多空對沖策略:應用循環神經網路於量化交易

Abstract

循環神經網路(RNN)由於其分析序列數據和從資料中提取模式以進行預測的能 力，而在量化交易中變得越來越受歡迎。RNN 的模型，例如長短期記憶(LSTM)和 門控循環單元(GRU)，已經在金融界被廣泛使用，包括股票價格預測、投資組合優 化和風險管理等等。本文介紹了 RNN 被應用在量化交易中的概念，包括 LSTM 和 GRU 模型的結構和訓練過程，也利用模型對於隔日報酬的預測，建立並回測每日多空 對沖策略的表現;結果顯示 LSTM 在報酬、回撤和其他信息相關的指標方面，均優於 GRU 模型和基準指數。此外，為了控制市場風險，也將 Fama-French 三因子模型應用 至預測報酬，而構建的投資組合在納入因子後，仍表現穩定持續的高報酬。總體而言， 本文提供了在量化交易中使用 RNN 的模擬實例，並突出了 LSTM 和 GRU 模型在實 現盈利交易策略方面的潛力。

Recurrent Neural Networks (RNNs) have gained popularity in quantitative trading due to their ability to analyze sequential data and extract patterns to make predictions. RNN- based models, such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), have been applied to various financial applications, including stock price prediction, portfolio optimization, and risk management. This paper provides an overview of RNNs in quantitative trading, including the architecture and training process of LSTM and GRU models. Daily long-short hedging strategies are built according to model predictions on next-day returns, with LSTM outperforming GRU models and benchmark indicators in terms of return, draw- down, and other information-related metrics. In addition, Fama-French three-factor model is applied to predicted returns to account for market risk, and the constructed portfolios demon- strate consistent high returns after the factor inclusion. Overall, this paper provides valuable insights into the use of RNNs in quantitative trading and highlights the potential of LSTM and GRU models for achieving profitable trading strategies.