110 strategy ideas against the same strict checks
retained text
That list is at 110 ideas now. 103 rejected, 5 still running forward on paper, 2 of them are futures and cant be traded with Alapaca. The checks: \- costs charged on both sides of every trade \- held-back data the idea never saw while I tuned it \- has to beat just holding the same thing \- has to beat random entries with the same exits \- if I tried 500 versions, the best one has to clear a higher bar, because the best of 500 looks good by luck \- for the newer tests, the pass/fail rule is committed to git before the test runs What killed them (chart above, 99 of the rejections reviewed by hand): \- 49 had no edge before costs. nothing there, or too weak to tell from noise \- 21 fell apart on the held-back data \- 12 died to costs and realistic fills \- 7 didn't beat holding \- the rest: best of many tries, no better than random entries, drawdown, or one market episode carrying the whole result I expected costs to be the main killer. They weren't. Half the ideas had nothing to rescue even at zero cost. A few specific ones: \- opening range breakouts on stocks, 672 versions. 0 passed, not one positive after costs \- bull flag continuation: Sharpe 0.35 vs 0.81 for SPY, with a deeper drawdown (-50.5% vs -34.2%) \- buying after two 2%+ down days looked strong (t 3.1) until I grouped trades by episode. then t was 1.98 against a bar of 2.0, and the best 5% of trade days carried more than all the profit. counted it as a fail \- picking last quarter's best strategies and holding them next quarter looked persistent, then 5 more bps of cost erased it \- defensive sectors on high vol days: no better than just holding less SPY at the same risk
document fetched 2026-11-18 00:48 · canonical https://reddit.com/r/algotrading/comments/1xyr3sm/110_strategy_ideas_against_the_same_strict_checks