Research map / Time series / Long-term forecasting
MASTER: Market-Guided Stock Transformer for Stock Price Forecasting
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MASTER, a market-guided stock transformer that alternates intra-stock and inter-stock attention over time steps and uses a market-status gate for feature selection, improving stock return ranking and portfolio metrics on CSI300/CSI800.
Problem
Stock price forecasting fails to capture momentary and cross-time inter-stock correlations and ignores how changing market conditions alter feature effectiveness.
Approach
Alternate intra-stock aggregation (across time steps per stock) and inter-stock aggregation (across stocks per time step) to approximate the full stock-by-time attention field cheaply, and gate input features with a softmax scaling vector computed from market index statistics.
Key results
- On CSI300, MASTER reaches IC 0.064 and RankIC 0.076 versus best baseline IC 0.054 (GAT) and RankIC 0.052 (DTML/GRU).
- On CSI800, MASTER achieves AR 0.28 and IR 2.3, against the best baseline AR 0.16 and IR 1.3 (DTML).
- The paper reports about 13% average improvement on ranking metrics and 47% on portfolio metrics over the second-best results.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MASTER instead of GAT | stock price forecasting · CSI300 | IC: 0.054±0.002 -> 0.064*±0.006 | Helped |
| Use MASTER instead of GRU | stock price forecasting · CSI300 | RankIC: 0.052± 0.005 -> 0.076*± 0.005 | Helped |
| Use MASTER instead of XGBoost | stock price forecasting · CSI300 | AR: 0.23±0.03 -> 0.27±0.05 | Helped |
| Use MASTER instead of LSTM | stock price forecasting · CSI300 | IR: 2.0 ± 0.4 -> 2.4±0.4 | Helped |
| Use MASTER instead of GAT | stock price forecasting · CSI800 | IC: 0.043± 0.002 -> 0.052*±0.006 | Helped |
| Use MASTER instead of DTML | stock price forecasting · CSI800 | RankIC: 0.053± 0.008 -> 0.066± 0.007 | Helped |
| Use MASTER instead of DTML | stock price forecasting · CSI800 | AR: 0.16±0.03 -> 0.28*±0.02 | Helped |
| Use MASTER instead of DTML | stock price forecasting · CSI800 | IR: 1.3±0.2 -> 2.3*±0.3 | Helped |
| Add market-guided gating (ablation) | MASTER · CSI300 | IC: 0.064±0.003 -> 0.064±0.003 | No clear effect |
| Add market-guided gating (ablation) | MASTER · CSI300 | AR: 0.25±0.03 -> 0.27±0.05 | Helped |
| Replace the transformer encoder with a bidirectional LSTM (ablation) | MASTER · CSI300 | IC: 0.064±0.003 -> 0.058±0.005 | Hurt |
| Replace alternating intra-stock and inter-stock aggregation with direct attention over all time-stock tokens (ablation) | MASTER · CSI300 | IC: 0.064±0.003 -> 0.041±0.008 | Hurt |
Surprising findings
- Plain Transformer attains the highest ICIR and RankICIR on CSI800 among baselines, slightly above MASTER on those two metrics.
Limitations
- Naive full-field attention and clustering ablations were run only on CSI300 due to resource limits.
- Relies on Alpha158 features and market index data; improvements are measured only on the Chinese market.
Experimental setup
Datasets: CSI300 CSI800 Alpha158
Benchmarks: CSI300 CSI800
Metrics: IC ICIR RankIC RankICIR AR IR
Baselines: XGBoost LSTM GRU TCN Transformer GAT DTML Naive Clustering (MA)STER-Bi
Builds on: Transformer Qlib DTML Local Sensitive Hashing
Closest papers by meaning
- StockMixer: A Simple Yet Strong MLP-Based Architecture for Stock Price Forecasting (AAAI 2024 · similarity 0.88)
- CI-STHPAN: Pre-trained Attention Network for Stock Selection with Channel-Independent Spatio-Temporal Hypergraph (AAAI 2024 · similarity 0.87)
- DHMoE: Diffusion Generated Hierarchical Multi-Granular Expertise for Stock Prediction (AAAI 2025 · similarity 0.85)
- MDGNN: Multi-Relational Dynamic Graph Neural Network for Comprehensive and Dynamic Stock Investment Prediction (AAAI 2024 · similarity 0.85)
- ECHO-GL: Earnings Calls-Driven Heterogeneous Graph Learning for Stock Movement Prediction (AAAI 2024 · similarity 0.84)
- Market-GAN: Adding Control to Financial Market Data Generation with Semantic Context (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.