Bitcoin Indicator Scorecard
Take a signal people quote — Pi Cycle, MVRV, RSI, the golden cross. Count how often the thing it predicted actually happened. Then count how often that same thing happened anyway, on the days the signal could have fired but didn't. If the two numbers are the same, the signal told you nothing.
The scorecard
11 signals examined · 2 fired often enough to judge · 0 beat doing nothing
Most of these signals describe cycle turns, and bitcoin has had four cycles. A signal that has fired five times cannot be told apart from luck, however clean its record looks — so those rows carry figures but no verdict.
| Indicator | What it claims | Times fired | It happened | Happened anyway | Edge | Verdict |
|---|---|---|---|---|---|---|
| Enough history to judge | ||||||
| RSI-14 below 30 | An oversold RSI precedes a rally. | 21 | 47.6% | 59.9% | -12.3 pts | Worse than the baseline |
| RSI-14 above 70 | An overbought RSI precedes a fall. | 20 | 30.0% | 38.6% | -8.6 pts | Worse than the baseline |
| Too few times to judge — figures shown, no verdict | ||||||
| LPPLS bubble model | A log-periodic power-law fit identifies a bubble approaching its critical time. | 18 | 28.0% | 34.4% | -6.5 pts | No verdict |
| Golden cross | The 50-day crossing above the 200-day average starts a bull phase. | 12 | 66.7% | 57.0% | +9.6 pts | No verdict |
| Death cross | The 50-day crossing below the 200-day average starts a bear phase. | 12 | 58.3% | 28.5% | +29.9 pts | No verdict |
| Mayer multiple above 2.4 | Price at 2.4x its 200-day average is an overheated market. | 8 | 25.0% | 26.3% | -1.3 pts | No verdict |
| Pi Cycle Top | The 111-day average crossing above twice the 350-day average marks a cycle top. | 5 | 100.0% | 25.9% | +74.1 pts | No verdict |
| MVRV above 3.7 | An MVRV above 3.7 marks the top of the cycle. | 5 | 20.0% | 28.0% | -8.0 pts | No verdict |
| MVRV below 1 | An MVRV below 1 means the average holder is under water: a bottom. | 5 | 60.0% | 54.9% | +5.1 pts | No verdict |
| Price below the 200-week average | Price under its 200-week average is a generational buying zone. | 4 | 75.0% | 46.1% | +28.9 pts | No verdict |
| Price below the 2-year average | The 2-year moving average multiplier marks the accumulation band. | 4 | 25.0% | 44.9% | -19.9 pts | No verdict |
Computed from the daily public snapshot through 2026-09-05, over 5,863 daily closes. Outcome window 365 days; triggers within 90 days are one episode; a verdict needs at least 20 episodes.
- Times fired — separate occasions, not days. A signal true for six months counts once, not 180 times.
- It happened — how often the predicted outcome followed within a year: a 40% fall for a top signal, a doubling for a bottom signal.
- Happened anyway — how often that same outcome followed on comparable days when the signal did not fire.
- Edge — the gap between the two. Near zero means the signal added nothing.
Every claim, one page each
Each of these is a question people actually ask. Each page gives the claim, where it came from, exactly how it was tested, and what the numbers do and do not support.
How a rule is scored
Each rule is measured only on days it could have fired — days where its inputs existed and where a full outcome window follows. A 200-week average cannot be scored before there are 200 weeks of price.
Consecutive triggers are grouped into episodes: a trigger, plus any further trigger within 90 days, counts once. Without this, a rule that stays true for six months would be counted as 180 independent successes.
The baseline is the eligible days on which the rule did not fire, excluding each trigger and the 90 days after it. That exclusion matters: the days immediately after a top signal are unusual precisely because a top signal just fired, and leaving them in would flatter the comparison.
Followed means a 40% fall from that day's close within 365 days for a top rule, or a doubling within 365 days for a bottom rule. Episodes whose 365-day window has not completed are excluded and counted as censored.
A rule with fewer than 20 scoreable episodes gets no verdict. Its figures are still shown, greyed, so you can see why.
What this cannot tell you
This is descriptive and in sample. It measures what followed these signals in the history we have. It is not a forecast, and a rule that beat its baseline historically may simply have been lucky across four cycles.
The outcome windows overlap heavily, so episodes are not independent draws. Treat differences of a few points as noise.
Most of these rules describe cycle-scale events, and bitcoin has had four cycles. That is why so many rows read not enough episodes. That result is the finding, not a gap in the work: a signal that has fired five times cannot be distinguished from luck, however impressive its record looks.
Every figure here is recomputed from the public daily snapshot. The code is in the data repository; the specifications are on the methods page.