국내 금현물 프리미엄을 활용한 AI 기반 페어트레이딩 전략

초록

This study proposes an AI-based pairs trading strategy that exploits the premium between the Korea Exchange (KRX) gold spot market and international gold prices. To address the limitations of fixed-threshold strategies under changing volatility and mean-reversion dynamics, the study develops a dynamic threshold framework that learns threshold coefficients directly from historical premium time series using AI models. Using daily data from 2016 to 2025, the results show that the domestic gold premium exhibits statistically significant mean reversion and predictable temporal dependence. AI-based dynamic threshold strategies outperform fixed-threshold strategies in risk-adjusted performance, and the proposed Fusion model records the highest Sharpe and Sortino ratios under both LBMA- and Tenfore-based signals. SPA and MCS tests confirm that the proposed strategy remains within the superior model set, indicating that adaptive threshold learning can enhance relative-value trading performance under changing market conditions.

키워드