Paper: arXiv 2611.26180

Authors: Ahmad Qadi, Hui Gong, Francesca Medda

Abstract

This paper asks whether stablecoin runs carry information about U.S. Treasury yields. Using daily data on USDT, USDC and BUSD from 2021 to 2025 and five stress episodes, it measures flows from changes in token units rather than market capitalisation, which removes the phantom inflows the standard measure records when tokens lose their peg. Flows have no stable relationship with yields in normal periods. During the three panic-driven runs, a one-billion-dollar five-day redemption is associated with an 8.5 basis point fall in the three-month bill yield within ten trading days, while two regulatory episodes carry no such signal. The response is broad across maturities and has the opposite sign to what reserve liquidation predicts, which points to flight to quality. The effect is led by the TerraUSD collapse and corroborated by the USDC/SVB run. Five-day redemptions of roughly USD 2 to 3 billion during a panic mark the scale at which stablecoin stress becomes informative for surveillance.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper demonstrates strong empirical rigor through its data-driven approach, novel flow correction, and state-dependent analysis. While the mathematical complexity is moderate, the clear methodology and significant findings contribute to a high overall score, particularly in its practical implications for central bank surveillance.

Research Flowchart

  flowchart TD
    A["Research Goal: Do stablecoin runs signal US Treasury yields?"] --> B("Methodology: Measure stablecoin flows from changes in token units");
    B --> C{"Data: Daily USDT, USDC, BUSD flows (2021-2025) & US Treasury yields"};
    C --> D["Computational Process: Analyze stablecoin flows vs. yield changes during 5 stress episodes"];
    D --> E{"Findings: No stable relationship in normal periods; Panic-driven runs (3 episodes) signal flight to quality"};
    E --> F["Specific Outcome: $1B 5-day redemption -> 8.5bps fall in 3-month bill yield within 10 days during panic"];
    F --> G["Threshold: $2-3B 5-day redemptions during panic mark informative scale for surveillance"];

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes using five‑day stablecoin redemption flows as a high‑frequency indicator of safe‑asset stress for central‑bank surveillance.

Source sentence
This paper asks whether stablecoin runs carry information about U.S. Treasury yields. The right use of the flow series is as a high-frequency marker that a safe-asset stress regime has begun.

Assumes 6

  • The largest stablecoins hold their reserves mainly in U.S. Treasury bills and other liquid dollar assets. market
    Source sentence
    The largest issuers, Tether (USDT) and Circle (USDC), hold their reserves mainly in U.S. Treasury bills, reverse repurchase agreements and other liquid dollar assets.
  • In normal times stablecoin issuance and redemption are a small source of marginal demand for safe assets. market
    Source sentence
    In normal times stablecoin issuance and redemption are a small source of marginal demand for safe assets.
  • Flows are measured from units outstanding valued at par, assuming token price is at or near $1. data
    Source sentence
    Flows are therefore measured from units outstanding, valued at par: ∆Ni,t = MCapi,t Pi,t −MCapi,t−1 Pi,t−1 , (2) in billions of dollars.
  • All series are aligned to Treasury trading days, implying that stablecoin flow dates are matched to the same calendar as Treasury yields. data
    Source sentence
    All series are aligned to Treasury trading days.
Show 2 more
  • Redemption magnitudes are used as the headline regressor, treating a value of one as a one‑billion‑dollar net redemption. statistical
    Source sentence
    Because a signed flow makes the coefficient on outflows awkward to read, the main regressions use redemption magnitudes: Redemptiont = max(−AggFlowt, 0) and Redemption5d t = max(−AggFlow5d t , 0), so that a value of one is a one-billion-dollar net redemption and a negative coefficient means redemptions go with lower yields.
  • Flow and peg variables are winsorised at the 1st and 99th percentiles, but results are unchanged without winsorising. statistical
    Source sentence
    Flow and peg variables are winsorised at the 1st and 99th percentiles; results are unchanged without 7 winsorising.

Data

  • Stablecoin price, market capitalisation and volume (USDT, USDC, BUSD) · CoinGecko crypto-onchain public
    2021‑01‑04 to 2025‑03‑31 · USDT, USDC, BUSD
    Source sentence
    3 Data and Variable Construction The sample is a daily business-day panel from 4 January 2021 to 31 March 2025, cover-ing the growth of the stablecoin market, the 2022 to 2023 tightening cycle and the five episodes. Daily price, market capitalisation and volume for USDT, USDC and BUSD come from CoinGecko.
  • U.S. Treasury yields (3‑month, 6‑month, 2‑year, 10‑year) · FRED macro public
    U.S. Treasury securities
    Source sentence
    Daily yields on three-month bills (FRED DTB3), six-month bills (DTB6), two-year notes (DGS2) and ten-year notes (DGS10) come from FRED; the three-month bill is the main dependent variable and the others are used for the maturity and spread tests.
  • Effective federal funds rate (DFF) · FRED macro public
    U.S. monetary policy rate
    Source sentence
    Controls are the effective federal funds rate (DFF), the VIX (VIXCLS) and the daily S&P 500 return.
  • VIX index (VIXCLS) · FRED macro public
    U.S. market volatility
    Source sentence
    Controls are the effective federal funds rate (DFF), the VIX (VIXCLS) and the daily S&P 500 return.
  • Daily S&P 500 return · S&P Global macro public
    S&P 500 index
    Source sentence
    Controls are the effective federal funds rate (DFF), the VIX (VIXCLS) and the daily S&P 500 return.

Recognised sources: FRED CoinGecko

Costs

Not applicable The paper reports no trading or execution result.

Finds 5

  • Flows have no stable relationship with Treasury yields in normal periods. statistical
    Source sentence
    Flows have no stable relationship with yields in normal periods.
  • During panic‑driven runs, a one‑billion‑dollar five‑day redemption is associated with an 8.5 basis‑point fall in the three‑month bill yield within ten trading days. statistical
    Source sentence
    During the three panic-driven runs, a one-billion-dollar five-day redemption is associated with an 8.5 basis point fall in the three-month bill yield within ten trading days, while two regulatory episodes carry no such signal. In the three panic episodes the effect is clear: at h = 10 a one-billion-dollar five-day redemption is associated with an 8.5 basis point fall in the three-month bill yield (βrun = −0.085, p =
  • The yield response is broad across maturities and opposite in sign to reserve‑liquidation predictions, indicating a flight‑to‑quality effect. statistical
    Source sentence
    The response is broad across maturities and has the opposite sign to what reserve liquidation predicts, which points to flight to quality. In the panic episodes they lower them.
  • Regulatory episodes show no significant yield response to stablecoin redemptions. negative
    Source sentence
    In the two regulatory episodes there is no effect at any horizon: the point estimates at short horizons are positive, none is significant, and the interaction is never significant.
  • Five‑day redemptions of roughly $2‑3 billion mark the scale at which stablecoin stress becomes informative for surveillance. performance
    Source sentence
    Five-day redemptions of roughly USD 2 to 3 billion during a panic mark the scale at which stablecoin stress becomes informative for surveillance. Five-day aggregate redemptions of roughly $2 to 3 billion across the major tokens are the scale reached in 21 the three panic episodes and the scale at which the signal becomes informative.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 16 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-20 · same card as JSON · A language model chose these items: check the source sentence before relying on one.