How is the number of clusters (k) selected?

k is selected using the elbow method: we compute the within-cluster sum of squares (WCSS) for k=2 through 6 and find the point where adding another cluster yields diminishing returns (largest second-derivative of the WCSS curve). The elbow chart is displayed on the page so the selection is fully auditable.

What is the "Institutional / High-Velocity" cluster?

Coins in this cluster have above-average velocity (volume/market cap ratio), meaning they circulate actively in payment and settlement flows. These tend to be large, liquid stablecoins used by exchanges and institutional counterparties.

What does it mean when a stablecoin migrates between clusters?

Cluster migration signals a genuine behavioral shift. A coin moving from "Growth Phase" to "Institutional" has matured — its growth rate normalized and velocity stabilized. A coin migrating toward "Declining / Contracting" is losing market share consistently across multiple windows, which may precede a deeper structural decline.

Indicator

Behavioral Clustering: What Type Is Each Stablecoin?

K-means groups stablecoins by how they actually behave — not by how large they are. The four behavioral features are velocity (how actively a coin circulates), 30-day growth rate (gaining or losing market share), market cap volatility (supply stability), and market beta (sensitivity to market-wide flows). All features are z-score normalised before clustering, so a $5B coin with the same behavioral profile as USDC will appear in the same cluster. The optimal number of clusters is selected using the elbow method on the within-cluster sum of squares — no arbitrary k. As of Apr 2026, the 90-day window produces 4 distinct behavioral clusters.

Clusters (90D)

—

optimal k (elbow)

Dominant Cluster

—

largest by member count

Coins Clustered

—

with ≥ 60% window coverage

Window

90D

behavioral look-back

Behavioral Cluster Map (PC1 × PC2)

Each dot is a stablecoin, positioned by its first two principal components of the behavioral feature matrix (velocity, growth, volatility, beta). Coins close together behave similarly. Colors indicate cluster membership. The axes capture the two dominant dimensions of behavioral variation — not market cap or price.

Elbow Chart — Optimal Cluster Count

Within-cluster sum of squares (WCSS) for k=2 to 6. The elbow — where adding another cluster yields the smallest marginal improvement — determines the optimal k. Showing this chart makes the k selection auditable rather than arbitrary.

Cluster Profiles

Average behavioral metrics per cluster (un-normalised). Velocity = V/Mcap. Growth = 30/60/90D market cap change. Volatility = σ of daily % changes. Beta = market sensitivity.

Coin-Level Assignments

Every coin's cluster assignment and raw behavioral metrics for the selected window. Sorted by market cap descending.

Interpretation Guide

Institutional / High-Velocity

High velocity (volume/mcap) signals active use in payments, settlement, and exchange flows. These coins are the workhorses of the stablecoin system. The cluster does not require large size — a newer coin that circulates actively will appear here.

Stable / DeFi Native

Lower velocity, moderate growth, and low beta. These coins tend to be locked in DeFi protocols (lending vaults, AMM liquidity) rather than circulating in payment flows. Supply is stable and relatively independent of market-wide events.

Growth Phase

Above-average growth rate and often elevated beta. These coins are gaining market share rapidly — typical of newly launched or recently adopted stablecoins. High beta means their expansion closely tracks broader market inflows.

Declining / Contracting

Negative or near-zero growth sustained over the selected window. These coins are losing market share. The cluster does not imply insolvency — it may reflect competition, regulatory pressure, or capital rotation to alternatives.

Methodology

Four behavioral features are computed per coin for the selected look-back window (30D/60D/90D): (1) Velocity = mean(volume_24h ÷ market_cap) — how actively the coin circulates; (2) Growth rate = (latest mcap − first mcap) ÷ first mcap; (3) Mcap volatility = standard deviation of daily market cap % changes; (4) Market beta = cov(coin daily %, market daily %) ÷ var(market daily %). All four are z-score normalised (StandardScaler) so no single feature dominates by scale. Coins with fewer than two-thirds of the window covered by valid data are excluded.

Clustering uses K-means with k-means++ initialisation (scikit-learn, 30 restarts). Optimal k is selected via the elbow method: WCSS is computed for k=2..6 and the k with the largest second-difference (sharpest bend) is chosen. The 2D scatter uses PCA on the normalised behavioral feature matrix for axis placement — distinct from the market-cap-returns PCA on the Market Factor Analysis page. Data: CoinGecko daily snapshots. Updated daily at 15:20 UTC.

Frequently Asked Questions

What does stablecoin behavioral clustering show?

K-means groups stablecoins by how they actually behave — their velocity, growth trajectory, supply stability, and market sensitivity. Because all features are size-normalised, a small coin that behaves identically to USDC will appear in the same cluster. This makes it possible to identify behavioral peers and outliers that raw market cap rankings obscure.

Why is price volatility not a clustering feature?

For dollar-pegged stablecoins, price variance is near-zero for every coin — it differentiates nothing. For yield-bearing tokens like sUSDE, rising "price" just reflects yield accrual, not a behavioral signal. Market cap volatility is used instead: it captures genuine supply dynamics — minting surges, redemption waves, and market stress — which are far more informative about behavioral type.

How is the number of clusters selected?

The elbow method: we compute within-cluster sum of squares (WCSS) for k=2 through 6, then find the k where marginal improvement drops most sharply (the largest second-derivative of the WCSS curve). The elbow chart is shown on this page so the selection is fully auditable. With the current stablecoin set, the elbow typically lands at k=3 or k=4.

What does it mean when a stablecoin changes cluster between windows?

Behavioral shifts are real signals. A coin moving from "Growth Phase" into "Institutional / High-Velocity" has matured — its growth rate normalized and circulation deepened. A coin drifting toward "Declining / Contracting" is losing share consistently across multiple time horizons, which may reflect competitive pressure, regulatory action, or structural redemptions. The 30D/60D/90D toggle lets you distinguish short-term noise from persistent behavioral trends.

How is the scatter plot constructed?

The two axes are the first two principal components of the normalised behavioral feature matrix — not the market-cap-returns PCA on the Market Factor Analysis page. PCA here serves purely as a 2D projection to make the four-dimensional feature space visually interpretable. Coins close together on the scatter have similar behavioral profiles.