Paper: arXiv 2507.01973

Authors: Junjie Guo

Abstract

Portfolio allocation via stock price prediction is inherently difficult due to the notoriously low signal-to-noise ratio of stock time series. This paper proposes a method by integrating wavelet transform convolution and channel attention with LSTM to implement stock price prediction based portfolio allocation. Stock time series data first are processed by wavelet transform convolution to reduce the noise. Processed features are then reconstructed by channel attention. LSTM is utilized to predict the stock price using the final processed features. We construct a portfolio consists of four stocks with trading signals predicted by model. Experiments are conducted by evaluating the return, Sharpe ratio and max drawdown performance. The results indicate that our method achieves robust performance even during period of post-pandemic downward market.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 6.0/10 (capped from 6.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 6.5)

Why this score: The paper uses advanced mathematical concepts like wavelet transforms and attention mechanisms, pushing the complexity beyond basic ML. It also includes specific portfolio metrics (return, Sharpe ratio, max drawdown) and backtesting on real data, demonstrating practical implementation rigor.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Portfolio Allocation via<br>Stock Price Prediction"] --> B["Input Data<br>Four Stocks Time Series"]
  B --> C["Wavelet Transform Convolution<br>Noise Reduction"]
  C --> D["Channel Attention<br>Feature Reconstruction"]
  D --> E["LSTM Model<br>Price Prediction"]
  E --> F["Trading Signals<br>Portfolio Construction"]
  F --> G["Evaluation Metrics<br>Return, Sharpe Ratio, Max Drawdown"]
  G --> H["Key Finding<br>Robust Performance in<br>Post-Pandemic Downturn Market"]

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 method that integrates wavelet transform convolution and channel attention with LSTM for stock price prediction based portfolio allocation.

Source sentence
This paper proposes a method by integrating wavelet transform convolution and channel attention with LSTM to implement stock price prediction based portfolio allocation.

Assumes 4

  • The portfolio uses equal weighting for each stock. execution
    Source sentence
    This study uses equal weight for each stock…
  • The set of stocks is fixed and the portfolio is rebalanced daily. execution
    Source sentence
    In stock price prediction based portfolio allocation, at each time step the portfolio is rebalanced by assigning stocks as long or short positions. We study the portfolio with fixed stocks and rebalance frequency of one day.
  • Trading signals are generated from one‑step‑ahead price predictions. model
    Source sentence
    This paper studies one-step ahead stock price prediction.
  • No transaction costs are considered in the reported results. execution
    Source sentence
    The results indicate that our method achieves robust performance even during period of post‑pandemic downward market.

Data

Backtest sample: S&P 500 stocks · equities · daily

  • Historical daily data for AAPL, AMZN, GE, MSFT market-prices unknown
    October 1, 2013 to September 29, 2023 · Four major stocks (AAPL, AMZN, GE, MSFT)
    Source sentence
    The empirical analysis is based on historical daily data for four major stocks: AAPL, AMZN, GE, and MSFT. The data spans from October 1, 2013, to September 29, 2023, providing a comprehensive view of market behavior over nearly a decade.

Costs

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

Finds 4

Reported performance (Wavelet Transform Convolution + Channel Attention LSTM): Sharpe 1.84 (benchmark -0.52) · annual return 23.7% · max drawdown 14.0%

  • The proposed method’s portfolio achieves an annualized return of 18.19%, Sharpe ratio of 1.47, and maximum drawdown of 20.83%, outperforming the buy‑and‑hold benchmark (3.73% return, -0.52 Sharpe, 32.35% MDD). performance
    Source sentence
    Portfolio 18.19% 1.47 20.83% B&H Portfolio 3.73% -0.52 32.35%
  • Across individual stocks, the proposed method yields higher returns and Sharpe ratios than baseline models (e.g., AAPL 23.08% return, 2.21 Sharpe). performance
    Source sentence
    Our Method AAPL 23.08% 2.21 18.78% AMZN 21.56% …
  • The equal‑weighted portfolio trading strategy provides more robust returns and lower maximum drawdown compared to individual stock trades and buy‑and‑hold. performance
    Source sentence
    The results indicate that our portfolio trading strategy achieves a more robust return while maintaining a lower maximum drawdown compared to the individual stock trades.
  • The method maintains strong performance in a high‑volatility, post‑pandemic market environment. performance
    Source sentence
    The results indicate that our method achieves robust performance even during period of post‑pandemic downward market.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 24 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.

Cited References (31)

  1. Quant 4.0: engineering quantitative investment with automated, explainable, and knowledge-driven artificial intelligence — → 2301.04020
  2. Portfolio selection (1952)
  3. Capital Asset Prices: A Theory of Market Equilibrium under Conditions of Risk (1964)
  4. Common risk factors in the returns on stocks and bonds (1993)
  5. Deep Learning in Long-Short Stock Portfolio Allocation: An Empirical Study — → 2411.13555
  6. Deep Learning for Finance: Deep Portfolios (2017)
  7. Application of a Multi-Factor Linear Regression Model for Stock Portfolio Optimization (2018)
  8. Deep learning (2015)
and 23 more
  1. Adam: A Method for Stochastic Optimization (2015)
  2. Squeeze-and-Excitation Networks (2020)
  3. A theory for multiresolution signal decomposition: the wavelet representation (1989)
  4. An Introduction to Wavelets and Other Filtering Methods in Finance and Economics (2001)
  5. Deep learning with long short-term memory networks for financial market predictions (2018)
  6. Empirical Asset Pricing via Machine Learning (2020)
  7. Deep Learning in Asset Pricing (2021)
  8. A LSTM-based method for stock returns prediction: A case study on S & P 500 (2019)
  9. Time Series Analysis: Forecasting and Control (2015)
  10. Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of the United Kingdom Inflation (1982)
  11. Financial time series forecasting using support vector machines (2003)
  12. Application of support vector machines in financial time series forecasting (2001)
  13. Deep learning with long short-term memory networks for financial market predictions (2018)
  14. Stock market’s price movement prediction with LSTM neural networks (2017)
  15. Portfolio Selection (1952)
  16. Deep learning in finance (2017) — link
  17. A deep reinforcement learning framework for the financial portfolio management problem (2017) — link
  18. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion (2010)
  19. A deep learning framework for financial time series using stacked autoencoders and long short-term memory (2017)
  20. An Introduction to Wavelet Theory in Finance: A Wavelet Multiscale Approach (2013)
  21. Stock price prediction via discovering multi-frequency trading patterns (2017)
  22. CBAM : Convolutional Block Attention Module (2018)
  23. Channel-wise attention network for time series classification (2020)

2 of 31 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.