Alpha Strike

Alpha Strike backtests a cointegrated pairs-trading strategy across the S&P 500: sector-grouped candidate pairs are filtered by correlation and Engle-Granger cointegration, then evaluated on a 24-variant parameter grid (z-score lookbacks, entry/exit thresholds) with a 2-year rolling train / 6-month test walk-forward backtest, 5bps transaction costs, and vol-targeted position sizing. Of 5,112 strategy variants tested from 2020 to 2026, 1,051 (20.6%) look profitable on a naive Sharpe ratio, but zero survive a Deflated Sharpe Ratio correction above the 0.95 significance bar — the actual point of the project: a backtest that tries enough parameter combinations will always produce something that looks like edge. The single best naive Sharpe in the universe (4.38, on utility pair AEE/LNT) carries only a 41% probability of reflecting real skill once the Deflated Sharpe Ratio accounts for it being the best of 5,112 attempts rather than the only one, even though a permutation test on that pair in isolation rejects the noise null on its own terms (p < 0.002), illustrating how selection bias and single-test significance answer different questions. Results ship with explicit caveats for survivorship bias, non-point-in-time sector classification, and data-vendor methodology differences. Backed by 28 unit tests, mypy and ruff clean, with the full pipeline reproducible in about 2 minutes on a warm cache.