High-Frequency Stock Pairs Trading: Final Research Report

He-He Ren

Abstract

This report presents a high-frequency pairs-trading study centered on tickers 600606 and 600604. The project progressed from a literature review and initial minute-level demonstration to a systematic analysis of 3,626 daily MATLAB files covering 2005 through 2020. Returns were constructed at six sampling frequencies to examine how measured comovement changes with the observation interval. The analysis then compared return correlations, normalized the two price series, constructed their relative spread, and adjusted historical prices for corporate actions using the latest available split and dividend dictionary. A rolling 20-trading-day Z-score framework was subsequently applied to generate mean-reversion trading signals. The backtest used fixed $500 long and $500 short positions and evaluated cumulative PnL, total return, winning probability, win-to-loss ratio, maximum drawdown, and Sharpe ratios. The results showed that return correlation generally increased as the sampling interval became longer. After corporate-action adjustment, the trading strategy produced net PnL of $1,665.10 and a net return of 166.51% at a 0.07% round-trip transaction cost. Transaction-cost robustness remained positive through 0.20%, with an estimated break-even round-trip cost of 0.90254%. Overall, the study provides a transparent empirical implementation of high-frequency pairs trading and evaluates its performance under realistic transaction-cost assumptions.

Source: semanticscholar · PDF

Read the AI summary and key takeaways for traders on WOBR Quant Research.