We tested 8 popular trading rules against buy and hold, 19 times
None of them beat holding. All 19 tests ended with less money than simply buying the same fund on the first day and holding it, after costs. Before any costs, 19 of 19 still trailed. The rules gave up between 1.6 and 12.0 percentage points of return per year. What they did do: 19 of 19 had a smaller worst drop than holding, while sitting in cash for part of the time.
Historical simulation of the past. It does not predict the future and is not advice. Trading involves risk of loss.
Return given up per year
- QQQ - RSI(2) under 10:12.0 pts per year below holding
- QQQ - lower Bollinger band:10.6 pts per year below holding
- QQQ - RSI(14) under 30:9.4 pts per year below holding
- QQQ - MACD cross:8.1 pts per year below holding
- SPY - RSI(14) under 30:7.4 pts per year below holding
- SPY - 20-day average:6.6 pts per year below holding
- QQQ - 20-day average:6.6 pts per year below holding
- SPY - lower Bollinger band:6.2 pts per year below holding
- SPY - 50-day average:5.4 pts per year below holding
- SPY - MACD cross:5.4 pts per year below holding
- SPY - RSI(2) under 10:5.0 pts per year below holding
- IWM - RSI(2) under 10:4.5 pts per year below holding
- QQQ - 200-day average:4.2 pts per year below holding
- IWM - 200-day average:3.4 pts per year below holding
- DIA - golden cross*:2.7 pts per year below holding
- SPY - 200-day average:2.6 pts per year below holding
- SPY - golden cross*:2.0 pts per year below holding
- GLD - 200-day average:1.6 pts per year below holding
- QQQ - golden cross*:1.6 pts per year below holding
The results
Each row is one rule on one fund, from Nov 13, 2006 to Nov 13, 2026, starting with $100,000. "Rule" is the total return after costs; "Holding" is buying the fund on the first day and never selling. Open a fund to see its full test.
What the numbers say
- All 19 tests ended below holding after costs, and 19 of 19 were below it even before costs, so trading costs were not what decided the result.
- The rules were invested between 3% and 76% of the time. Holding is invested every day, so each rule had to make up the days it sat out by avoiding enough of the falls.
- 19 of 19 had a smaller worst drop from a high than holding. A smaller drop is easier to sit through, but here it came with a lower ending balance.
- 3 of 19 had a higher Sharpe ratio (return per unit of risk, over T-bills) than holding: SPY golden cross (0.57 vs 0.54), QQQ golden cross (0.75 vs 0.68) and SPY 200 day average (0.60 vs 0.54). The differences are small, and they still ended with less money.
- SPY golden cross, QQQ golden cross and DIA golden cross made fewer than 30 trades in 20 years (9, 11 and 9). That is too few to tell a rule from luck, in either direction.
How we tested
- The list of 19 tests was fixed before any of them was run, and every result is published as it came out. One planned test (TLT on its 200-day average) was dropped before publishing because our data provider is missing most of its 2007 to 2012 dividends, which would have understated both returns.
- Data: 5,032 daily sessions per fund from Twelve Data, Nov 13, 2006 to Nov 13, 2026, adjusted for splits, with dividends added back for both the rule and holding.
- Each rule is read on the day's close and trades at the next day's open, long only, with the whole account and no leverage. While out of the market, cash earned the 3-month Treasury bill rate (an average of 1.59% a year over the window).
- "After costs" means every entry and exit paid modelled slippage of 2 bp per side on SPY, QQQ, IWM and DIA and 5 bp per side on GLD, with no commission. Holding is one buy on the first day with no costs. Taxes are not modelled for either.
- At a harsher 10 bp of slippage per side, 19 of 19 still ended below holding. Each fund's page shows its own number.
- The same engine runs the free Reality Check, so you can test a rule of your own the same way.
Limits
- One window. A different start or end date can change every number here; these 20 years include a long rise in US stocks after 2009.
- Each rule was tested with its common textbook settings only. Other settings were not tried, on purpose, so the result was not picked after the fact.
- Fills are modelled at the next open plus slippage, not real orders.
- Our data provider has no QQQ dividend records for 2007 to 2011, so both QQQ returns are slightly understated.
- This describes the past. It does not predict the future and is not advice.