The one-minute backtest trap: how a fee-only backtest invents profit
With only the exchange fee charged, our 1-minute backtests produced returns in the tens of thousands of percent a year. Adding one price tick of slippage per trade erased almost all of it. The profit had never existed.
How this was made: the code, the research runs and the first drafts of this text were produced by an AI assistant (Claude) working under the operator's direction; the operator made every decision and checked the numbers. All trading was on paper: real prices, no real money. Nothing here is investment advice.
What we saw
Charging Upbit's 0.05% fee per trade, the median Williams %R setting on 1-minute bars returned +53,042% a year. Charging the fee plus one tick per trade, the same settings lost everything. The gap shrinks as the bars get longer:
| Bars | Above water, fee only | Above water, fee and one tick | Median a year, fee only | Median a year, fee and one tick |
|---|---|---|---|---|
| 1-minute | 36.0% | 4.9% | −58.5% | −100.0% |
| 3-minute | 58.3% | 6.2% | +7.2% | −94.2% |
| 5-minute | 35.7% | 2.7% | −2.6% | −44.4% |
| 10-minute | 50.5% | 6.3% | 0.0% | −30.5% |
| 15-minute | 81.5% | 34.2% | +10.4% | −5.7% |
| 30-minute | 80.8% | 54.6% | +9.2% | +1.5% |
| 1-hour | 89.4% | 73.6% | +15.1% | +7.0% |
| 4-hour | 83.4% | 80.6% | +29.1% | +22.3% |
Why it happens
Every trade on an exchange happens at either the best bid or the best ask. A 1-minute bar's closing price is simply the last trade of that minute, so it lands on the bid or on the ask more or less at random. On a coin priced at 128 won with a 1-won tick, the two are 0.78% apart. Consecutive 1-minute returns in our data had an autocorrelation of −0.42: a close on the bid tends to be followed by a close on the ask, and the other way round.
A backtest that buys "at the close" after a down-tick and sells "at the next open" is, in effect, buying at the bid and selling at the ask. It earns the spread on paper, on every trade. Oversold rules such as Williams %R are perfectly tuned to find those down-ticks. In real trading a market order buys at the ask and sells at the bid, so the spread is a cost, not a gain.
How to spot the trap
- Fee-only results that look too good on very short bars, with tens of thousands of trades.
- Profits concentrated in cheap coins, where one tick is a large share of the price.
- Many trades that last a single bar.
- Results that collapse when you charge one tick per trade, or test on bid and ask prices instead of closes.
What was left
With one tick charged, 4.9% of 1-minute setups made money. The best that passed our two-period test made +50% a year, during a six-month window in which simply holding the same coins made +513% a year. None beat holding. Results by timeframe, and the tick size calculator for your own coins.