Dynamic Cointegration in Pairs Trading: Evidence from Treasury and Equity Markets

This paper presents a comprehensive empirical study on pairs trading strategies using a dynamic cointegration framework applied to both U.S. Treasury yield curves and Chinese equity markets. Unlike traditional static approaches, we introduce a rolling-window cointegration analysis to monitor the temporal stability of statistical relationships, coupled with an adaptive backtesting system that incorporates Kalman and particle filters for dynamic parameter estimation. Our methodology rigorously tests the Engle-Granger and Johansen cointegration procedures, estimates mean reversion parameters via the Ornstein-Uhlenbeck process, and optimizes trading thresholds using a grid-search mechanism. Key innovations include the integration of time-varying parameter stability analysis, real-time structural break detection, and a multi-asset comparative performance evaluation. Empirical results reveal a stark performance divergence: while the equity pair (600036.SS–000001.SS) achieved a Sharpe ratio of 1.103 and a total return of 28.98%, the Treasury yield pair (TNX–TYX) yielded a negative Sharpe ratio of -0.767, underscoring the critical role of asset class selection and regime adaptability. The study concludes that static cointegration models are insufficient in modern markets, advocating instead for dynamic, regime-aware frameworks to enhance the robustness and profitability of statistical arbitrage strategies.