本研究旨在建立一套結合 AI 機器學習與統計模型的配對交易策略，針對台灣股市進行實證分析。傳統配對交易多依賴靜態價格偏離邏輯，忽略市場波動環境變化與進場風險控制。為改善此問題，本研究運用 OPTICS 分群演算法篩選潛在配對組合，並以 VECM 協整模型建立價格長期均衡關係，再透過 AI/機器學習模型中的 XGBoost模型預測短期波動度，設計四種具動態反應能力的進場邏輯。實證結果顯示，結合短期波動預測可有效提升交易訊號品質與報酬穩定性，亦補足既有文獻在策略進場時機與亞洲市場應用上的研究空白。;This study aims to develop a pairs trading strategy that integrates AI-based machine learning with statistical modeling, and conducts an empirical analysis on the Taiwan stock market. Traditional pairs trading strategies often rely on static price deviation logic, overlooking the effects of market volatility and entry risk control. To address these issues, this research employs the OPTICS clustering algorithm to identify potential trading pairs, and uses the Vector Error Correction Model (VECM) to establish long-term price equilibrium relationships. Furthermore, the XGBoost model, a machine learning technique, is applied to forecast short-term volatility and design four dynamic entry strategies. Empirical results show that incorporating short-term volatility predictions significantly improves the quality of trading signals and the stability of returns. This approach also fills the research gap in the timing of trade entry and the application of pairs trading strategies in Asian markets.
