Weekly Monthly Insights Charts Tracker Companies Networks Glossary
Indicator

Granger Causality: Lead-Lag Across Major Stablecoin Pairs

Granger causality tests whether one stablecoin's market cap changes predict another's, beyond the second coin's own history (Granger, 1969). It is not classical causation, it is predictive precedence in the time-series sense. This page tracks lead-lag relationships across four pairs spanning the major stablecoin design types: fiat-backed incumbents (USDT/USDC), incumbent vs algorithmic (USDT/DAI, USDC/DAI), and incumbent vs yield-bearing synthetic (USDT/USDE). Select a pair below to inspect its dynamics over the rolling window of your choice. The "All Pairs at a Glance" panel shows where significant lead-lag relationships currently exist across the market.

F(A → B)
loading…
F(B → A)
loading…
Dominant Direction
predictive precedence
Strongest Pair (now)
across all 4 pairs

Rolling Granger F-Statistic

Rolling Granger F-statistic for both directions of the selected pair over the selected window. Amber dashes mark F = 3.84 (p < 0.05); coral dashes mark F = 6.63 (p < 0.01). Values above a threshold reject the null that the leading series adds no predictive value. Sustained crossing of the 3.84 line signals an emerging structural lead-lag relationship.

Current F-Statistics. Both Directions

F-statistics for both directions of the selected pair at the selected window. Bars are coloured when above F = 3.84 (p < 0.05). A higher bar in one direction indicates that direction as the primary predictor. Amber dashes mark the significance threshold.

Window Sensitivity, 30D / 60D / 90D

F-statistics for the selected pair across all three rolling windows. Consistency across windows signals a robust structural relationship. Divergence suggests the lead-lag may be regime-specific rather than persistent.

All Pairs at a Glance

Current F-statistics across all four pairs and both directions at the selected window. Bars above the amber line (F = 3.84) are statistically significant lead-lag relationships. This panel reveals where in the stablecoin market predictive precedence currently exists, independent of which pair you have selected above.

How to Interpret Granger Causality
F > 6.63 (p < 0.01)
Strong Lead Signal

The F-statistic exceeds the 99% significance threshold. Within the tested window, lagged changes in the leading series contain statistically significant predictive information about the following series. Granger results indicate predictive content within the sample, not economic causality, and the relationship can be unstable across regimes.

For enterprise: Lagged changes in the leading issuer's series contain predictive information for the follower within this sample; treat as one input, not a confirmed leading indicator. For policymakers: The leading issuer's flows are worth monitoring alongside other liquidity and concentration metrics.

F 3.84–6.63 (p 0.01–0.05)
Marginal / Emerging Lead

F-statistic is above 95% threshold but below 99%. The relationship is statistically significant but may be driven by a specific regime or short period. If this coincides with a market expansion or stress episode, it may reflect a temporary information spillover rather than a structural lead.

For enterprise: Treat as confirming, not standalone, signal. Monitor across subsequent windows. For policymakers: Worth tracking; may solidify into a robust lead over time.

F < 3.84 (p > 0.05)
Synchronized, No Significant Lead

Neither series provides statistically significant predictive information about the other. The two coins are likely responding to the same contemporaneous signals, macro liquidity conditions, crypto market flows, ETF or treasury flows, rather than to each other's lagged signals. This is the expected baseline for mature, liquid markets where information propagates quickly across issuers.

For enterprise: No exploitable lead-lag at this window, both issuers react simultaneously to demand. For policymakers: Market is informationally efficient across the two issuers; neither sets a leading signal.

Methodology

Test: Granger causality F-test (Granger, 1969) with 1 lag. Tests whether lagged values of X significantly improve predictions of Y beyond Y's own history. Implemented by comparing a restricted model (AR(1) in Y) to an unrestricted model (AR(1) in Y + lagged X). F-statistic: F = ((RSS_r − RSS_u) / 1) / (RSS_u / (n − 3)), where q = 1 restriction, k = 3 parameters (intercept + Y_lag + X_lag). The Frisch–Waugh–Lovell theorem is used for numerical stability: the X coefficient is estimated from residuals after partialling out Y_lag from both Y and X.

Data: Daily market cap % changes (ΔMcap/Mcap) for USDT, USDC, DAI, and USDE from CoinGecko daily snapshots. The test is computed independently for each pair across both directions. Market cap % changes are equivalent to net minting/redemption flow signals for dollar-pegged stablecoins.

Pairs covered: USDT↔USDC (the two incumbent fiat-backed dollars), USDT↔DAI and USDC↔DAI (incumbent vs algorithmic / crypto-collateralized), USDT↔USDE (incumbent vs yield-bearing synthetic). Selected to span the major stablecoin design types.

Thresholds: F = 3.84 (p < 0.05 for F(1, ∞)) and F = 6.63 (p < 0.01 for F(1, ∞)). Exact p-values require numerical CDF evaluation; these thresholds are used as practical decision boundaries. For short windows (n < 50), treat results with caution, finite-sample critical values are slightly higher.

Important caveat: Granger causality is not classical causation. "USDT Granger-causes USDC" means USDT's past values help predict USDC's future values, not that Tether causes Circle to issue stablecoins. It reveals information spillovers and issuance timing dynamics.

Data source: CoinGecko API (daily snapshots). Update frequency: Daily at ~15:30 UTC. Coverage: Apr 2025 – present (449 snapshots). Computation runs entirely in the browser.

Frequently Asked Questions
What is Granger causality and why does it matter for stablecoins?
Granger causality tests whether one time series provides statistically significant predictive information about another, beyond the series' own history (Granger, 1969). For stablecoins, "USDT Granger-causes DAI" means that knowing yesterday's USDT market cap change helps predict today's DAI change, suggesting Tether responds to market conditions before MakerDAO, or that USDT issuance signals that DAI subsequently follows. This reveals monetary leadership dynamics: which issuer sets the pace, and which reacts.
Why test multiple pairs rather than just USDT/USDC?
USDT and USDC are both large fiat-backed dollars and tend to be highly synchronized, they often respond to the same macro signals at the same time, producing near-zero F-statistics. Testing additional pairs (USDT↔DAI, USDC↔DAI, USDT↔USDE) spans different stablecoin design types, fiat-backed, algorithmic/crypto-collateralized, and yield-bearing synthetic, which can have genuinely different response speeds and timing. Lead-lag relationships are more likely to appear across design types than within them.
Is this actual causation, does USDT cause USDC to be issued?
No. Granger causality is predictive precedence, not classical causation. "USDT Granger-causes USDC" means USDT's past values help predict USDC's future values. The economic interpretation is that USDT issuance carries information about market demand that USDC subsequently responds to, or that both issuers react to a common underlying factor (crypto market liquidity demand), but with a 1-day difference in response speed. Do not interpret the F-statistic as proof that Tether controls Circle's operations.
What does the F-statistic mean in practice?
The F-statistic measures how much better the unrestricted model (using both Y's own lag and X's lag) predicts Y, compared to the restricted model (using only Y's own lag). F = 3.84 corresponds to p = 0.05 for large samples, a 5% probability of seeing this result by chance if there is no true relationship. F = 6.63 corresponds to p = 0.01. Practically: F above 3.84 means the lagged variable adds statistically significant predictive power. F below 3.84 means no significant predictive benefit from the lagged series.
Why use market cap % changes instead of raw market cap levels?
Granger causality tests assume the time series are stationary (no unit root). Raw market cap levels are non-stationary, they trend upward over time, which produces spurious statistical relationships. Daily percentage changes are approximately stationary and represent genuine minting/redemption flow signals for dollar-pegged stablecoins. This is the standard approach in empirical monetary economics for testing lead-lag relationships.
Why is the test computed with 1 lag only?
One lag (yesterday's value) tests whether information from one issuer reaches the other within 24 hours, the minimum granularity for CoinGecko daily snapshots. A 1-day lag captures the most direct information spillover between issuers. Adding more lags (2, 3 days) can improve statistical power but risks overfitting with smaller samples. For robustness, the window sensitivity chart shows whether the 1-lag relationship is consistent across 30D, 60D, and 90D rolling windows.