER  -