Paper: arXiv 2401.06139
Authors: Bohan Ma, Yushan Xue, Yuan Lu, Jing Chen
Abstract
As the Chinese stock market continues to evolve and its market structure grows increasingly complex, traditional quantitative trading methods are facing escalating challenges. Particularly, due to policy uncertainty and the frequent market fluctuations triggered by sudden economic events, existing models often struggle to accurately predict market dynamics. To address these challenges, this paper introduces Stockformer, a price-volume factor stock selection model that integrates wavelet transformation and a multitask self-attention network, aimed at enhancing responsiveness and predictive accuracy regarding market instabilities. Through discrete wavelet transform, Stockformer decomposes stock returns into high and low frequencies, meticulously capturing long-term market trends and short-term fluctuations, including abrupt events. Moreover, the model incorporates a Dual-Frequency Spatiotemporal Encoder and graph embedding techniques to effectively capture complex temporal and spatial relationships among stocks. Employing a multitask learning strategy, it simultaneously predicts stock returns and directional trends. Experimental results show that Stockformer outperforms existing advanced methods on multiple real stock market datasets. In strategy backtesting, Stockformer consistently demonstrates exceptional stability and reliability across market conditions-whether rising, falling, or fluctuating-particularly maintaining high performance during downturns or volatile periods, indicating a high adaptability to market fluctuations. To foster innovation and collaboration in the financial analysis sector, the Stockformer model’s code has been open-sourced and is available on the GitHub repository: https://github.com/Eric991005/Multitask-Stockformer.
Complexity vs Empirical Score
- Math Complexity: 7.5/10
- Empirical Rigor: 6.0/10 (capped from 8.0: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.0)
Why this score: The model employs advanced techniques like wavelet transforms, self-attention networks, and graph embeddings, requiring significant mathematical sophistication. It is highly empirical with a reported backtested strategy on real market data, open-sourced code, and specific performance metrics across different market conditions.
Research Flowchart
flowchart TD
A["Research Goal:<br>Enhance stock selection in volatile markets"] --> B["Input Data:<br>Chinese Stock Market Prices & Volumes"]
B --> C["Key Methodology:<br>Discrete Wavelet Transform DWT"]
C --> D{"Decomposition"}
D --> E["High-Freq: Short-term fluctuations & abrupt events"]
D --> F["Low-Freq: Long-term market trends"]
E & F --> G["Multi-Task Self-Attention Network"]
G --> H["Computational Process:<br>Dual-Frequency Spatiotemporal Encoder & Graph Embedding"]
H --> I["Key Outcomes:<br>Simultaneous Prediction of Returns & Direction"]
I --> J["Findings:<br>Superior performance & stability across market conditions"]
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 Stockformer, a price‑volume factor stock selection model that combines wavelet transformation with a multitask self‑attention network.
Source sentence
this paper introduces Stockformer, a price-volume factor stock selection model that integrates wavelet transformation and a multitask self-attention network
Assumes 5
- The model assumes industry and market value neutralization removes industry and size effects. model
Source sentence
Industry and market value neutralization are employed to eliminate the influence of a stock’s industry and market value.
- The backtesting assumes a transaction cost of 0.1% per side (with stamp duty changes). execution
Source sentence
The transaction cost is 0.1% per side, with stamp duty at 0.1% before August 27, 2023, and 0.05% thereafter.
- The data split uses a rolling window of two‑year training, three‑month validation, and three‑month test periods. data
Source sentence
the data is divided into two years of trading data as the training set (486 days), followed by three months of trading data for the validation set (81 days), and another three months for the test set (81 days)
- Only stocks that are not delisted, have positive net assets, are not newly issued, and are not under risk warning are included. data
Source sentence
Stocks that were delisted, had negative net assets, were newly issued, or were under risk warning were excluded.
Show 1 more
- The TopK‑Dropout strategy assumes that selecting the top‑k ranked stocks yields better performance. model
Source sentence
The core idea of the Topk-Dropout strategy is to select stocks that rank high in factor values, assuming these stocks are more likely to perform well in the future.
Data
- CSI 300 index constituents market-prices public
Source sentence
We utilize stock market data from March 1, 2018, to January 30, 2024, for analysis.
Qlib datasets
Costs
Unclear - sources disagree
Source sentence (the authors' words)
The transaction cost is 0.1% per side, with stamp duty at 0.1% before August 27, 2023, and 0.05% thereafter.
Finds 3
- Stockformer achieves a directional accuracy of 57.46%, higher than all baselines. predictive-accuracy
Source sentence
Notably, it achieves a directional accuracy of 57.46%, significantly higher than other models.
- In backtesting, Stockformer yields annualized returns of 239.73% (rising market), -15.18% (falling market), and 44.48% (volatile market). performance
Source sentence
Backtesting results indicate that the Stockformer-based stock selection model achieves annualized returns of 239.73%, -15.18%, and 44.48% in rising, falling, and fluctuating markets, respectively
- Maximum drawdowns are 3.07% (rising), 19.6% (falling), and 4.14% (volatile) with Sharpe ratios of 8.46, -0.56, and 2.71 respectively. performance
Source sentence
with maximum drawdowns of 3.07%, 19.6%, and 4.14%; and Sharpe ratios of 8.46, -0.56, and 2.71.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 37 related papers). Coverage is partial, so this is neither replication nor refutation.
3 papers in the archive
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.