EWY Three-Leg Statistical Arbitrage: Execution, Model Drift, and Failure

Summary

This case study follows a live statistical-arbitrage strategy trading the residual between the EWY Korea ETF and Samsung Electronics and SK Hynix. A rolling log-price regression estimates hedge ratios; the residual is standardized into a z-score, with thresholds governing entry, staged grid exposure, and exit. The author describes changes from simulation to live execution, including concurrent orders, position reconciliation against exchange state, frozen regression parameters while a trade is open, and drift monitoring.

The account reports a profitable period followed by a sharp loss when the relationship broke down and the frozen z-score no longer represented a valid mean-reverting spread. It emphasizes that simulations omit real execution frictions, multi-leg positions carry unhedged exposure during fills, and rolling or frozen model parameters can disconnect signals from realized P&L. The author proposes drift-based trading halts, residual half-life monitoring, stability checks, and alternative pair or regression designs. This is one trader’s short, leveraged live record, not evidence that the strategy generalizes; execution costs, structural breaks, and model failure remain central risks.

Key ideas

• The strategy trades a regression residual between EWY and two major Korean equity constituents.
• Log prices make hedge coefficients interpretable as relative sensitivities, while residual z-scores drive mean-reversion entries and exits.
• Concurrent execution and repeated reconciliation reduce multi-leg exposure but do not make orders atomic.
• Freezing regression parameters aligns trade thresholds with a consistent coordinate system, but that model can become obsolete.
• Beta and volatility drift, residual half-life, and stability checks can help detect when the mean-reversion premise has failed.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.