Author(s)

ZiLin Zhou

Affiliation(s)

La Salle College Preparatory High School, Pasadena 91107, United States.

Corresponding Author

ZiLin Zhou

ABSTRACT

Asset allocation is a fundamental challenge in investment management, traditionally addressed through models such as mean-variance optimization. However, dynamic market environments and multi-period investment horizons limit the effectiveness of static methods. In recent years, the emergence of deep reinforcement learning (DRL) has provided a powerful tool for addressing complex sequential decision-making problems in finance. This paper conducts a comprehensive academic analysis of the application of DRL in asset allocation. First, we introduce the asset allocation problem and its challenges, then review the basic concepts of DRL and its relevance to financial decision-making. Next, we propose a theoretical framework for transforming the asset allocation problem into a Markov decision process and describe in detail how DRL agents learn optimal investment strategies under various assumptions and structures within this framework. Subsequently, through a review of foreign academic literature, this paper examines existing findings on the application of DRL in asset allocation from a qualitative perspective, including the superior performance of DRL strategies relative to traditional methods in certain scenarios and cautionary results where DRL remains competitive even under simple benchmarks. We discuss the current limitations of DRL methods, high transaction costs, and potential directions for improvement and future research priorities. The study concludes that while DRL holds great potential for enhancing asset allocation theory and practice, several key practical challenges must be addressed before its full potential can be realized.

KEYWORDS

Asset allocation; Deep Reinforcement Learning (DRL); Markov decision process

CITE THIS PAPER

ZiLin Zhou. The application of deep reinforcement learning in asset allocation: a theoretical framework and empirical analysis. Journal of Trends in Financial and Economics. 2025, 2(2): 44-50. DOI: https://doi.org/10.61784/jtfe3044.

REFERENCES

[1] Markowitz H M. Foundations of portfolio theory. The journal of finance, 1991, 46(2): 469-477.

[2] Merton R C. Optimum consumption and portfolio rules in a continuous-time model. In Stochastic optimization models in finance. Academic Press, 1975: 621-661.

[3] Sutton R S, Barto A G. Reinforcement learning: An introduction. Cambridge: MIT press, 1998, 1(1): 9-11.

[4] Benhamou E, Guez B, Ohana J J. Deep Reinforcement Learning: Extending Traditional Financial Portfolio Methods. Available at SSRN, 2024.

[5] Jiang Y, Olmo J, Atwi M. High-dimensional multi-period portfolio allocation using deep reinforcement learning. International Review of Economics & Finance, 20025, 98: 103996.

[6] Aboussalah A M, Xu Z, Lee C G. What is the value of the cross-sectional approach to deep reinforcement learning? Quantitative Finance, 2021, 22(6): 1091–1111.

[7] Chen S, Ge L. A learning-based strategy for portfolio selection. International Review of Economics & Finance, 2021, 71: 936-942.

[8] Wei J, Yang Y X, Jiang M, et al. Dynamic multi-period sparse portfolio selection model with asymmetric investors’ sentiments. Expert Systems with Applications, 2021, 177: 114945.

[9] Kruthof G, Müller S. Can deep reinforcement learning beat 1N. Finance Research Letters, 2025: 106866.

[10] Cong L W, Tang K, Wang J, et al. AlphaPortfolio: Direct construction through deep reinforcement learning and interpretable AI. Available at SSRN, 2021: 3554486.