Paper: arXiv 2601.13435

Authors: Shuozhe Li, Du Cheng, Leqi Liu

Abstract

Learning profitable intraday trading policies from financial time series is challenging due to heavy noise, non-stationarity, and strong cross-sectional dependence among related assets. We propose \emph{WaveLSFormer}, a learnable wavelet-based long-short Transformer that jointly performs multi-scale decomposition and return-oriented decision learning. Unlike standard time-series forecasting that optimizes prediction error and typically requires a separate position-sizing or portfolio-construction step, our model directly outputs a market-neutral long/short portfolio and is trained end-to-end on a trading objective with risk-aware regularization. Specifically, a learnable wavelet front-end generates low-/high-frequency components via an end-to-end trained filter bank, guided by spectral regularizers that encourage stable and well-separated frequency bands. To fuse multi-scale information, we introduce a low-guided high-frequency injection (LGHI) module that refines low-frequency representations with high-frequency cues while controlling training stability. The model outputs a portfolio of long/short positions that is rescaled to satisfy a fixed risk budget and is optimized directly with a trading objective and risk-aware regularization. Extensive experiments on five years of hourly data across six industry groups, evaluated over ten random seeds, demonstrate that WaveLSFormer consistently outperforms MLP, LSTM and Transformer backbones, with and without fixed discrete wavelet front-ends. On average in all industries, WaveLSFormer achieves a cumulative overall strategy return of $0.607 \pm 0.045$ and a Sharpe ratio of $2.157 \pm 0.166$, substantially improving both profitability and risk-adjusted returns over the strongest baselines.

Complexity vs Empirical Score

  • Math Complexity: 5.5/10
  • Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs not modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper introduces an innovative learnable wavelet-Transformer (WaveLSFormer) that replaces fixed DWT preprocessing and directly optimizes trading P&L rather than forecasting MSE. The empirical evaluation is strong, featuring five years of hourly intraday U.S. equity data across multiple industry sectors with explicit variance reporting across 10 random seeds.

Research Flowchart

  flowchart TD
    A["Research Goal: Direct End-to-End Learning of Market-Neutral Long-Short Trading Positions"] --> B["Input Data: 5 Years Hourly Financial Time Series (6 Industry Groups)"]
    
    B --> C["Learnable Wavelet Front-End<br/>(Multi-scale decomposition with spectral regularizers)"]
    
    C --> D["Low-Guided High-Frequency Injection (LGHI)<br/>(Fuses high-frequency cues into low-frequency representations)"]
    
    D --> E["Portfolio Output & Rescaling<br/>(Market-neutral long/short weights bounded by risk budget)"]
    
    E --> F["End-to-End Optimization<br/>(Directly optimizes trading objective with risk-aware regularization)"]
    
    F --> G["Key Outcomes: Outperforms MLP/LSTM/Transformers<br/>(Cumulative Return: 0.607 ± 0.045, Sharpe Ratio: 2.157 ± 0.166)"]

Cited References (48)

  1. \href http://ijr.sagepub.com/content/9/2/62.abstract Passive Dynamic Walking (1990) — link
  2. A New Approach to Linear Filtering and Prediction Problems (1960) — link
  3. Stock market prediction performance of neural networks: A literature review (2017) — link
  4. Simple technical trading rules and the stochastic properties of stock returns (1992) — link
  5. Persuasive evidence of market inefficiency (1998)
  6. A comparison of artificial neural network and time series models for forecasting commodity prices (1996) — link
  7. Informer: Beyond efficient transformer for long sequence time-series forecasting (2021) — link
  8. A study on fundamental and technical analysis (2013)
and 40 more
  1. Stock market prediction using different neural network classification architectures (1996) — link
  2. Integrating Fundamental and Technical Analysis of Stock Market through Multi-layer Perceptron (2018) — link
  3. Deep learning for time series classification: a review (2019) — link
  4. Time series classification from scratch with deep neural networks: A strong baseline (2017) — link
  5. Gradient-based learning applied to document recognition (1998) — link
  6. Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling (2014) — link
  7. Long short-term memory (1997) — link
  8. CNNpred: CNN-based stock market prediction using a diverse set of variables (2019) — link
  9. Recurrent Neural Networks in Time Series Prediction (2018)
  10. A CNN-LSTM-based model to forecast stock prices (2020) — link
  11. Attention is all you need (2017) — link
  12. Gradient flow in recurrent nets: the difficulty of learning long-term dependencies (2001)
  13. A new statistic and practical guidelines for nonparametric Granger causality testing (2006) — link
  14. Time2Vec: Learning a Vector Representation of Time (2019) — link
  15. Efficient Capital Markets: A Review of Theory and Empirical Work (1970) — link
  16. A Comparison of Artificial Neural Network and Time Series Models for Forecasting Commodity Prices (1996) — link
  17. Deep Learning for Financial Applications: A Survey (2020) — link
  18. Deep Learning Models for Price Forecasting of Financial Time Series: A Review of Recent Advancements (2024) — link
  19. Deep Learning for Financial Forecasting: A Review of Recent Trends (2025) — link
  20. Deep Neural Networks, Gradient-Boosted Trees, Random Forests: Statistical Arbitrage on the S&P 500 (2017) — link
  21. Deep Learning Models for Financial Time Series Forecasting: A Comparative Study (2023)
  22. Learning to Trade in Financial Time Series Using High-Frequency Through Wavelet Transformation and Deep Reinforcement Learning (2021) — link
  23. Predicting Stock Movements: Using Multiresolution Wavelet Reconstruction and Deep Learning in Neural Networks (2021) — link
  24. Financial Time Series Forecasting Using Optimised Wavelet Transform and Neural Network Models (2022) — link
  25. A Hybrid Approach of Wavelet Transform, ARIMA and LSTM Model for the Share Price Index Futures Forecasting (2024)
  26. Enhancing Financial Time Series Forecasting: A Comparative Study of Discrete Wavelet Transform and LSTM Models for Selected Global Indices (2025) — link
  27. Wavelet-Integrated Deep Neural Networks: A Systematic Review of Applications and Synergistic Architectures (2025) — link
  28. W-Transformers: A Wavelet-Based Transformer Framework for Univariate Time Series Forecasting (2022) — link
  29. Wavelet Transformer: An Effective Method on Multiple Periodic Decomposition for Time Series Forecasting (2025) — link
  30. Wavelet-Enhanced Transformer for Adaptive Multi-Period Time Series Forecasting (2025)
  31. Stockformer: A price-volume factor stock selection model based on wavelet decomposition and multi-task self-attention (2025) — link
  32. Prediction of Stock Market Price Using Hybrid of Wavelet Transform and Artificial Neural Network (2016) — link
  33. Stock market’s price movement prediction with LSTM neural networks (2017) — link
  34. Stock Market Prediction Using LSTM Recurrent Neural Network (2020) — link
  35. A deep learning framework for financial time series using stacked autoencoders and long-short term memory (2017) — link
  36. Wavelet Neural Network Prediction Method of Stock Price Trend Based on Rough Set Attribute Reduction (2018) — link
  37. Predicting Stock Movements: Using Multiresolution Wavelet Reconstruction and Deep Learning in Neural Networks (2021) — link
  38. Deep Learning and Wavelets for High-Frequency Price Forecasting (2018) — link
  39. LSTM -Transformer-Based Robust Hybrid Deep Learning Model for Financial Time Series Forecasting (2025) — link
  40. Predictive Modeling of Stock Prices Using Transformer Model (2024) — link

0 of 48 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.