Research · Base rates

What happened anyway

Every claim on the scorecard is measured against the days it could have fired but didn’t. This is what those days did. A signal that is followed by a doubling 53% of the time sounds impressive until you know that doubling within a year happened on 53% of all days. This page is the denominator.

Computed from the daily snapshot for 10 September 2026

On every day since 2013, a 40% fall arrived within the next year 34.6% of the time and a doubling 53.0%. The median one-year forward return was 71.9%, and it was positive on 69.6% of starting days.

5001 starting days from 2013-01-01 to 2026-09-10. Windows overlap, so the effective sample is four cycles, not four thousand days.

Forward returns from any day

Every day since 2013 as a starting point, per cent change over the horizon
Distribution of forward return, all starting days
HorizonWorst twentieth
5th percentile
/th>
Worst quarter
25th percentile
/th>
Middle
median
Best quarter
75th percentile
/th>
Best twentieth
95th percentile
/th>
Share positiveWindows
30 days-24.9%-8.5%2.7%17.5%56.4%56.4%4971
90 days-40.6%-15.0%7.6%44.0%173.6%57.6%4911
180 days-48.4%-18.9%27.2%75.2%376.9%62.4%4821
365 days-62.3%-17.3%71.9%170.5%771.3%69.6%4636

The distribution is skewed: at one year the p5 is -62.3% and the p95 is 771.3%. A median of 71.9% with a quarter of starting days losing 17.3% or more is what a volatile asset in a rising trend looks like, and it is the shape every claim on the scorecard is measured against.

The two outcomes the scorecard tests

Unconditionally, from every day
40% fall within a year34.6%4636 windows
Doubled within a year53.0%4636 windows

These are the two outcomes the scorecard tests every rule against: a top rule is judged on the first, a bottom rule on the second. A rule has to beat these on the days it could have fired, or it has told you nothing.

By starting cycle

The same outcomes, grouped by when the window opened
One-year forward return and outcome rates by cycle
Windows startingMedian 1y40% fallDoubledWindows
2013-16138.9%31.8%63.2%1461
2017-1947.6%48.5%55.4%1095
2020-2398.9%33.9%56.1%1461
2024- (open)15.7%18.1%17.3%619

By months since the last halving

The only cycle marker on this site whose dates were not chosen after seeing the data

Every other cycle marker on this site — Pi Cycle’s moving averages, MVRV’s 3.7, the Mayer multiple’s 2.4 — has a threshold someone chose after seeing bitcoin’s history. The halving dates are set by the protocol. This table conditions on them and nothing else.

81%6%
0–6 months698 days
58%12%
6–12 months732 days
24%71%
12–18 months689 days
1%100%
18–24 months549 days
27%37%
24–30 months549 days
88%7%
30–36 months546 days
72%24%
36–42 months549 days
84%18%
42–48 months324 days
doubled within a yearfell 40% within a year

Every day since 2013, grouped by how long after a halving it fell, and what followed within the next year. Four halvings, so each bar rests on at most one observation per cycle — the shape is suggestive, the sample is not.

Outcomes by position in the halving cycle, six-month bins
Months after halvingMedian 1y return40% fall within 1yDoubled within 1yWindows
00-06318.0%6.0%80.9%698
06-12114.6%12.4%57.9%732
12-18-44.9%71.3%23.8%689
18-24-47.4%100.0%1.1%549
24-3047.9%37.3%26.8%549
30-36128.3%7.1%88.1%546
36-42125.2%23.5%72.5%549
42-48223.9%17.6%83.6%324

Starting 00-06 months after a halving, a 40% fall within a year happened 6.0% of the time; starting 18-24 months after, 100.0%. That is the four-year cycle in one table, and it is four cycles. A bin of 549 windows is four independent observations. The pattern is unusually clean for a sample that small, and a clean pattern in four observations is exactly what this site declines to call a rule everywhere else.

By where the cycle was when you started

Starting MVRV, in quartiles of its own history
Outcomes by starting MVRV quartile
Starting MVRVRangeMedian 1y return40% fallDoubledWindows
Q1 lowestbelow 1.2984.9%22.0%60.8%1125
Q21.29 to 1.70117.7%41.0%60.6%1078
Q31.70 to 2.1877.4%36.5%47.7%1196
Q4 highestabove 2.1812.6%38.6%44.4%1237

Starting in the cheapest quartile of MVRV, the median one-year return was 84.9%; in the dearest, 12.6%. This is the realised-value thesis stated as a base rate rather than a signal: no threshold, no trigger, just the four quartiles and what followed each.

How to read this against the scorecard

The scorecard does not use these figures directly. It restricts the baseline to the days a rule could have fired but didn’t — the transition-matched set — so that a rule is compared with its own opportunity set, not with every day. That is why a rule’s “happened anyway” column differs from the unconditional rate here. The unconditional rate is the floor under both.

To test a rule of your own against these base rates, use the rule builder; to see what the site said on a given day and how it turned out, the forward record.

Windows overlap: a 365-day window starting on Monday shares 364 days with one starting on Tuesday. Thousands of windows are a handful of independent regimes. Every rate on this page is a fact about four cycles.

What this is not

Not a forecast. A forward-return distribution is a description of the past. The halving table in particular reads like a schedule and is four observations of a pattern that has no guarantee of persisting; the page says so beside it.

Recomputed daily by fetch/baserate.py and published as baserate.json. Method on the methods page.