Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MIGT: Memory Instance Gated Transformer Framework for Financial Portfolio Management

Fengchen Gu, Angelos Stefanidis, Ángel García-Fernández, Jionglong Su, Huakang Li

TL;DR

This work tackles the instability and limited generalization of DRL in volatile financial markets by introducing MIGT, a Memory Instance Gated Transformer for portfolio management. The model integrates a Gated Instance Attention module, memory-trajectory information, and gated memory-based feeding through PW-MLP and Logits MLP to produce robust, high-return trading policies under PPO optimization. Ablation studies show that instance normalization, the LGU gating layer, and the Transformer variant each meaningfully contribute to performance, with MIGT achieving superior cumulative returns and risk-adjusted metrics across DJIA data. The framework demonstrates strong long-horizon profitability and stability, though future work could enrich risk modeling and enable more flexible long/short strategies with appropriate risk controls.

Abstract

Deep reinforcement learning (DRL) has been applied in financial portfolio management to improve returns in changing market conditions. However, unlike most fields where DRL is widely used, the stock market is more volatile and dynamic as it is affected by several factors such as global events and investor sentiment. Therefore, it remains a challenge to construct a DRL-based portfolio management framework with strong return capability, stable training, and generalization ability. This study introduces a new framework utilizing the Memory Instance Gated Transformer (MIGT) for effective portfolio management. By incorporating a novel Gated Instance Attention module, which combines a transformer variant, instance normalization, and a Lite Gate Unit, our approach aims to maximize investment returns while ensuring the learning process's stability and reducing outlier impacts. Tested on the Dow Jones Industrial Average 30, our framework's performance is evaluated against fifteen other strategies using key financial metrics like the cumulative return and risk-return ratios (Sharpe, Sortino, and Omega ratios). The results highlight MIGT's advantage, showcasing at least a 9.75% improvement in cumulative returns and a minimum 2.36% increase in risk-return ratios over competing strategies, marking a significant advancement in DRL for portfolio management.

MIGT: Memory Instance Gated Transformer Framework for Financial Portfolio Management

TL;DR

This work tackles the instability and limited generalization of DRL in volatile financial markets by introducing MIGT, a Memory Instance Gated Transformer for portfolio management. The model integrates a Gated Instance Attention module, memory-trajectory information, and gated memory-based feeding through PW-MLP and Logits MLP to produce robust, high-return trading policies under PPO optimization. Ablation studies show that instance normalization, the LGU gating layer, and the Transformer variant each meaningfully contribute to performance, with MIGT achieving superior cumulative returns and risk-adjusted metrics across DJIA data. The framework demonstrates strong long-horizon profitability and stability, though future work could enrich risk modeling and enable more flexible long/short strategies with appropriate risk controls.

Abstract

Deep reinforcement learning (DRL) has been applied in financial portfolio management to improve returns in changing market conditions. However, unlike most fields where DRL is widely used, the stock market is more volatile and dynamic as it is affected by several factors such as global events and investor sentiment. Therefore, it remains a challenge to construct a DRL-based portfolio management framework with strong return capability, stable training, and generalization ability. This study introduces a new framework utilizing the Memory Instance Gated Transformer (MIGT) for effective portfolio management. By incorporating a novel Gated Instance Attention module, which combines a transformer variant, instance normalization, and a Lite Gate Unit, our approach aims to maximize investment returns while ensuring the learning process's stability and reducing outlier impacts. Tested on the Dow Jones Industrial Average 30, our framework's performance is evaluated against fifteen other strategies using key financial metrics like the cumulative return and risk-return ratios (Sharpe, Sortino, and Omega ratios). The results highlight MIGT's advantage, showcasing at least a 9.75% improvement in cumulative returns and a minimum 2.36% increase in risk-return ratios over competing strategies, marking a significant advancement in DRL for portfolio management.

Paper Structure

This paper contains 23 sections, 16 equations, 13 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Definition
  3. Definition of Data Input, Trading Period and Process
  4. Assumptions about the Experimental Environment
  5. Portfolio Management Objective
  6. DRL Environment
  7. The Markov decision process of portfolio management
  8. State, Action, Reward Function, and transaction agent
  9. Policy Network
  10. Research Problem
  11. MIGT Framework
  12. Gated Instance Attention
  13. PW-MLP, Logits MLP and Memory Trajectory
  14. Experiments
  15. Data Selection
  16. ...and 8 more sections

Figures (13)

  • Figure 1: Interaction Process for Time Series
  • Figure 2: The Markov decision process of the DRL Environment, reflects the interaction of state, reward, action, DRL agents and the stock market environment.
  • Figure 3: the structure of MIGT policy
  • Figure 4: the structure of Gated Instance Attention
  • Figure 5: the Scaled Dot-Product Attention structure
  • ...and 8 more figures