Trading ETF-Stock Residuals with a Three-Leg Statistical Arbitrage Model
Summary
The article proposes a relative-value strategy linking the South Korea ETF EWY with Samsung Electronics and SK Hynix. It explains why the author avoids trading the visible cash-equity versus perpetual-contract basis: trading-hour mismatches, currency conversion, funding, liquidity, and execution constraints make the quoted spread difficult to capture, while overnight pricing may reflect new information rather than mispricing. The alternative trades three USDT-margined contracts in one account. A rolling regression of log EWY price on the log prices of the two stocks estimates an implied basket value; the standardized regression residual is the signal, with hedge sizes derived from the coefficients.
The article describes data and numerical safeguards, including rejecting ill-conditioned regressions, as well as risks from omitted ETF holdings, asynchronous markets, and multi-leg execution. It reports that the strategy was run in simulation and mentions a small simulated profit, but the excerpt does not provide enough performance detail to establish robustness. The author stresses that this is statistical relative-value trading, not risk-free arbitrage, and that economic rationale, out-of-sample persistence, and costs must all be validated before live use.
Key ideas
- The proposed trade models EWY as a function of Samsung Electronics and SK Hynix using log prices.
- The standardized regression residual, rather than the outright price direction, is the intended trading signal.
- Keeping all three legs in one execution environment reduces some operational frictions but does not remove market risk.
- Collinearity can destabilize hedge coefficients, so ill-conditioned regression results invalidate a signal.
- Simulation evidence is limited, and live viability depends on out-of-sample stability and execution costs.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.