Cointegration vs. Correlation in Pairs Trading
0:00 / 4:23
Necessary but not sufficient
0:00 - 0:27
Pairs trading needs a stationary spread; Brandes tests seven U.S. equity pairs and shows correlation and costs can still break the trade.
Correlation is not cointegration
0:27 - 1:02
SPY/QQQ look nearly perfectly correlated but fail every cointegration test and produce a negative walk-forward Sharpe.
The ≥2-of-3 gate
1:02 - 1:39
Engle-Granger, Johansen, and Phillips-Ouliaris in log-price space; only GOOG/GOOGL and V/MA pass.
β stability vs. regime luck
1:39 - 2:20
GOOG/GOOGL’s hedge ratio barely moves across folds; GLD/GDX’s drifts and its Sharpe rides gold-bull regimes.
Cost-fragility
2:20 - 3:08
Tight GOOG/GOOGL spreads break even near 15 bps; wider V/MA survives to about 37 bps.
Signal decay
3:08 - 3:46
Rule entries and a placebo of extreme z-scores decay the same way—reversion comes from the spread.
Gate, then stress-test
3:46 - 4:22
Cointegration filters fake pairs; β stability and cost tolerance decide what is actually tradeable.