Journal of Engineering, Project, and Production Management, 2027, 17(2), 2025-304
Developing Financial Intelligent Investment Methods Using an Improved-PPO Algorithm
1 Associate Professor, Director
of Professional Degree Office, School of Accounting, College of
Intelligent Accounting Management, Guangdong University of Finance and
Economics, Guangzhou, 510320, China
2 Master’s Candidate in Accounting, School of Accounting,
Guangdong University of Finance and Economics, Guangzhou 510320, China
3 Master’s Candidate in Accounting, School of Accounting,
Guangdong University of Finance and Economics, Guangzhou 510320, China
4 Doctoral Candidate, School of Management, Jinan University,
Guangzhou, 510632, China, E-mail:
YC20250227@126.com (corresponding author).
Project Management
Received December 5, 2025; revised February 5, 2026; July 2, 2026; accepted October 13, 2026
Available online October 26, 2026
Abstract: In the context of dynamic financial markets and the limitations of traditional and existing intelligent methods in balancing risk and return, this study targets the poor generalization and weak risk resistance of investment strategies in dynamic stock market environments. It proposes a Proximal Policy Optimization (PPO)–based investment model that integrates a deep sequence-encoding module and a dynamic, risk-adjusted advantage mechanism. The model employs a hybrid encoder that combines Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) architectures to process multi-granularity time-series data (minute, daily, weekly), effectively compensating for deficiencies in state observations. Meanwhile, a dynamic risk adjustment advantage function is introduced, in which a Multi-Layer Perceptron (MLP) generates dynamic penalty coefficients to enhance risk control, working synergistically with the actor–critic structure of PPO to optimize decisions. Experimental results show that incorporating the deep sequence encoding module yields the fastest convergence in ablation studies, achieving an average round reward of 1.7. In cross-industry Exchange Traded Fund (ETF) portfolio management, the model achieves a cumulative return of 35.85% and a 99% risk tolerance of –2.50%, outperforming comparable models. During simulated high-volatility periods, peak volatility is capped at 5.5%, and the dynamic risk penalty coefficient closely tracks market fluctuations. This research provides a scientific and practical tool for A-share stock timing and portfolio management, enriches methodologies for financial time-series processing, and promotes the practical application of deep reinforcement learning in intelligent finance.
Keywords: Intelligent financial investment; improved proximal strategy optimization algorithm; deep sequence encoding module; advantages of dynamic risk adjustment; multi-granularity time-series financial data.
Copyright © Journal of Engineering, Project, and Production Management (EPPM-Journal).
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.
Requests for reprints and permissions at eppm.journal@gmail.com.
Citation: Yang, C., Zeng, H., Han, S., and Yu, Y. (2027). Developing Financial Intelligent Investment Methods Using an Improved-PPO Algorithm. Journal of Engineering, Project, and Production Management, 17(2), 2025-304.
