A Simple 7-Day ROC Rebound Rule on SPY: 71% Win Rate Across 33 Years and 146 Trades

The question is disarmingly simple and spans 33 years of SPY from 1993 to 2026: after a 7-day slide of 3%, does the market tend to snap back? Quantified Strategies does not use ROC (Rate of Change — a momentum gauge that compares today's close to the close n days ago) to chase strength; it flips the lens and asks whether a sharp weekly decline creates a short-term rebound edge. SPY, launched in January 1993 as the first major U.S. exchange-traded fund tracking the S\u0026P 500 and now holding well over $500 billion in assets with enormous daily volume, offers a continuous, highly liquid price history that makes such a long-horizon test feasible. That history is the laboratory — not a promise — for the hypothesis.

The math is elementary: ROC = [(today's close - close n days ago) / close n days ago] x 100. A 7-day reading of -3% means price fell 3% in a week (say $100 to $97); +3% means it recovered 3% over the same window (back from $97 toward $100); zero means unchanged. Think of a rubber band stretched too far and snapping back — that is the mean-reversion intuition (prices reverting toward an average) versus momentum (buying what already rose). Framing it this way makes the thresholds intuitive: -3% marks a stretched dip, +3% marks the snap-back confirmation, and the distance between them is the expected rebound path, not a trend-following breakout.

The rules are kept deliberately bare and run in three steps: 1) when 7-day ROC closes below -3% and the system is flat, buy SPY at the next session's open; 2) hold the single long position until 7-day ROC closes above +3%; 3) on that close, sell at the next session's open and return to cash. Signals form at the close, execution happens at the next open — no look-ahead to the same close. Only one position is held at a time, with no stop-loss and no profit target; the opposite ROC threshold is the sole exit. The simplicity leaves little room for hidden optimization, but it also means intraday reversals inside the -3% day are ignored and the next-open assumption can add slippage in fast markets.

The headline result before costs and dividends is striking: $1 compounded through the completed trades grows to about $9.15. Across 146 trades the win rate is 71.2%, the profit factor (gross profit divided by gross loss — above 1.0 means profitable, 2.0 means $2 earned per $1 lost) is 2.40, and the maximum drawdown (largest peak-to-trough decline) is about 29%. The equity curve climbs over the full span but with real valleys; a 29% pullback means months of underwater patience. For context, buy-and-hold SPY turned $1 into roughly $30 with dividends reinvested over the same era — this rule spends far less time in the market, harvesting only sharp dips, so it compounds to a lower but still positive level while sidestepping long sideways stretches.

Robustness is tested by varying the lookback from 4 to 11 days while keeping the -3% / +3% thresholds. Every variant stays profitable, with profit factors between 1.75 and 2.40 and trade counts between 122 and 157. The 7-day window is the strongest at 2.40, but neighbors are close, which suggests the outcome is not perched on a single isolated setting — the core signal survives a neighborhood of parameters. Low parameter sensitivity matters because market rhythm drifts; a rule that needs exactly seven days would be fragile (a sign of overfitting), while a broad plateau hints at a more durable edge, even if the exact peak shifts in the next decade.

A second sensitivity sweep fixes the entry below -3% and moves only the exit from +1% to +5%. The range remains profitable, with profit factors from 1.96 at the loosest +1% exit to 2.53 at the strictest +5% exit, while trade count falls from 183 to 80. A more demanding exit lifts per-trade expectancy but cuts frequency by more than half and keeps capital deployed longer between round-trips. The trade-off is clear: fewer, more selective rebounds versus more frequent, smaller snaps. Around +3% the balance sits in the middle — not too noisy like +1% and not too sparse like +5% — which is why the video anchors on it as a pragmatic compromise.

Split by era, the edge persists: 1993-2004 produced 54 trades with a 72.2% win rate and a 2.71 profit factor (dot-com bubble, crash and rebound), 2005-2015 delivered 51 trades with a 64.7% win rate and a 2.07 profit factor (global financial crisis, European debt stress and a long recovery), and 2016-2026 gave 41 trades with a 78.0% win rate and a 2.45 profit factor (pandemic crash, rapid recovery and the AI-led rally). Profit factors stay above 2.0 in all three, even though win rates swing; the weakest period, 2005-2015, is still profitable. That cross-regime consistency weakens the "one lucky bull market" objection, though the declining trade frequency to 41 in the most recent decade hints that sharp 7-day drops became rarer as volatility regimes changed.

The video is careful to frame this as a narrow, in-sample backtest and lists what it excludes: commissions, slippage (the gap between signal price and filled price), taxes and dividends, with no live or walk-forward out-of-sample validation. In other words, the rules fit the history they were measured on but have not been proven on unseen future data. The practical read is nuanced: for a disciplined, rule-based investor who wants a simple mean-reversion skeleton with few, well-defined signals, the framework is instructive; for anyone sensitive to costs, dividend drag or a 29% drawdown, it needs a cost- and dividend-adjusted walk-forward test before capital is risked — the laboratory curve is not a brokerage statement.