20240328-浙商证券-金融工程深度_学界纵横系列之七-含交叉注意力机制的趋势预测模型_14页_1mb

Summary of X-Trend: Cross-Attentive Time-Series Trend Network

Core Viewpoints

• X-Trend, a model with a cross-attention mechanism, enhances trend prediction in financial markets, offering superior risk-adjusted returns and faster recovery from drawdowns.

Key Features

• Adapts to market changes via context sets and cross-attention, outperforming traditional models and deep momentum networks.
• Achieves higher Sharpe ratios (up to 189% improvement) and quicker recovery during volatile periods.

Findings

• In the few-shot scenario, X-Trend significantly outperforms benchmarks, demonstrating strong small-sample learning capabilities.
• The cross-attention mechanism allows the model to apply insights from historical patterns to new assets and market conditions.
• Performance varies in zero-shot settings due to sample distribution differences.

Risk Considerations

• Backtested results may not reflect future performance. The model is not an investment recommendation.

Conclusion

X-Trend improves market condition adaptation through cross-attention, providing a robust method for trend-based trading strategies.