Two rules-based models trading crypto perpetual futures. No discretion, no forecasts — five years of evidence and a live forward test.
Balanced setting · July 2021 – September 2026 · simulated on real Binance futures data, fees included · compounded
Performance
| a year | worst drawdown | Sortino | Calmar | |
|---|---|---|---|---|
| Strategy · balanced | +76.5% | −11.8% | 7.99 | 6.50 |
| BTC | +16.9% | −76.7% | 0.75 | 0.22 |
| Mag 7 stocks | +26.4% | −50.8% | 1.42 | 0.52 |
Account value on a log scale. Sortino: monthly returns above a 4% risk-free rate, against the losing months only. Calmar: the annual return divided by the worst drawdown over the whole period. Benchmarks are buy-and-hold over the same window; the Mag 7 stocks (Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, Tesla) are an equal-weight basket bought once, dividends reinvested. 2021 starts in July and 2026 ends in September.
Bad markets
BTC fell 77% and 53%. Over the same dates the strategy made +72% and +139%.
| Every BTC fall of 20% or more | BTC | strategy |
|---|---|---|
| Sep 2021 → Sep 2021 | -23% | -3% |
| Nov 2021 → Nov 2022 | -77% | +72% |
| Mar 2024 → Sep 2024 | -26% | +19% |
| Jan 2025 → Apr 2025 | -28% | +7% |
| Oct 2025 → Jun 2026 | -53% | +139% |
A flush day is any day BTC traded 8% or more below the previous close; 7 of them fell 15% or more, including September 2021, December 2021, June 2022, November 2022 (FTX), August 2024 and 10 October 2025. On 2 of them the strategy lost a single full stop, and on one-minute data those stops filled at their level. Most of the time it was simply not in a trade. Balanced setting, compounded; BTC on daily closes.
Read these numbers carefully
Ratios this high are unusual, and they deserve suspicion. Most of it comes from a few days: take out the ten best days of the five years and the strategy makes +34% a year; take out twenty and it makes +16%.
| Balanced setting | a year | worst drawdown | Sortino | Calmar |
|---|---|---|---|---|
| As published | +76.5% | −11.8% | 8.0 | 6.50 |
| Without its 10 best days | +33.9% | −13.5% | 3.5 | 2.52 |
| Without its 20 best days | +15.8% | −16.2% | 1.4 | 0.98 |
The best 10 of 265 trading days made about 49% of all the growth — spread across every year, not one crash. Long flat stretches between them are normal.
Tight stops mean large positions for the account size. The full book stops scaling at an account of roughly $1.5M–$2.0M, because its smallest market is thin; beyond that, orders would have to be split and give up edge.
The numbers →Drawdowns are measured on closed trades. A position that went against us intraday and recovered is not in them, so the real figure is never smaller — and Calmar never larger.
How it was tested
Every model is re-run seeing only what was knowable at the time. One that failed was deleted.
The check →The trading program is replayed against the tested version over five years, trade for trade.
The check →Rules are checked on years they were never built on, and sized on an edge smaller than measured.
The check →Fees come out of every trade before anything is measured, at a rate worse than the exchange charges.
The check →Live forward test
Real signals on a simulated prop-firm account. Too few trades to judge yet — it is here so it can be checked.
Prop-firm accounts
1.25% a Model A trade · 0.41% a Model B trade
1.00% a Model A trade · 0.35% a Model B trade
20,000 reshuffled paths of the five-year record. Funded: typical payout to the trader per $100,000, profit above the start paid out monthly at an 80% split; a funded account has no target. Challenge: the 10% target. Account lost = the 6% drawdown or 3% daily limit is hit.
FAQ
The performance figures are a five-year simulation on real exchange data, with fees taken out of every trade. The forward test has been running since 13 August 2026 on a simulated prop-firm account, and is reported separately on the Live page, never mixed into the simulation.
Crypto perpetual futures: two independent rules-based models across three markets. About 57 trades a year, often none for weeks.
Every model is re-run so it can only see what was knowable at the time; the live code is replayed against the tested code; results are split into the years a rule was built on and years it never saw; and position sizes are set on an edge deliberately smaller than measured. The How it was tested page shows the work, including what failed.
At the balanced setting the deepest fall was 11.8% and the worst single day -4.82%. Losses are capped per trade by a stop held at the exchange, so the account never depends on our software being online.
They are the two obvious alternatives: holding the crypto market, or holding the most volatile corner of the stock market. Over the same window BTC returned +17% a year with a 77% worst drawdown, and the Mag 7 stocks +26% with a 51% worst drawdown. The strategy returned +77% with an 11.8% worst drawdown.
Two ways of measuring return against risk. Sortino takes the average monthly return above a 4% risk-free rate and divides it by the size of the losing months only, so a big winning month does not count against the strategy the way it does in the more common Sharpe ratio. Calmar divides the annual return by the worst drawdown: how many times over a year pays for the deepest fall. Above 1 is good for either; above 3 is rare — which is why the same figures are also shown without the best days.
Less than you might think. At the balanced size the full book stops scaling at an account of roughly $1.5M–$2.0M (the account, not the position: tight stops make each position larger than the account), because its smallest market is thin; without that market, roughly $5.9M–$17.6M; BTC alone, $25.6M–$76.7M. Beyond that, orders would have to be split over time, which costs exactly the precise entries these setups rely on.
Not publicly. The strategies and the software that trades them are run privately and are not for sale, but I am open to a conversation about access, licensing or working together — reach out on X (@RayCapitol) or Discord (@r_capital).
Yes, at smaller sizes. At the challenge sizing, 89% of simulated challenges reached the profit target, typically in 96 days. Once funded, at the smaller funded sizing, 2.4% of simulated accounts were lost within a year, and the typical payout was $15,325 a year per $100,000.