Three straight down days on IWM — Published Backtest & Verdict

The bounce is real but small — about 0.7 points of edge per signal over 10 sessions versus a random day. Compounded across 77 trades at only 39.9% time in market, the system beat buy-and-hold by 57 points (+153.5% vs +96.8%) and cut the max drawdown roughly in half (-20.7% vs -41.1%): the win came as much from sitting out 60% of the time as from the bounce itself. The worst trade bought three down days into the COVID crash and lost -14.0% in 10 sessions — three down days can always become ten. No costs are modeled; 77 round trips of real costs and slippage would take a real bite out of the gap. We publish losers too — this one happened to win.

• Total return: +153.5% (beat buy-and-hold at +96.8%)
• CAGR: +12.9% vs +9.2% buy-and-hold
• Max drawdown: -20.7% vs -41.1% buy-and-hold
• Win rate: 61% over 77 trades
• Period: 2019-03-06 to 2026-11-06 (IWM), published 2026-11-07

Cost assumption: slippage and commissions aren't modeled — fills are assumed at the exact closing price with no trading fees, the same assumption stated on How We Backtest. Real execution will differ, usually for the worse.

The claim on trial

Small caps bounce after three straight down days — buying the third down close and holding 10 sessions beats buy-and-hold.

The exact rules

• A down session is a close below the prior session's close. Signal = the close of the third consecutive down session, first signal per streak only (a longer streak does not signal again).
• Claim test (event study): forward returns at 5 and 10 sessions from all 101 signal closes vs the all-days baseline; split by whether the signal close was below its 200-day SMA (93 classifiable; the first 8 signals predate a computable 200-day average).
• Tradable version: buy at the signal close, hold exactly 10 sessions, exit at that close; signals while invested are ignored. Entry starts earning the next bar (no look-ahead); exit retains the exit bar's return.
• Long only, fully invested when in, cash earns 0%. No transaction costs, no taxes modeled (disclosed — 77 round trips would incur real costs).

Our honest verdict

The bounce is real but small — about 0.7 points of edge per signal over 10 sessions versus a random day (101 signals: +1.20% mean vs +0.51% baseline). Compounded across 77 trades at only 39.9% time in market, the system beat buy-and-hold by 57 points (+153.5% vs +96.8%) and cut the max drawdown roughly in half (-20.7% vs -41.1%): the win came as much from sitting out 60% of the time as from the bounce itself. Signals below the 200-day average did at least as well as those above it (+1.39% vs +1.24% mean at 10 sessions). Worst trade bought three down days into the COVID crash (-14.00%). No costs modeled; 77 round trips of real costs would take a real bite.

Full trade log on Substack · Original X thread · StratIQ track record · All published backtests

Charted on TradingView (affiliate link — we may earn a commission at no extra cost to you): tradingview.com

Not financial advice — research and education only; do your own due diligence.