Your strategy sells just before the rally and buys just before the crash. After the tenth time, the thought is close at hand: then I will just do the opposite. Buy when it wants to sell, sell when it wants to buy.
The thought is not stupid. A signal that is reliably wrong would be worth exactly as much as one that is reliably right. You would only have to flip it.
Tested on the three weakest daily strategies on the platform, across ten large coins since 2017, the flipped version beats simply buying and holding in 5 of 30 cases. For the weakest of the three, in none.
The reason is a mix-up. For a strategy, "bad" almost never means "reliably wrong". It usually means "arbitrary, and trading costs money".
Three strategies the platform itself gave up on
Looking for the worst strategy means saying how "worst" is measured. So three candidates run here, each for a different reason:
- Stoch-RSI/SMA. Buys when the stochastic RSI crosses its own three-day average from below and sells on the cross from above. The platform switched it off on 23 June 2026 as its weakest active strategy: after fees, it beat buy and hold in none of 28 tested combinations.
- Bollinger squeeze. Buys when the Bollinger bands narrow and price then breaks above the upper band, sells below the middle line. Switched off the same day as the weakest strategy on the platform.
- RSI(2) mean reversion. Buys short dips inside an uptrend and sells on the bounce. Among the strategies still running and tested on many coins, it is the weakest against buy and hold on daily candles (825 runs on 136 coins).
Flipping here means: buy on the sell signal, sell on the buy signal, cash in between. That is how someone with an ordinary spot account would do it, without short selling and without leverage.
The rules were fixed before the first run started.
Three strategies, three kinds of bad. If flipping helps anywhere, it should help here.
The flipped version almost always loses to simply holding
| Strategy | Average trades | Flipped beats holding | Gap to holding, median |
|---|---|---|---|
| Stoch-RSI/SMA | 601 | 0 of 10 | โ28.6 percentage points a year |
| Bollinger squeeze | 16 | 4 of 10 | โ8.7 |
| RSI(2) | 59 | 1 of 10 | โ8.0 |
BTC, ETH, BNB, XRP, ADA, DOGE, TRX, LINK, LTC, XLM against USDT, Binance daily candles from the first candle to 18 Oct 2026, 0.1 % fee per buy and per sell. Gap = annual return (CAGR) of the flipped version minus buy and hold over the same window.
25 of 30 flipped versions trail simply holding, often by double-digit percentage points a year.
Against the original it looks kinder: in 19 of 30 cases the flipped version beats the strategy it came from. That is no credit to flipping. Bollinger squeeze and RSI(2) are rarely in the market, RSI(2) on bitcoin on 8.7 percent of days. Their flipped versions are therefore invested almost all the time. They are buy and hold with gaps, and every gap costs fees.
Flipping beats the bad strategy. Buy and hold beats both.
Why does the opposite of a bad signal not give a good one?
The original and its flip share the market
Long-only, a strategy and its flip complement each other. When one is in the market, the other is out, and the other way round. Together they are always invested, like someone who simply holds.
That sets a hard condition. The flipped version can only beat buy and hold if the original lost money during the time it was invested. Trailing buy and hold is not enough. A strategy that earns 50 percent while in the market, while holding earns 300 percent, is bad, but its flip only gets the other 250 percent, minus fees.
This condition was stated in advance, and the data bear it out. In only 6 of 30 cases did the original lose money in its market time before fees. In 4 of them, the flipped version beats buy and hold after fees as well. In the other 24, the original earns while invested, and the flip falls short in 23 of them. The one exception, the squeeze on ADA, arises only because the flipped version was still in cash when ADA crashed in 2018.
A bad strategy usually still makes money, just less than holding. Its opposite then gets the rest, not more.
The weakest of the three shows it most clearly.
With Stoch-RSI, the signal was not wrong. Trading was expensive
| Bitcoin | before fees | after fees |
|---|---|---|
| Stoch-RSI/SMA, per year | 19.4 % | 3.5 % |
| Flipped, per year | 16.6 % | 1.1 % |
| Buy and hold, per year | 40.7 % | 40.7 % |
BTC/USDT from 17 Sep 2017, 649 round trips, strategy in the market on 51 percent of days.
In its market time, Stoch-RSI on bitcoin earned 401 percent before fees. The flipped version was in profit during its half too. The strategy splits the market into two halves that both went up. It does not hit a direction, not even the wrong one.
What breaks it is 649 round trips. At 0.1 percent per buy and per sell, they cost about 16 percentage points of return a year. The flipped version trades just as often and pays the same bill.
On LINK and LTC, Stoch-RSI really did lose in its market time. Before fees, the flipped version beats buy and hold there by 10 and 9 percentage points a year. After fees, it trails by 13 and 3.
Flip noise and you get noise, with the same fees.
That leaves the cases where flipping worked.
Yesterday's winners do not last
With the Bollinger squeeze on ETH, XRP and LTC, the original lost money while invested, and the flipped version beats buy and hold, on XRP by 10 percentage points a year. Here was what the starting thought is looking for: a signal that is reliably wrong.
But that is measured in hindsight, over the whole period. Split it, and the picture changes. Up to the end of 2021, the flipped squeeze beats holding on 7 of 10 coins, from 2022 on only 4 of 10. Across all three strategies, the flipped version wins in 12 cases in the first half, and of those 12, 5 still win in the second.
That is barely more than a coin toss. Anyone who had flipped, in 2021, the strategy that had been reliably wrong until then would have won in fewer than half of the cases afterwards.
Something else changes hands too. The original often sits in cash and loses little: RSI(2) on bitcoin fell at most 34 percent. The flipped version is invested almost all the time and fell 76 percent, as deep as holding itself. You swap the original's shallow path for the deep one of holding, without getting holding's return.
A signal that is reliably wrong is as rare as one that is reliably right, and it lasts just as poorly.
Two numbers before you flip anything
- What did the original earn while it was invested, before fees? If the number is positive, the flipped version cannot beat buy and hold. The strategy is then not wrong, it is in the market too little or costs too much.
- How often does it trade? Each round trip costs 0.2 percent at 0.1 percent per side. At 70 round trips a year, that is 14 percent, and the flipped version pays it too.
Both numbers are in every backtest on the platform: time in market, number of trades, and return before and after fees. If the first is positive or the second is high, the flip is done before it starts. What remains is the choice between a better strategy and simply holding.
What stays open is whether a strategy that loses during its market time across several market phases keeps doing so afterwards. In this data, too few cases met that bar to decide it.
Method
Binance spot daily candles (UTC) against USDT, from the first candle to the close of 18 Oct 2026. Signal on the daily close, execution at the next day's open, 0.1 percent fee per side, long-only, no leverage. Platform default parameters, no optimisation. Buy and hold over the same window, after the indicators' warm-up. The flipped version starts in cash and buys on the original's first sell signal. A variant invested from the start gives 4 instead of 5 of 30. Platform cross-check: RSI(2) on bitcoin matches the backtester exactly before fees (70 round trips, 7.1 percent a year; after fees 5.4 instead of 5.2 percent). In the platform validation, the flipped version reaches 17.8 percent after fees against 25.6 percent for holding.
FAQ
Why not go short when the original buys? That would be the pure signal reversal. It needs assumptions on leverage, funding costs and liquidation that are not measured here. For a short, the condition has the same shape: it only earns if the original loses during its market time.
Why these ten coins? Fixed in advance: large coins already traded on Binance before 2020, so that every run spans several market phases.
Two of the strategies are switched off. How were they computed? Rebuilt from the published rules, with the same fees and the same execution. For Stoch-RSI on bitcoin, the rebuild reaches 3.5 percent a year after fees and 649 round trips, the stored platform runs over a slightly shorter window 3.9 percent and 664 trades.
Not investment advice, not a recommendation, not a forecast โ historical patterns are no promise.
Study the Past โ Improve your Future ๐ฅ