Paper: arXiv 2507.00332

Authors: Ruisi Li, Xinhui Gu

Abstract

Propose a deep learning driven multi factor investment model optimization method for risk control. By constructing a deep learning model based on Long Short Term Memory (LSTM) and combining it with a multi factor investment model, we optimize factor selection and weight determination to enhance the model’s adaptability and robustness to market changes. Empirical analysis shows that the LSTM model is significantly superior to the benchmark model in risk control indicators such as maximum retracement, Sharp ratio and value at risk (VaR), and shows strong adaptability and robustness in different market environments. Furthermore, the model is applied to the actual portfolio to optimize the asset allocation, which significantly improves the performance of the portfolio, provides investors with more scientific and accurate investment decision-making basis, and effectively balances the benefits and risks.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 6.0/10 (capped from 7.2: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 7.2)

Why this score: The paper applies a well-known deep learning architecture (LSTM) to a traditional finance problem (multi-factor investing), involving mathematical concepts like regression and time-series modeling; however, it provides empirical backtesting results with specific risk metrics (Max Drawdown, Sharpe, VaR), comparative tables, and discusses data preprocessing and application to portfolio optimization.

Research Flowchart

  flowchart TD
  A["Research Goal: Optimization of Multi-factor Investment Model<br>Driven by Deep Learning for Risk Control"] --> B["Data Preparation & Processing"]
  B --> C["Key Methodology: LSTM & Multi-factor Integration"]
  C --> D["Computational Process: Optimization of Factor<br>Selection & Weight Determination"]
  D --> E["Risk Control Application:<br>VaR & Max Drawdown"]
  E --> F["Outcome 1: Model Evaluation<br>Superior Sharp Ratio & Robustness"]
  E --> G["Outcome 2: Portfolio Optimization<br>Enhanced Asset Allocation"]

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 learning driven multi‑factor investment model optimization method for risk control.

Source sentence
Propose a deep learning driven multi factor investment model optimization method for risk control.

Assumes

No explicit assumptions found in the text.

Data

Backtest sample: 2010 – 2020 · stock market · equities · daily

Costs

Not modeled A trading or execution result is reported without costs.

Finds 3

Reported performance (LSTM-based multi-factor investment model): Sharpe 0.90 (benchmark 0.68) · max drawdown 8.8%

  • The LSTM model outperforms the benchmark model with higher average daily return and Sharpe ratio, and lower maximum retracement and VaR across market environments. performance
    Source sentence
    In the turbulent market environment, LSTM model also proves its better market opportunity capture ability and risk control level with higher average daily return and Sharp ratio, and lower maximum retracement and VaR.
  • Optimizing the LSTM model increases annualized return and Sharpe ratio, though the marginal benefit diminishes at higher optimization levels. performance
    Source sentence
    Figure 2 shows that with the increase of the optimization degree of the LSTM model, both the annualized rate of return and the Sharp ratio increase significantly,
  • Compared with a traditional linear regression model, the LSTM model reduces maximum retracement by 29.6%, increases Sharpe ratio by 32.4%, and decreases VaR by 20.0%. performance
    Source sentence
    The maximum retracement of LSTM model decreased by 29.6%, the Sharp ratio increased by 32.4%, and the VaR value decreased by 20.0%.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 31 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.