Is there a common force behind stablecoin supply changes? Principal components of daily market capitalization movements, tested against sampling noise.
As of September 8, 2026, one component exceeds the noise ceiling, by 1.1 percentage points of total variance, which is within one band-width of the edge. It currently loads most heavily on PYUSD (27% of the component).
Principal component analysis of daily market cap percent changes across the top 10 tracked stablecoins. Every component is tested against the sampling-noise band for 10 uncorrelated series over 503 days, 7.4% to 13.0% of total variance; components inside that band are consistent with pure noise.
One component exceeds the noise ceiling, by 1.1 percentage points of total variance — within one band-width of the edge. A reading this close to the threshold can be produced by an episode of unusual volatility in a subset of coins; persistence as the series lengthens is the test. It currently loads most heavily on PYUSD (27% of the component).
The spectrum and the band are recomputed daily as the series lengthens. This reading is consistent with the near-zero average pair correlation currently shown on the correlation page.
| Component | Explained variance | Against the band |
|---|---|---|
| PC1 | 14.1% | above |
| PC2 | 12.1% | inside |
| PC3 | 10.9% | inside |
| PC4 | 10.7% | inside |
| PC5 | 10.0% | inside |
| PC6 | 9.5% | inside |
| PC7 | 9.1% | inside |
| PC8 | 8.4% | inside |
| PC9 | 8.3% | inside |
| PC10 | 7.0% | inside |
| Coin | PC1 loading | Share of PC1 | PC2 loading |
|---|---|---|---|
| PYUSD | +0.52 | 27% | +0.34 |
| USDT | +0.43 | 18% | -0.38 |
| USDC | +0.42 | 18% | -0.10 |
| RLUSD | +0.35 | 12% | -0.33 |
| USDG | +0.34 | 11% | +0.27 |
| USDD | +0.25 | 6% | +0.11 |
| USDS | +0.23 | 5% | -0.25 |
| USDE | +0.11 | 1% | +0.55 |
| USD1 | -0.06 | 0% | -0.32 |
| DAI | -0.01 | 0% | -0.25 |
A market driven by one shared force concentrates variance in its first component; equity markets typically show a first component several times the noise threshold. Components inside the shaded band are statistically indistinguishable from what 10 independent series would produce by chance, so only components above it carry evidence of common structure. The band narrows as the observation window grows.
One component exceeds the noise ceiling, by 1.1 percentage points of total variance — within one band-width of the edge. A reading this close to the threshold can be produced by an episode of unusual volatility in a subset of coins; persistence as the series lengthens is the test. It currently loads most heavily on PYUSD (27% of the component).
Standardization. Each coin’s return series is centered and scaled to unit variance before the decomposition, so the components describe co-movement rather than volume of movement. Without the scaling the first component goes to whichever coin has the largest percentage variance, regardless of whether anything else moves with it.
Matrix and coverage. Daily market capitalization percent changes for the top 10 coins by market capitalization. A day enters only when every coin has that day and the prior day, giving 503 complete observations; a coin with no variance over the window is dropped rather than divided by. Levels are not used, because they trend and a decomposition of trending series produces structure that is not there.
The noise band. The band is the Marchenko–Pastur bound for a correlation matrix of 10 uncorrelated series over 503 observations. Components within it are consistent with sampling noise; the band is recomputed daily from the actual matrix dimensions.
Updated daily at 15:40 UTC. See the methodology for data sources and coverage.
Each coin’s daily market cap percent changes are centered and scaled to unit variance, and the eigendecomposition runs on the resulting correlation matrix, so a coin contributes according to how it moves rather than how volatile it is. A day enters only when every coin in the set has that day and the prior day, and a coin with no variance over the window is dropped rather than divided by. The figures on this page are computed when the page is built, not in the browser.
On current data, essentially no, pending persistence. The one component above the ceiling clears it by 1.1 percentage points, less than the width of the noise band itself, so it is the kind of reading a volatility episode in a few coins can produce.
Because the absence is the finding. Whether stablecoins behave as one market or as many independent instruments matters for how concentration, contagion and substitution should be read, and this page measures which is currently true.
The spectrum is recomputed daily and the noise band narrows as the series lengthens. A regime shift, toward or away from common movement, appears as components crossing the band boundary and staying across it; a single volatile episode appears as a crossing that fades once its window passes.