Backtested and hypothetical results, shown for educational purposes only — not investment advice. Past performance does not indicate future results.

Mean Reversion

show
hold for
US indices SPY QQQ IWM DIA International FXI EEM
GPI’s daily moves don’t reliably mean-revert. The dip-buy stats below are weak or noisy — at the daily close-to-close horizon this signal works cleanly on the broad equity indices (SPY, QQQ) and mostly breaks down on commodities and crypto.
Expected return · next day

GPI — Stretched, weak edge$278.27 · 2026-09-11

GPI follows through on the prior day (trends) at the daily horizon.

Current stretch: RSI(2) at 25 is leaning oversold, 2 down days in a row — but the daily edge here is thin.

Today
-1.3%
last close-to-close
RSI(2)
24.8
neutral
vs 50-day avg
-3.5%
-0.4σ
Streak
2 down
closes in a row
Off 20-day high
-9.2%
pullback depth
Reversal character: after an up day, next day averages +0.15% (51% up); after a down day, +0.03% (49% up). Reversal score -0.12%, 1-day autocorrelation 0.023 — positive = trending.
Growth of $1, close-to-close, since 1997. The two lines are held the day after their trigger and sit in cash otherwise, so they trade a similar number of days — the comparison is the dip-buy leg (green) against its opposite (red), not against buy & hold. No costs or slippage.
Forward return by RSI(2) at the close (toggle mean / median above). Low RSI(2) = short-term oversold. Bar outlined = where GPI is now.
Forward return by how far price sits from its 50-day average (in standard deviations). Outlined = now.
The next-day return after a run of consecutive up or down closes. A rising staircase toward the “down” end is mean reversion.
Mean-reversion edge by year — the dip-buy leg's next-day return minus its opposite (after a down day vs after an up day, or bottom- vs top-decile RSI(2)). Above zero = the name faded moves that year. Bars are the mean, which a crash can swing hard in either direction (2020); the tick marks are the median — the typical day.
Matched-exposure comparisons — each pair trades a similar number of days, so raw returns are comparable. RSI(2) deciles use thresholds ≤7.4 and ≥94.0.
RuleDays heldAvg / dayWin rate TotalCAGRSharpeMax DD
■ After a DOWN close 3,535 (49%) +0.032% 49.4% -42% -1.9% 0.11 -85.2%
■ After an UP close 3,627 (50%) +0.149% 50.6% +3,773% +13.5% 0.53 -69.7%
■ Bottom 10% RSI(2) 726 (10%) -0.010% 50.0% -46% -2.1% -0.01 -79.6%
■ Top 10% RSI(2) 726 (10%) +0.051% 49.9% -2% -0.1% 0.08 -60.7%
How to read this  ▾

Some assets mean-revert at short horizons (a sharp move tends to partly reverse); others trend (a move tends to continue). At the daily close-to-close horizon the reversal shows up cleanly and consistently in the broad equity indices and mostly breaks down elsewhere — commodities, crypto, and individual stocks carry their own momentum. The names above are the ones where it holds up.

RSI(2) is a 2-day momentum oscillator — below ~10 is short-term oversold, above ~90 overbought. Distance from the 50-day average is measured in standard deviations (z-score). Streak counts consecutive up or down closes. For each, we show what happened over the next 1–5 days, historically.

The strategy table pairs each dip-buy rule against its opposite — day-after-a-down-close vs day-after-an-up-close, and the bottom vs top RSI(2) decile. Both legs of a pair are held one day and trade a similar number of days, so raw returns compare fairly (unlike buy & hold, which wins on total return simply by always being invested). Avg / day — mean return on the days the rule is in the market — is the cleanest apples-to-apples number. Frictionless: no commissions or slippage. A character test, not a tradeable system. Past behaviour is not a forecast.

End-of-day adjusted closes, refreshed nightly. 7,260 trading days, 1997-10-30 – 2026-09-11. Backward-looking statistics, not investment advice.