Research ·
Does shorting help a crypto trend bot? Long-only vs both sides
Long-only made 35.5% a year with a −14.7% worst drop. Long and short made 45.7% with −18.5%. Almost all of the difference came from one year: 2022, when long-only lost 1.8% and the shorts made 36.7%.
Most crypto bots only buy. Ours can bet on prices falling too, and it does: about half of all its position-time in the backtest was short. Is that worth it? We ran the same rules three ways to find out.
The test
Same coins, same 16 votes, same four seats, same 25% stop-loss and profit lock, same 0.07% cost per trade, January 2021 to September 2026. The only change: in the long-only run every short signal was ignored; in the short-only run every long signal was.
| Jan 2021 – Sep 2026 | Long only | Short only | Both (live) |
|---|---|---|---|
| Return per year | 35.5% | 5.9% | 45.7% |
| Sharpe ratio | 1.98 | 0.41 | 1.67 |
| Worst drop | −14.7% | −18.7% | −18.5% |
* 2026 runs to 27 September. Hypothetical backtest, funding not included.
What it shows
- Long-only is the smoother ride. A better Sharpe ratio (1.98 against 1.67) and a shallower worst drop (−14.7% against −18.5%). Our guess at why: crypto's rising trends have tended to run longer than its falling ones, and shorts get caught in sharp rebounds.
- Shorting pays in one kind of year. In 2022, when our ten coins fell between 52% (BNB) and 94% (SOL), long-only lost 1.8% and short-only made 36.6%. The combined strategy made 36.3%, nearly all of it from shorts. In the other years the short positions inside the combined strategy added 2–3 points (2023, 2025), cost about 3 (2024), or helped more (about 9 points so far in 2026).
- Short-only doesn't work on its own. 5.9% a year with an 18.7% worst drop. Shorts are an insurance policy, not a strategy.
Why the bot keeps both sides
On the numbers alone there's a real case for long-only: better risk-adjusted return, smaller drops, simpler to explain. We kept both sides for one reason: a bear market is exactly when a long-only crypto strategy leaves you with nothing to do for a year or more, and exactly when this one made money. Six years contain only one proper bear market, so the evidence for that insurance rests on one year, and we'd rather say so than hide it.
The bot doesn't treat the two sides equally anyway. When Bitcoin is above its 50-day average, every short signal counts half, so it needs twice the conviction to short into a rising market. It's meant to keep shorts rare and small while the market is rising.
If you copy
Expect to see short positions, sometimes while coins are rising, and expect some of them to lose. Over six years they made the difference between a flat year and a good one exactly once. If you'd rather have the long-only profile, this bot isn't it, and that's a fair choice.
More on how the bot's exposure splits between long and short: leverage and exposure.