Paper: arXiv 2410.20679
Authors: Peng Zhu, Yuante Li, Yifan Hu, Sheng Xiang, Qinyuan Liu, Dawei Cheng, Yuqi Liang
Abstract
As financial markets grow increasingly complex in the big data era, accurate stock prediction has become more critical. Traditional time series models, such as GRUs, have been widely used but often struggle to capture the intricate nonlinear dynamics of markets, particularly in the flexible selection and effective utilization of key historical information. Recently, methods like Graph Neural Networks and Reinforcement Learning have shown promise in stock prediction but require high data quality and quantity, and they tend to exhibit instability when dealing with data sparsity and noise. Moreover, the training and inference processes for these models are typically complex and computationally expensive, limiting their broad deployment in practical applications. Existing approaches also generally struggle to capture unobservable latent market states effectively, such as market sentiment and expectations, microstructural factors, and participant behavior patterns, leading to an inadequate understanding of market dynamics and subsequently impact prediction accuracy. To address these challenges, this paper proposes a stock prediction model, MCI-GRU, based on a multi-head cross-attention mechanism and an improved GRU. First, we enhance the GRU model by replacing the reset gate with an attention mechanism, thereby increasing the model’s flexibility in selecting and utilizing historical information. Second, we design a multi-head cross-attention mechanism for learning unobservable latent market state representations, which are further enriched through interactions with both temporal features and cross-sectional features. Finally, extensive experiments on four main stock markets show that the proposed method outperforms SOTA techniques across multiple metrics. Additionally, its successful application in real-world fund management operations confirms its effectiveness and practicality.
Complexity vs Empirical Score
- Math Complexity: 6.0/10
- 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 paper introduces novel architectural modifications (attention-based GRU gate, multi-head cross-attention) and uses specific datasets (CSI 300/500, NASDAQ 100, S&P 500) with multiple metrics, coupled with a successful real-world fund management deployment, demonstrating high empirical rigor; the mathematical complexity is moderate to high due to the integration of GATs and custom attention mechanisms without overly dense derivations.
Research Flowchart
flowchart TD
A["Research Goal:<br>Improve Stock Prediction Accuracy"] --> B{"Data Input"}
B --> C["Four Main Stock Markets<br>Time Series & Cross-Sectional Data"]
C --> D["Step 1: Improved GRU<br>Reset Gate replaced with Attention"]
D --> E["Step 2: Multi-Head Cross-Attention<br>Learn Latent Market States"]
E --> F["Model Output:<br>MCI-GRU Architecture"]
F --> G["Key Outcomes"]
G --> H["SOTA Performance on<br>4 Stock Markets"]
G --> I["Successful Real-World<br>Fund Management Application"]
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 stock prediction model, MCI-GRU, that combines a multi-head cross-attention mechanism with an improved GRU architecture.
Source sentence
this paper proposes a stock prediction model, MCI-GRU, based on a multi-head cross-attention mechanism and an improved GRU.
Assumes 3
- Transaction costs are ignored in the simulated trading strategy. execution
Source sentence
Transaction costs are excluded from consideration in this simulation.
- Predictions are generated at the close of day t and trades are executed at the open of day t+1. execution
Source sentence
... At the opening of trading day t + 1, traders liquidate the stocks purchased on day t and acquire those ranked in the top-k.
- Historical features are taken from the previous 60 trading days to predict returns over the next 21 trading days. data
Source sentence
we employ features derived from the previous 60 trading days to predict stock return rankings over the next 21 trading days.
Data
equities · daily
- CSI300 market-prices unknown
Source sentence
... Coverage Ratio Covers 70% of China’s A-share market cap
- NASDAQ100 market-prices unknown
Source sentence
... Coverage Ratio Covers 90% of NASDAQ’s total market cap
Costs
Not modeled
Source sentence
Transaction costs are excluded from consideration in this simulation.
Finds 3
- MCI-GRU achieves the highest annualized rate of return (ARR) and superior risk‑adjusted metrics across the CSI300 and CSI500 datasets. performance
Source sentence
Our proposed model demonstrates a significant performance advantage, achieving the highest ARR (0.352 for CSI 300 and 0.330 for CSI 500) and exhibiting superior risk‑adjusted returns, as indicated by the highest ASR (1.559 for CSI 300 and 1.626 for CSI 500), CR (2.776 for CSI 300 and 1.663 for CSI 500), and IR (1.526 f
- MCI‑GRU attains the lowest MAE and MSE on the NASDAQ 100 dataset, indicating superior prediction accuracy. predictive-accuracy
Source sentence
Among all the models, MCI‑GRU performs exceptionally well in terms of MAE and MSE, particularly on the NASDAQ 100 dataset, where its MAE and MSE are the lowest, indicating a clear advantage in prediction accuracy.
- The model outperforms existing state‑of‑the‑art methods across multiple metrics on four major stock markets. performance
Source sentence
the experimental results demonstrate that the proposed MCI‑GRU model outperforms existing state‑of‑the‑art methods across multiple key performance metrics.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 29 related papers). Coverage is partial, so this is neither replication nor refutation.
Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.