Research · Macro
The series people overlay on bitcoin charts — the Fed funds rate, M2, net liquidity, credit spreads, the VIX — published raw, and then tested. For each series, the page tests whether recent changes were associated with bitcoin's return over the following quarter, at leads from zero to twenty-six weeks, with a bootstrap interval and a baseline from the same series shifted in time. Whether “liquidity leads bitcoin” is a claim with a lead and an interval, not a chart with two lines.
Reading the daily snapshot.
| Series | Change measured | Best lead | Correlation there | Bootstrap 90% interval | Shifted-series band | Verdict |
|---|
All series come from FRED as keyless CSV and are published raw in macro.json. Weekly alignment takes the last available value in each week ending Wednesday; monthly M2 steps once a month; net liquidity is assembled from its three legs (Fed assets and the Treasury account are published in millions of dollars, reverse repo in billions; all are converted to trillions). The macro “impulse” is the thirteen-week change (percentage change for M2 and net liquidity, change in points for rates and spreads, log level for the VIX). The bitcoin target is the forward thirteen-week log return of the daily-average price. Correlations are Spearman with midranks. The bootstrap resamples 26-week moving blocks of the paired weekly observations, 200 draws, and reports the 5th and 95th percentiles; the baseline band comes from 200 circular shifts of the macro series by 52 to 260 weeks, which keeps each series' own autocorrelation and removes any relationship to bitcoin. A verdict of evidence requires the observed correlation to sit outside the shifted band by a margin of 0.03 with the bootstrap interval excluding zero, at three consecutive leads; a reading within 0.03 of the band for three leads is reported as marginal rather than rounded up to evidence. Each series uses its own fixed random seed, so adding a series cannot change another's verdict.
Roughly 700 weekly observations contain three or four macro cycles, and the 2020–21 episode — liquidity and bitcoin rising together — is one episode however many weeks it spans; the block bootstrap treats it accordingly, which is why the intervals are wide. A correlation at a lead is not a mechanism, and a series that led in one cycle need not lead in the next. The page reports what the data support at each lead and how wide the uncertainty is; it does not turn that into a forecast, and neither should a reader.