Provyan Research · figures as of the 15 Nov 2026 close
We tested 4,132 crypto signal setups on 9 years of data. 95% show no reliable edge.
44 signals and price conditions, run on 94 coins, every occurrence since 2017 counted and compared with what the same coin does on any ordinary day. After correcting for the number of tests, 202 of the 4,132 setups have evidence behind them. The golden cross is not one of them.
What the count says
- Most setups carry no information. 1,794 of 4,132 are indistinguishable from picking days at random, 1,569 more are too weak to rely on, and 567 have happened fewer than 10 times. That leaves 202.
- What survives runs against the move that came before. Of the 202 setups with evidence, 102 lean down and 100 lean up. After rallies the evidence points down in 70 of 90 setups, and after drops it points up in 27 of 32.
- The famous ones do poorly. Golden cross: judged on 9 coins, evidence on 0. Death cross: judged on 8 coins, evidence on 0. RSI drops below 30: judged on 92 coins, evidence on 4. RSI rises above 70: judged on 93 coins, evidence on 10.
- Fibonacci levels are not special. Retracements end near them no more often than near any other level: the test.
- On data it had never seen, the rule was right 60% of the time. With every study rebuilt on candles up to 15 Nov 2025 and the calls of the following 12 months scored, 634 calls averaged +1.5% after costs. Down calls were right 76% of the time and up calls 45%, in a falling year. Replay and live ledger.
- A level most calls reach is not a profit. A first level fixed in advance was reached by 81% and then 83% of calls, and leaving there still averaged +0.2% and +0.2% per call after costs. The check.
By family of signal
| Family | Setups | Judged | With evidence | Lean |
|---|---|---|---|---|
| Rallies | 940 | 840 | 90 (11%) | 20 up · 70 down |
| Volume | 188 | 181 | 15 (8%) | 12 up · 3 down |
| Momentum | 188 | 185 | 14 (8%) | 12 up · 2 down |
| Price shock | 376 | 354 | 24 (7%) | 10 up · 14 down |
| Trend | 376 | 185 | 10 (5%) | 9 up · 1 down |
| Drops | 1034 | 882 | 32 (4%) | 27 up · 5 down |
| Bitcoin reactions | 372 | 292 | 10 (3%) | 5 up · 5 down |
| Streak | 188 | 184 | 5 (3%) | 5 up |
| Whale footprint | 188 | 187 | 1 (1%) | 1 down |
| Order flow | 188 | 188 | 1 (1%) | 1 down |
| Volatility | 94 | 87 | 0 (0%) | – |
All 44 signals ranked one by one
How it was done
- Data. Daily spot candles for 94 USDT pairs on Binance, from each pair's first trading day.
- Exact rules. Each signal has a written rule, confirmed at the daily close, and cannot fire twice on a coin within 7 days.
- Compared with the coin itself. The 7 days after a signal are measured against the same coin's 7-day record from every other day, so a coin that mostly rose does not flatter its signals.
- Corrected for luck. A test of 10,000 random draws per study, then a Benjamini-Hochberg correction across all 3,565 studies with enough cases.
- Checked over time. A setup whose direction flips between the earlier and later part of its history loses its score.
The full method, with its limits, is on the methodology page. The main limits: the coins are today's survivors, one exchange, daily candles, and a history that is not a live track record. The live record is the sealed ledger, which is still young.
Use and citation
Free to quote, chart and republish with a link to this page. Suggested credit: Provyan, “We tested 4,132 crypto signal setups on 9 years of data: 95% show no reliable edge”, https://provyan.com/research/crypto-signals-tested/. Figures update after every daily close; the date at the top of the page says which close they describe.
Download every study (JSON) All open data Questions: hello@provyan.com
Statistics about the past, not financial advice.