ConnorsRSI short on crypto: 7 in 10 trades won, and a loss

The answer first: a short strategy that wins seven trades in ten can still lose money, and this one did. It sells coins that have just run up hard and buys them back once they cool off — which usually works, by about 10% a trade. But a short has no floor under its loss: when the coin keeps rising, the loss grows until a stop or a liquidation ends it. Without a stop that happened six times in five years, and those six trades took more than half of everything the strategy lost. A wide stop-loss turned all six into ordinary losses and the book from a small loss into a small gain.

The rule comes from an algorithm published on QuantConnect¹ that shorts US stocks, built on the ConnorsRSI indicator of Larry Connors. Here it runs on crypto perpetuals, with the platform's own CRSI indicator.

The rule and the test

• Entry: at a daily close where ConnorsRSI(3, 2, 100) is above 90 and the volatility of the last 99 daily log returns is above 5.261% a day — about 100% a year on a 365-day year — place a sell limit 3% above that close, valid for the next day. It fills only if the price trades through the level, at the level (or at the open if the day opens above it).
• Exit: at the first daily close with ConnorsRSI below 30. No stop-loss, no take-profit, no time limit — the stops come later as variants.
• Size and margin: one short at a time per coin, 95% of the run's equity, 1x leverage, isolated margin.
• Costs: the limit entry pays the 0.02% maker fee with no slippage; the exit pays the 0.05% taker fee and 0.05% slippage; funding is charged and paid.
• Coins: the 60 Binance USDT-M perpetuals with the highest turnover in the 30 days before 2021-10-22, as the platform lists them for that date. 22 of the 60 have been delisted since and have no daily candles on the platform, which leaves 38 survivors — a survivors-only list; the 22 dead coins are checked separately below.
• Period: 2021-10-22 → 2026-10-22, daily candles, 10,000 USDT per run. Indicators warm up on the bars before the start. DYDX and GALA, listed in September 2021, begin in December 2021, once the 100-day rank inside ConnorsRSI has history; ICP's daily candles on the platform start in September 2022, so its run begins in January 2023. BTC, ETH and BNB never met both conditions and made no trades.
• Runs: strategy #156 without a stop (runs #2110–#2147), #160 with a 25% stop and #161 with a 40% stop. The stop runs were stored only for coins where the stop changed something — 24 and 16 coins; on the others the trades are the same as without it, which quota-free re-runs of all 38 coins confirmed.

The book. The original holds up to 40 shorts at 1/40 of the account each. A platform run trades one coin with 95% of its own account, so a liquidation shows up on its card as a loss of about 95%. To see the strategy the way the original sizes it, each of the 38 runs gets one of 40 equal slots: the book's daily return is the sum of the 38 runs' daily returns divided by 40, two slots stay in cash, and the slots are made equal again every day. An open short is then about 2.4% of the book (95% of a 1/40 slot). The book's max drawdown is measured on daily closes, and its Sharpe ratio is computed the way the platform computes it (daily returns, √365). A short script does this from the stored runs' daily equity, so every book number can be redone from the runs linked below.

How much was tried. Eleven versions of this rule have been run on the platform: five in an earlier port that had to approximate ConnorsRSI (strategy #100 — no stop, stops of 25%, 40% and 60%, and the original's stricter volatility threshold) and six here (no stop, 25% and 40% stops, each also with the stricter threshold). The 25% and 40% levels were fixed before these runs, but the earlier port had already tried them on the same coins, so they are not out of sample.

Seven in ten trades won, the book lost

Without a stop, 126 of the 181 trades won. The average winner made +10.15% on its position; the average loser lost -21.77%, and six of the losers were liquidated, each losing 99.82% of its margin. In money: with 10,000 USDT in each of the 38 runs, the winning trades made 114,136 USDT and the losing ones lost 123,936 USDT — and 63,732 USDT of that went in the six liquidations, one trade in thirty. Of the 286 sell limits the rule placed, 181 filled; the other 105 expired because the coin did not rise another 3% the next day.

Those six trades decide the result. The other 49 losing trades averaged -12.21%, small enough for the winners to carry: the book made +1.93% in what was left of 2021 — the first liquidation, 1000SHIB, included — and +7.95% in 2022, a falling year for most coins. The other five liquidations came from 2023 on, and from its peak in November 2022 the book lost 13.10% by November 2025.

Six shorts that doubled

A 1x short on isolated margin is liquidated when the price roughly doubles from the entry; on these six the liquidation price was 1.96 to 2.03 times the entry, funding moving it a little. Each was entered exactly as the rule intends — a coin already stretched to a ConnorsRSI above 90 that rose another 3% the next day — and kept rising. The exit waits for ConnorsRSI to drop below 30, and the margin ran out first: on 1000SHIB within a day, on CRV after 17. One more came within a percent of it: GRT, shorted on 2023-02-05, rose 99.10% above its entry before the rule bought it back at a loss of 26.11% — a near miss that the win rate does not show either.

Backtest

ConnorsRSI crash short: CRSI(3,2,100) > 90, no stop

HBARUSDT · 1d · Futures · 2021-10-22 → 2026-10-22

Return

-90.71%

Max drawdown

95.17%

Sharpe

-0.39

Win rate

80.0%

Trades

5

Open the report →

What the stop changed

A stop-loss 25% or 40% above the entry price turns each of the six liquidations into an ordinary loss. No run with a stop was liquidated, and every one of the 57 stop-outs filled at its stop plus the 0.05% slippage. These runs use the current engine, in which a stop that the price crosses on its way to the liquidation level fills first: 1000SHIB, liquidated without a stop on 2021-10-27, is stopped out the same day in the 25% run.

Backtest

ConnorsRSI crash short: CRSI(3,2,100) > 90, stop-loss 25%

HBARUSDT · 1d · Futures · 2021-10-22 → 2026-10-22

Return

+36.86%

Max drawdown

28.77%

Sharpe

0.46

Win rate

80.0%

Trades

5

Open the report →

The stop does not pick its trades, though. Matching each stop-out with the same trade in the run without a stop:

Most stop-outs cut a trade that would have come back: 22 of 38 at 25%, 10 of 19 at 40%. Measured trade by trade, the 25% stop saved 444.9 percentage points on the six liquidated trades and gave back 259.5 on the 30 other trades it closed. That is why the number of coins in profit went down with the stop, from 21 to 19 and 20: the 25% stop made the six liquidated coins much better and twelve others worse. DYDX, the best coin without a stop at +97.23%, made +23.88% with the 25% stop — one stop-out on 2026-07-01 cut a trade that would have ended at +22.29%.

Backtest

ConnorsRSI crash short: CRSI(3,2,100) > 90, stop-loss 25%

DYDXUSDT · 1d · Futures · 2021-10-22 → 2026-10-22

Return

+23.88%

Max drawdown

23.89%

Sharpe

0.32

Win rate

66.7%

Trades

12

Open the report →

The two levels ended almost level in the book, at +8.37% and +8.43%. The 25% stop had the smaller drawdown (6.21% against 8.09%) and the higher Sharpe ratio (0.57 against 0.50); with only six trades doing the work, that difference is not something to choose a level by.

Year by year

The book made its money in the falling market of 2022 and gave it back in 2023 and 2024, when the coins it shorted kept rising. The stops did not change that shape; they made the losing years shallower. Over five years the books with a stop made about 8% in all — with each short at only 2.4% of the account, neither the gains nor the losses could be large.

Survivors only? The 22 delisted coins

Of the 60 coins of October 2021, 22 have been delisted since — LUNA, FTM, MATIC, EOS, WAVES and 17 more. The platform keeps them on 1h and 4h candles only, so the main test above could not include them. For a short strategy leaving out the dead coins could flatter it or hurt it: a coin that collapses is a good short, unless it rallies first — all three shorts of LUNA in the check below lost money.

To include them, the rule was re-run on 4h candles reading its indicators from daily bars built out of them (the timeframe: "1d" option on each indicator). The limit then rests for six 4h bars from 04:00 instead of the calendar day, and the exit fills four hours later. As a check, the same proxy on the 38 survivors found six liquidations too (five on the same coins), 158 trades instead of 181 and 20 coins in profit instead of 21 — but single coins moved a lot (their returns correlate at 0.74 with the daily runs), so it is good for counts, not for coin-by-coin numbers. These are quota-free re-runs, not stored.

None of the delisted coins doubled while it was shorted, so there was no liquidation for a stop to prevent: with the 25% stop their average return per coin was -0.33%, with the 40% stop +1.69%. Adding the dead coins makes the result a little better, not worse, and changes nothing in the story — the six liquidations were all on coins that are still listed.

A stricter volatility filter, and the earlier port

The original annualizes volatility with √252, the trading days of a stock market. Taken literally on crypto, that raises the bar from 5.261% to 6.331% a day. With the stricter filter the rule trades less than half as often, and only on the most extreme coins — XRP, LTC, LINK, VET and AAVE drop out along with BTC, ETH and BNB. Without the daily book (these are quota-free re-runs, not stored), the fair comparison is equal money in each of the 38 runs:

The first three rows are the stored runs above. The stricter filter won four trades in five and still had three liquidations — on SOL, 1000SHIB and HBAR, three of the same six coins — which either stop removed again. The pattern holds; the level does not mean much: the stricter filter came out ahead on this history with or without a stop, which makes it one more variant that looks better in hindsight, not a reason to prefer it.

The earlier port of the same algorithm on the same 38 coins (strategy #100, runs #1522–#1559), which had to approximate ConnorsRSI with RSI and percent-rank conditions and the 3% offset with 0.375 ATR, came out the same way: 168 trades, 73.21% of them won, six liquidations, and a book of -1.84% with a 16.60% drawdown, by the same script.

Conclusion

• A win rate of 69.61% said almost nothing about this short strategy: the average winner made +10.15%, and six liquidations at -99.82% took more than half of all losses, leaving the book at -1.39%.
• A wide stop-loss — 25% or 40% above the entry — removed every liquidation and turned the book to +8.37% and +8.43%. It did that by catching six trades; most of its other stop-outs cut trades that would have recovered.
• For a short, the question to ask of a high win rate is what the worst trade was. A long can lose at most what it put in; a short's loss grows as long as the price rises, and only a stop or a liquidation ends it.

On a long trend rule, whose own exit already cut losers short, a stop mostly cost money: does a stop-loss help. Another oscillator rule that wins most of its trades: RSI 30/70 on crypto. How to read the cards: backtest metrics explained.

All 38 coins

"Same" means the stop never fired on that coin, so its trades are those of the run without a stop. "Trades won" counts trades that closed with a profit after fees and funding.

Notes

1. The algorithm "ConnorsCrash", published on QuantConnect as a backtest on US stocks (September–December 2024): up to 40 shorts at 1/40 each, ConnorsRSI(3, 2, 100) above 90, annualized volatility above 100%, a sell limit 3% above the price, out when ConnorsRSI falls below 30. Checked on 2026-10-30. Its volatility is annualized with √252 trading days, which on crypto is the stricter filter tested above, and its limit order carries no time limit; here the limit is valid for one day. ↩