Adding Mean Reversion to a Momentum System: Cointegrated Pairs as Diversifier

Why a Momentum System Needs Mean Reversion

V7 is fundamentally a momentum system. L1 identifies directional signals, L2 gates timing, L3 manages exits. All layers assume the market will move in the predicted direction. But momentum systems have a known weakness: they suffer during range-bound periods when prices chop without direction.

S18 adds a complementary strategy: cointegrated pairs trading. When two instruments share a long-run equilibrium relationship (cointegration), deviations from that equilibrium tend to revert. This mean-reversion alpha is uncorrelated with momentum alpha, providing genuine diversification.

Engle-Granger and the 12 Validated Pairs

S18 uses the Engle-Granger two-step test to identify cointegrated pairs. First, regress one instrument on the other. Second, test the residuals for stationarity using the ADF test. Pairs with ADF p-value below 0.05 are considered cointegrated.

Out of all possible instrument combinations, 12 pairs passed cointegration tests consistently across multiple walk-forward windows. These include obvious pairs (EURUSD/GBPUSD, AUDUSD/NZDUSD) and some less obvious ones (USDJPY/US30, GOLD/SILVER). The 12 pairs generated 15.2R additional return over the backtest period.

A Diversifier, Not a Replacement

The 15.2R contribution is small relative to the 533.9R total from the main momentum pipeline. S18 is not trying to compete with the core system. It is providing returns during the specific periods when the core system is flat or losing. During range-bound regimes (identified by S01, S04, and S09), the pairs strategy often generates positive R while the momentum system sits on the sidelines. This counter-cyclical alpha is the whole point. Over any single month, S18 might add or subtract 1-2R. Over years, the diversification benefit shows up in smoother equity curves and lower max drawdown.