Does buying 2σ dips actually work?

One rule, five years of daily bars, nine tickers — and the honest answer, including the cases where it loses badly. Last run 11 Aug 2026.

z = (close − SMA20) / stdev20  ·  BUY when z < −2.0  ·  SELL when z ≥ 0.0  ·  long only, 100% notional, next-bar fill

No stops, no leverage, no position sizing, no parameter fitting — the same 20/−2/0 settings on every ticker so nothing is tuned per-symbol. Data is 5 years of daily closes (11 Aug 2021 → 11 Aug 2026).

Results

Ticker	CAGR	Sharpe	Max DD	Win %	Trades	Time in mkt	B&H CAGR	B&H Sharpe	B&H Max DD
XLE	25.4%	1.60	−11.7%	79%	24	18%	29.1%	0.98	−26.9%
SPY	10.0%	0.76	−9.2%	69%	26	21%	17.6%	0.88	−25.4%
IWM	9.6%	0.68	−11.8%	67%	24	23%	8.9%	0.45	−33.1%
XLU	6.7%	0.64	−15.3%	62%	26	25%	7.1%	0.43	−28.1%
GLD	3.8%	0.45	−8.0%	84%	19	18%	29.9%	1.28	−26.4%
QQQ	6.1%	0.42	−18.2%	75%	20	21%	21.8%	0.85	−35.6%
BTC-USD	2.1%	0.22	−44.9%	52%	31	26%	7.0%	0.39	−76.6%
TLT	−2.2%	−0.12	−24.7%	59%	27	35%	−15.5%	−0.80	−46.8%
ETH-USD	−9.9%	−0.05	−61.2%	47%	32	28%	−9.9%	0.20	−79.4%

Green Sharpe = the dip rule beat buy & hold on a risk-adjusted basis. Red = it lost.

What the numbers actually say

It is a drawdown machine, not an alpha machine. On every single ticker the max drawdown is smaller than buy & hold — often by half (SPY −9.2% vs −25.4%, BTC −44.9% vs −76.6%). That is the one robust result here.
It beats buy & hold on Sharpe in 4 of 9 (XLE, IWM, XLU, TLT) — the mean-reverting, range-bound stuff. It gets crushed on the trending stuff.
It loses on raw CAGR almost everywhere, because it is only invested ~20% of the time. If you compare it to buy & hold on total return you are comparing a 20%-exposure strategy to a 100%-exposure one, which is not a fair fight in either direction.
GLD is the cautionary tale: an 84% win rate and a 3.8% CAGR against buy & hold's 29.9%. High win rate tells you nothing on its own.
Crypto is where dip-buying goes to die. ETH is negative on a rule that is negative anyway; BTC keeps a third of the return for a bit more than half the drawdown. Neither is a strategy, both are a risk filter.
These are gross of costs. At ~25 round trips over five years, commissions are noise but slippage on the open is not — assume a few bps a trade.

Run it on your own ticker

The backtester that produced this table is free and takes no signup. Type a symbol, pick your lookback and z-thresholds, get the equity curve, the trade list and the current signal.

Mean-Reversion Backtester → Any crypto, stock or ETF · up to 5y daily · Sharpe / CAGR / max DD / win rate vs buy & hold · free, no signup

If you want the version I would actually trade off

The single-setting table above deliberately does not fit parameters, because a fitted number on one sample is worth nothing. The useful version is a parameter sweep with the out-of-sample check attached, so you can see whether the best setting is a real ridge or a lucky spike.

The Full Optimization Report ($29, one ticker) is that: a sweep over lookback × entry z × exit z, walk-forward in-sample vs out-of-sample results for the best settings, the complete trade ledger as CSV, and a standalone runnable Python file so you can re-run and modify it yourself. Delivered by email within 24 hours — if the ticker's best parameters do not hold up out of sample, the report says so.

Get the $29 report →

Secure Stripe checkout. Put the ticker you want in the checkout notes, or reply to the receipt.