Paper: arXiv 2503.09655

Authors: Faezeh Sarlakifar, Mohammadreza Mohammadzadeh Asl, Sajjad Rezvani Khaledi, Armin Salimi-Badr

Abstract

Traditional Long Short-Term Memory (LSTM) networks are effective for handling sequential data but have limitations such as gradient vanishing and difficulty in capturing long-term dependencies, which can impact their performance in dynamic and risky environments like stock trading. To address these limitations, this study explores the usage of the newly introduced Extended Long Short Term Memory (xLSTM) network in combination with a deep reinforcement learning (DRL) approach for automated stock trading. Our proposed method utilizes xLSTM networks in both actor and critic components, enabling effective handling of time series data and dynamic market environments. Proximal Policy Optimization (PPO), with its ability to balance exploration and exploitation, is employed to optimize the trading strategy. Experiments were conducted using financial data from major tech companies over a comprehensive timeline, demonstrating that the xLSTM-based model outperforms LSTM-based methods in key trading evaluation metrics, including cumulative return, average profitability per trade, maximum earning rate, maximum pullback, and Sharpe ratio. These findings mark the potential of xLSTM for enhancing DRL-based stock trading systems.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 6.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced concepts like PPO and xLSTM architectures with mathematical formulations, warranting a high math score. It includes backtesting on real financial data with specific metrics (Sharpe ratio, etc.), establishing a moderate empirical rigor level.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Enhance DRL-based Stock Trading<br>by using xLSTM Networks"] --> B["Methodology: xLSTM + Deep RL<br>(Proximal Policy Optimization)"]
  
  B --> C["Data Input:<br>Financial Data of Major Tech Companies"]
  
  C --> D["Computational Process:<br>xLSTM networks serve as<br>Actor and Critic components<br>for time-series analysis"]
  
  D --> E["Key Findings:<br>xLSTM outperforms LSTM<br>in Cumulative Return,<br>Sharpe Ratio, and other metrics"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a deep reinforcement learning trading system that uses Extended LSTM (xLSTM) networks for both the actor and critic within a Proximal Policy Optimization (PPO) framework.

Source sentence
Our proposed method utilizes xLSTM networks in both actor and critic components, enabling effective handling of time series data and dynamic market environment. Proximal Policy Optimization (PPO), with its ability to balance exploration and exploitation, is employed to optimize the trading strategy.

Assumes 2

  • The trading environment starts with an initial cash balance of $1 million. execution
    Source sentence
    For more details about this environment, we refer to the initial balance amount, which is $1 million to become compatible with the J. Zou et al. research study [5].
  • A turbulence index threshold is used to penalize actions in unstable market conditions. market
    Source sentence
    As shown in Algorithm 2, our reward function first checks the market's turbulence index against a predefined threshold to avoid high-risk situations. We penalize the model for choosing actions in unstable market conditions.

Data

  • Yahoo market data for five major tech companies · Yahoo market-prices public
    2009/01/01 to 2022/01/01 · NVIDIA, Apple, Microsoft, Google, Amazon
    Source sentence
    We use market data from Yahoo for five major companies in the tech industry including NVIDIA, Apple, Microsoft, Google, and Amazon. For these companies, we’ve selected the timeframe from 2009/01/01 to 2022/01/01 as the training data

Costs

Modeled Costs are part of the reported headline results.

Source sentence
transaction_cost = Compute_Transaction_Cost(action, current_stock_prices);

Finds 2

  • The xLSTM‑based model achieves higher cumulative return, max earning rate, average profitability per trade, and Sharpe ratio than the LSTM baseline. performance
    Source sentence
    The results clearly demonstrate that incorporating xLSTM into our architecture outperforms the classic LSTM networks across all evaluation metrics.
  • In the reported configuration (time window 15, batch size 32) the proposed model attains a cumulative return of 39.28 % and a Sharpe ratio of 0.718. performance
    Source sentence
    15 32 39.28 43.17 32.28 129.82 0.718

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 11 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.