Paper: DeFi-ying the Fed? Monetary policy transmission to stablecoin deposit rates
Authors: Andrea Barbon, Jean Barthélemy, Benoit Nguyen
Date: 2026-09-14
Estimated Reading Time: 25-30 minutes
Major stablecoin issuers pay no interest, so the yield a holder of USDC, USDT or DAI earns comes from lending the tokens on DeFi protocols, and the paper asks whether Federal Reserve policy reaches that yield. On Aave v2 and v3 on Ethereum, from January 2021 to January 2026, deposit rates averaged about 100 bps above the Fed policy rate, with peaks above 1,000 bps, and sat on average 160 bps below it from mid-2022 to the end of 2023. The authors build a model of DeFi lending with two opposing forces: liquidity providers with access to traditional finance withdraw stablecoins after a hike, which raises utilisation and rates, while leveraged crypto investors borrow less when tightening lowers expected crypto returns and collateral values, which lowers them. A Bayesian proxy-VAR identified with high-frequency FOMC surprises finds that a contractionary shock reduces both supply and borrowing, with the USDC deposit rate rising on impact and the USDT rate falling, and USDT converging to the policy variable only beyond a 42-week horizon. Error-correction regressions put the half-life of shocks to the spread between 2.2 and 5 weeks, and for USDC and USDT, cointegration with US money-market rates becomes strong once 2021 is excluded. Wallet-level and high-frequency evidence attributes the persistent spread to sticky liquidity supply, transaction costs and a small share of liquidity providers integrated with traditional finance. The authors conclude that the leverage channel dominates in the short run, and that tokenized T-bills and money market funds could tighten the link, subject to KYC and whitelisting requirements.
Core insights
- Deposit yield is set by leverage demand Aave sets rates algorithmically from utilisation, so that deposit_rate = utilisation * borrow_rate * (1 - reserve_factor), with no reference to any policy rate. As of March 2025 the USDC and USDT curves carried a 0% base rate, slopes of 4% and 10% either side of a 92% kink and a 10% reserve factor, against 5%, 35% and 25% for DAI. Four risk proxies (liquidation, depeg, hack and rate volatility) explain only about a third of the variation in the spread to the 2-year Treasury yield.
- One shock, opposite signs by coin With the shock normalised to a 100 bp rise in the 2-year Treasury yield, USDC borrowing falls by 40% and supply by around 50% in the first week, lifting the USDC deposit rate by roughly 100 bps, before later deleveraging (borrowing down 90% against supply down 80%) attenuates the gain. For USDT, borrowing contracts faster than supply, the deposit rate falls in the short run, and convergence takes beyond 42 weeks. The weekly first-stage F-statistic is 7, slightly below the rule-of-thumb threshold for a strong instrument, and the monthly check runs on a sample the authors call small.
- Unconditional pass-through is weak and often inverted Around a median 50 bp hike, deposit rates rise by about 15 bps after one week, a 0.3-to-1 pass-through. Between March 2022 and July 2023 the Fed raised rates eleven times for a total of 525 bps, while DeFi rates fell from 2.5% to 0.5% by September 2022 and returned to 2.5% only by July 2023; between September and December 2024 a 100 bp cut coincided with a 600 bp rise, from 3% to 9%. Over the full sample the long-run coefficients are small, and cointegration becomes strong for USDC and USDT only once 2021, a year of near-zero policy rates, is excluded.
- Supply responds slowly to demand shocks Of 89 daily rate jumps above 20 percentage points, 59% of the 49 increases were demand-driven and 60% of the 40 decreases came from new supply. When USDC depegged after the SVB failure in March 2023, USDT borrowing surged while USDT supply lagged, and the USDT deposit rate passed 60% annualised. For spread spikes between 5% and 25%, gas prices during the decay window explain between 33% and 51% of the variation in decay time, falling to 10% at a 30% threshold and to 0% and 4% at 35% and 40%.
- A minority of liquidity providers does the arbitrage The TradFi proxy is 8,985 wallets that supplied only USDC on Aave and never held USDT or DAI on-chain, 7.29% of liquidity providers active around policy announcements and roughly 15% of USDC volume. Relative to other liquidity providers these wallets withdraw after hikes and add after cuts, which the authors translate as 1 to 2 percentage points more withdrawal after a hike, and the result holds with day and wallet fixed effects. Proxies built on transaction size or frequency return insignificant coefficients.
The supply-side reward in this market is a share of what leveraged borrowers pay. A depositor on Aave is paid out of borrow interest, borrow demand comes from investors taking leveraged long positions in crypto, and in this sample Bitcoin falls after a contractionary shock, so a hike can cut the reward to suppliers while raising their outside option. The yield on a dollar stablecoin in DeFi behaves as a claim on crypto risk appetite until enough suppliers can arbitrage it against Treasuries. The reserve factor passes a fixed share of borrow interest to the protocol. The paper does not measure protocol revenue, so this is my extrapolation and an assumption: with reserve factors held fixed, Aave's income from these pools follows leverage demand rather than the policy rate, and can fall during a tightening cycle.
The model states what closes the gap. When liquidity providers are risk-neutral and every one of them can move between DeFi and traditional finance without cost, the equilibrium deposit rate equals the risk-free rate plus a premium for expected losses, a one-for-one pass-through. When access is constrained, a higher risk-free rate removes some suppliers but does not change how much each remaining supplier lends, so supply becomes less sensitive to policy precisely when DeFi rates sit below it. The empirical counterpart is the 7.29% of wallets acting as arbitrageurs (with the caveat that a USDC-only wallet is as much a wallet that never needed a second stablecoin as a wallet with a bank account). What share of supply would have to sit with such investors before the arbitrage channel outweighs the leverage channel? The paper does not estimate that threshold, and its closing policy question about the size at which DeFi would challenge rate controllability is the same question from the other side.
The magnitudes need more care than the signs. The VAR is identified from 40 monetary policy surprises, a mean of 0.87 bps and a 5th to 95th percentile range of -3.56 to 5.47 bps, and every impulse response is normalised to a 100 bp move in the 2-year yield. Borrowing down 90% and supply down 80% are therefore responses to a shock far larger than any the instrument contains, and reading them as what a typical FOMC meeting does assumes the response scales linearly down to a few basis points, which the paper does not test. With a first-stage F-statistic of 7 on top, I would treat the difference in sign between USDC and USDT as established and the sizes as indicative. DAI is thinner still: Section 7.3 calls it leverage-dominated, but the reported impulse responses cover only USDC and USDT.
The wallet-level outcome is defined as the change in log10(1 + S), and the authors read coefficients of -0.018 and -0.019 as 1 to 2 percentage points. If the definition is the one estimated (an assumption, since the paper does not restate the units), then change_in_ln = ln(10) * change_in_log10, and the differential withdrawal is larger than stated rather than smaller. Separately, the rate-jump section argues that a 20-point fall requires a prior rise because rates average around 1% in the sample, while the descriptive statistics report mean deposit rates between 4.15% and 4.69% and medians between 3.11% and 3.65%. The argument survives either figure, so the inconsistency affects the text rather than the classification.
The explanations for sticky supply also pull in two directions. For the SVB episode the authors suggest that incumbent liquidity providers with market power may hold back deposits to protect the return on their existing balances, and an appendix model predicts that higher concentration raises the deposit rate and its sensitivity to the risk-free rate. For gas fees, the same paper cites the small scale of individual liquidity positions as corroboration that transaction costs bind. The two can coexist across pools, but neither is tested against the other. It is a bit ironic that the rate which passed 60% during a USDC depeg was the USDT rate, and the trade the authors propose behind it, borrowing USDT to buy discounted USDC, is a leverage trade rather than a rate trade.
Retail users carry this exposure too. With the GENIUS Act prohibiting regulated issuers from paying interest and MiCA prohibiting it since 2024, trading platforms increasingly route stablecoin balances into lending protocols, and Coinbase announced in September 2025 an integration that allocates USDC to Morpho vaults. On the paper's evidence, and assuming Morpho markets respond to policy as Aave's do (the paper studies Aave only), an earn product built this way inherits a yield that tracks crypto leverage and can fall while the policy rate rises. The authors point to tokenized T-bills and money market funds, whose rates follow policy more directly. If those require KYC and whitelisting, do they supply the arbitrageur the paper finds missing, or do they reproduce the segmentation between USDC and USDT inside a new pair of pools?
