Investment approach
Swarm trades the most liquid crypto USDT perpetuals, long and short, following trends that persist for weeks to months. Every position is sized from volatility before entry and closed by rule; signals are never overridden by hand.
Trend over several horizons
Each asset is ranked on the strength and consistency of its trend over horizons from days to months, and capital goes only to the strongest few. A position is opened, reversed or closed only when its trend signal changes; in between, it is resized to hold its volatility target. Turnover in the backtest was about 45× equity a year, most of it from those resizes.
Volatility targeting
Positions are sized inversely to each asset's volatility, with the portfolio targeting roughly 20% annualised; exposure rises in quiet markets and falls in violent ones. Realised volatility over the backtest was 24.3%. Gross exposure averaged 0.30× equity and peaked at 1.02×, with 5× available on the venue. Note 04: gross exposure vs venue leverage.
Long and short
Shorts are taken on the same signals as longs, with a stricter condition for shorting while the broad market is rising. In 2022, a year in which the ten assets fell between 52% and 94%, the backtest returned 32.5% net, almost all of it from short positions. Note 05: long-only vs long/short.
Risk limits
- Every position has a stop order resting on the exchange at all times.
- No single asset above 25% of equity.
- A daily loss limit and a drawdown circuit breaker stop new risk being added.
- Funding and liquidity filters keep the book out of crowded or thin markets.
The limits block new risk only. Exit orders are never blocked.
Backtest
January 2021 to September 2026, ten Binance USDT perpetuals on 4-hour bars, net of 0.07% per unit traded and Binance's historical funding. Hypothetical: the rules were chosen with this data already available. Backtest of the rule as traded until 14 October 2026, when the resize rule was widened (journal); the change is within the figures' noise.
| 2021 | 2022 | 2023 | 2024 | 2025 | 2026* | |
|---|---|---|---|---|---|---|
| Net return | +101.5% | +32.5% | +53.9% | +39.6% | +6.1% | +22.0% |
* To 27 September 2026. 2021 was an unusually strong year for trend following. From 2022 to 2025 the backtest compounded at about 32% a year, and 2025 on its own returned 6%. By period, hypothetical: Sharpe ratio, from the tear sheet's monthly returns: about 2.1 in 2021–2023, about 0.8 in 2024–2026 and about 0.7 in 2025–2026, net of costs and funding; recent years have been weaker. No return is targeted or promised. The full-period CAGR of 41.8% overstates what recent years produced, and the backtest had drawdowns of 15–20%. Drawdown history · Full tear sheet
Execution
The system runs unattended, 24 hours a day. It places its own orders, keeps a stop resting on the venue for each position, reconciles its positions with the exchange every cycle, and records the slippage of each fill against the price at decision time.
The slippage record has already changed execution once. Passive (maker) orders saved fees on entry but cost 9–22 basis points on exit, so exits now cross the spread. Note 07