Skip to content
DeFi Reference

How pegs hold — arbitrage, PSMs, and de-pegs

1. TL;DR

Section titled “1. TL;DR”

A stablecoin’s peg is not enforced by a rule that forbids other prices — it holds because anyone can profit by trading against a deviation (buying below $1 and redeeming at $1, or minting at $1 and selling above it) until the deviation closes. A Peg Stability Module (PSM) puts that arbitrage on-chain and makes it instant and permissionless for a specific pair, such as DAI/USDS against USDC, at the cost of tying the stablecoin’s fate to whatever it’s pegged against. Real incidents show both failure modes: USDC itself fell to roughly $0.87 in March 2023 when $3.3 billion of its reserves were frozen at a failed bank (a reserve/liquidity problem that propagated into DAI through its PSM), while Solana’s Cashio (CASH) went to zero the same year for an entirely different reason — a smart-contract bug that let an attacker mint tokens against fake collateral, with no reserve shortfall involved at all.

2. Explain it simply

Section titled “2. Explain it simply”

Analogy

Section titled “Analogy”

A peg is like a store’s price-match guarantee: “find it cheaper elsewhere and we’ll match it, refunding the difference.” As long as the store has enough inventory to honor every match request, nobody rational buys elsewhere for less, and nobody sells to the store for more than the match price — so the market price converges to the store’s price without the store ever needing to set prices everywhere. A PSM is that price-match booth running permanently, with no manager approval needed and a fixed exchange partner (usually USDC) instead of “elsewhere.”

ELI5

Section titled “ELI5”

A digital dollar stays worth about $1 because people can trade it in for a real dollar (or a very similar token) whenever it’s worth less, and trade a real dollar in for it whenever it’s worth more — both moves make money for whoever does them, and doing them pushes the price back toward $1. A “peg support tool” is just a piece of code that lets anyone do this trade instantly and automatically instead of asking a company for it. But this only works if there’s actually a real dollar (or similar token) on the other side of every trade — if that backing itself gets stuck or turns out to be worth less than a dollar, the digital dollar can lose its own value too.

Step-by-step walkthrough

Section titled “Step-by-step walkthrough”

Scenario: DAI is trading at $1.003 on the open market (above its $1 peg); Sky’s PSM has a sellGem fee (tin) of 0.01% (1 basis point) for USDC → DAI.

  1. Before. DAI trades at $1.003; the PSM holds USDC and DAI/USDS reserves; an arbitrageur, Sam, holds 1,000,000 USDC.
  2. sellGem. Sam calls sellGem, paying 1,000,000 USDC and a 0.01% fee (100 USDC), and receives 999,900 DAI from the PSM at the fixed rate.
  3. Sell on the open market. Sam sells the 999,900 DAI at the market price of $1.003 each, receiving $1,002,900 worth of value.
  4. Profit and price pressure. Sam’s profit is roughly 1,002,900−1,000,000=$2,9001{,}002{,}900 - 1{,}000{,}000 = \text{\textdollar}2{,}900 (before gas). His selling adds DAI supply to the open market, pushing its price down toward $1.
  5. After. As more arbitrageurs repeat this trade, DAI’s market price falls until it no longer exceeds $1 plus the PSM’s tin fee — the arbitrage profit disappears exactly at that point, which is why the PSM’s fee, not $1.00 exactly, defines the practical upper bound of DAI’s price.

Common misconceptions

Section titled “Common misconceptions”
  • Myth: A peg is a hard price enforced by code, like a price ceiling. Reality: it’s an equilibrium sustained by profit-seeking arbitrage; if arbitrage capital, redemption access, or reserves are constrained, price can and does deviate temporarily even for a fully collateralized coin.
  • Myth: A PSM guarantees its stablecoin always trades at exactly $1. Reality: the PSM only bounds price within its fee band, and only up to the PSM’s capacity (buf) and as long as the reference asset it swaps against (e.g., USDC) itself holds its peg.
  • Myth: USDC’s March 2023 de-peg meant Circle was insolvent. Reality: the $3.3 billion was a liquidity/access problem — funds frozen at a failed bank pending resolution — not a permanent loss; Circle stated it would “stand behind USDC and cover any shortfall using corporate resources, involving external capital if necessary” (Circle, via CoinDesk, 2023-03-11), and the peg recovered within days.
  • Myth: every de-peg is caused by insufficient reserves. Reality: Cashio’s CASH token collapsed not from a reserve shortfall but from a smart-contract validation bug that let an attacker mint tokens against fabricated collateral — an exploit, not an arbitrage or reserve failure (see §6).
  • Myth: a brief de-peg is harmless once the price recovers. Reality: even a short de-peg can trigger real, permanent liquidations in protocols that use the depegged asset as collateral or as a price-oracle input before it recovers (see /lending/liquidations/ and /oracles/oracle-manipulation/).

If you only remember one thing

Section titled “If you only remember one thing”

A peg is an arbitrage equilibrium, not a promise — a PSM makes that arbitrage instant and on-chain for one specific pair, which tightens the peg in normal times but transmits that pair’s own problems directly into the stablecoin during a crisis.

3. How it works

Section titled “3. How it works”

The no-arbitrage band

Section titled “The no-arbitrage band”

For a stablecoin redeemable at par against a reference asset with a round-trip fee ff (redemption fee plus gas/time cost), rational arbitrage keeps the market price pp within

1−f  ≤  p  ≤  1+f1 - f \;\le\; p \;\le\; 1 + f

Outside this band, buying (if p<1−fp < 1-f) or minting-and-selling (if p>1+fp > 1+f) is profitable and pulls price back inside it. Off-chain issuer redemption (see /stablecoins/fiat-backed-and-cctp/) has a wide effective ff because of minimum sizes, KYC delays, and settlement time; an on-chain PSM shrinks ff to just its stated swap fees, which is why PSM-backed pegs are typically far tighter day-to-day.

The PSM’s specific fee band

Section titled “The PSM’s specific fee band”

Sky’s LitePSM implements exactly this with two functions and two fees: sellGem (USDC → DAI/USDS, fee tin) and buyGem (DAI/USDS → USDC, fee tout). Per Sky’s documentation, “the tin fee is assessed on all sellGem transactions, and the tout fee is assessed on all buyGem transactions,” both “set by Sky Ecosystem governance” (Sky Protocol docs, fetched 2026-08-29), giving a governance-tunable band [1−tout, 1+tin][1-tout,\ 1+tin] around $1 specifically for DAI/USDS. Crucially, governance can widen this band to infinity as a circuit breaker: “governance can set either tin or tout to the value of type(uint256).max, halting sellGem and buyGem transactions respectively” (Sky Protocol docs, fetched 2026-08-29) — exactly the lever Sky used in March 2023 (see §6).

Why the PSM transmits its reference asset’s problems

Section titled “Why the PSM transmits its reference asset’s problems”

Because the PSM lets USDC flow freely into DAI/USDS’s own reserve, DAI’s own price becomes partially a weighted average of USDC’s price and the price of DAI’s other collateral. If USDC depegs to pUSDC<1p_{USDC} < 1, arbitrageurs will still sellGem USDC for DAI as long as DAI trades above pUSDCp_{USDC}, transmitting USDC’s depeg directly into DAI in proportion to how much of DAI’s backing sits in the PSM — this is the mechanism behind the incident in §6, not a separate bug.

PegKeepers and other on-chain peg tools

Section titled “PegKeepers and other on-chain peg tools”

Sky/Maker’s PSM is the most widely used design, but it is not unique: Curve’s crvUSD uses PegKeepers, contracts that can mint crvUSD directly into (or burn it out of) Curve’s crvUSD/stablecoin pools when price deviates, functioning as an AMM-embedded analogue to a PSM for defending crvUSD’s own peg (see /stablecoins/cdp-stablecoins/ for crvUSD’s broader design). The general pattern — a permissionless, on-chain, fee-bounded swap against a trusted reference asset — recurs across CDP-stablecoin designs precisely because off-chain issuer redemption (§3 above) is too slow and restrictive to hold a tight peg on its own.

4. Worked numeric example

Section titled “4. Worked numeric example”

Part A — routine arbitrage (continuing §2): with tin = 0.01%, the PSM keeps DAI within roughly [0.9999,1.0001][0.9999, 1.0001] of USDC’s own price under normal conditions, since any deviation beyond that band is arbitraged away net of the fee, as computed in §2 ($2,900 profit on a $1M round-trip at a $1.003 market price).

Part B — how much of USDC’s March 2023 de-peg passed through to DAI. At the time of the Silicon Valley Bank failure, USDC represented roughly 51.87% of DAI’s total collateral (about $4.42 billion), with the rest of DAI’s backing (48.13%) in other, unaffected collateral (Cointelegraph, 2023-03, secondary reporting of on-chain collateral data). If DAI’s price moved as a simple value-weighted average of its collateral’s marks, and USDC fell to its reported trough of about $0.87 while the rest of DAI’s collateral held at $1.00, the implied DAI price is

pDAI≈0.5187×0.87+0.4813×1.00=0.4513+0.4813=0.9326p_{DAI} \approx 0.5187 \times 0.87 + 0.4813 \times 1.00 = 0.4513 + 0.4813 = 0.9326

— a roughly 6.7% drop. This back-of-envelope figure lines up closely with the approximately 7.4% drop DAI actually experienced (Cointelegraph, 2023-03), suggesting the PSM’s proportional pass-through mechanism (§3) explains most of DAI’s move, with the small remaining gap plausibly due to additional market panic/liquidity effects beyond pure collateral-value arithmetic. This is presented as an illustrative reconstruction, not a claim that the actual price-formation process was this simple.

5. Where it’s used

Section titled “5. Where it’s used”

Ethereum

Section titled “Ethereum”
  • Sky (Maker) PSM / LitePSM — DAI/USDS ↔ USDC at a governance-set fee band; combined DAI supply stood at roughly $4.47 billion (as of 2026-08, CoinLaw, secondary aggregator). developers.skyeco.com
  • Curve PegKeepers — an AMM-embedded, mint/burn-based peg-defense tool for crvUSD, conceptually parallel to a PSM; see /stablecoins/cdp-stablecoins/.
  • Issuer redemption desks — Circle’s and Tether’s own redemption processes are the “off-chain PSM” for fiat-backed coins, with wider bands due to minimums and settlement delay; see /stablecoins/fiat-backed-and-cctp/.

Solana

Section titled “Solana”
  • n/a for an on-chain PSM-style contract at comparable scale. Solana’s fiat-backed stablecoins rely on off-chain issuer redemption and CCTP’s cross-chain burn-mint parity (see /stablecoins/fiat-backed-and-cctp/) rather than a dedicated on-chain swap-against-reference-asset contract; CCTP enforces 1:1 parity of the same asset across chains, which is a related but distinct guarantee from a PSM’s swap between two different stablecoins.
  • Cashio (CASH) — an over-collateralized (not PSM-defended) stablecoin whose collateral-validation bug destroyed its peg entirely in March 2022; see §6.

6. Risks, attacks, and incidents

Section titled “6. Risks, attacks, and incidents”
  • USDC de-peg, 10–13 March 2023. Silicon Valley Bank failed on 2023-03-10; Circle disclosed on 2023-03-11 that $3.3 billion of USDC’s roughly $40 billion in reserves (about 8%) was held at SVB. USDC traded down to approximately $0.87 (roughly a 13% deviation) amid the uncertainty (CoinDesk, 2023-03-10/11). Circle stated it would “stand behind USDC and cover any shortfall using corporate resources, involving external capital if necessary” (Circle, via CoinDesk, 2023-03-11). U.S. regulators guaranteed all SVB depositors on 2023-03-12, and USDC regained its peg by 2023-03-13 (CoinDesk, 2023-03-13). The depeg propagated into DAI, which lost about 7.4% of its value given USDC’s roughly 51.87% (~$4.42 billion) share of DAI’s collateral at the time via the PSM (Cointelegraph, 2023-03); MakerDAO governance responded within days by raising PSM swap fees and capping the daily USDC-to-DAI mint at 250 million DAI. Root cause: bank-failure-driven reserve illiquidity at the issuer, transmitted through the PSM’s proportional pass-through mechanism into a second stablecoin.
  • Cashio (CASH) infinite-mint exploit, March 2022, Solana. Cashio’s minting logic failed to validate the mint field on a saber_swap.arrow account, letting an attacker construct a fake collateral account and mint CASH against effectively worthless collateral. Estimated losses range from about $28 million (DeFiLlama) to roughly $50–52 million (samczsun and other reports) — sources disagree on the exact figure, consistent with the range typically cited for unaudited-protocol exploits of this kind. CASH’s price collapsed to near zero (around $0.00005) and did not recover. Root cause: a missing ownership/mint-field validation check in an unaudited contract integrating with Saber’s LP tokens as collateral — an exploit, not an arbitrage or reserve failure, and structurally unrelated to the USDC incident above.
  • Redemption friction as a slower-motion risk. Even solvent, well-collateralized stablecoins can trade off-peg for hours or days if redemption requires minimum sizes, KYC, or manual processing — a liquidity/frictions problem distinct from insolvency, and the reason on-chain PSMs (§3) exist at all.
  • Oracle and collateral contagion. A depegged stablecoin used as loan collateral or referenced by a price oracle elsewhere can trigger real liquidations in other protocols even after its own price recovers, if those protocols marked positions during the depeg window (see /lending/liquidations/ and /oracles/oracle-manipulation/).

7. Open problems

Section titled “7. Open problems”
  • How tight should a PSM’s fee band be? Tighter bands hold the peg more precisely in normal times but, as March 2023 showed, also mean a larger share of the stablecoin’s fate rides on a single reference asset; looser bands or lower caps reduce contagion risk at the cost of a looser everyday peg. Sky’s post-SVB parameter changes show this is actively tuned rather than solved.
  • Single-issuer concentration in PSM collateral. Whether CDP stablecoins should cap or diversify how much of their backing can flow through any single reference asset’s PSM is a live governance question these sources do not resolve definitively.
  • Would an audit have caught Cashio’s bug? Cashio was unaudited; whether standard audit practices reliably catch missing-validation bugs in cross-protocol collateral integrations (as opposed to more commonly audited-for issues) is debated rather than settled (see /security/bug-classes/).
  • Cross-chain peg consistency. CCTP (see /stablecoins/fiat-backed-and-cctp/) guarantees fungibility of the same token across chains, but does not guarantee that local market liquidity on every chain is deep enough to prevent a chain-specific price deviation even when the issuer itself is fully solvent — an open question for any multi-chain stablecoin’s peg stability.

8. Ethereum vs Solana

Section titled “8. Ethereum vs Solana”
AspectEthereumSolana
On-chain peg-defense contractPSM/LitePSM (DAI/USDS ↔ USDC), Curve PegKeepers (crvUSD)n/a at comparable scale
Cross-token arbitrage boundGovernance-set fee band (tin/tout)n/a
Cross-chain parity mechanismCCTP burn-mint (same token, across chains)CCTP burn-mint (same token, across chains)
Documented de-peg incidentUSDC/SVB, March 2023 (reserve illiquidity, propagated via PSM)Cashio, March 2022 (smart-contract exploit, unrelated to reserves)
Circuit breaker used in a real incidentYes — Sky raised PSM fees and capped daily mint within daysn/a — Cashio’s failure was not recoverable via a parameter change

The two chains’ documented incidents illustrate genuinely different failure classes rather than a difference in peg-defense sophistication: Ethereum’s PSM-based DAI/USDS experienced a slow-motion, reserve-driven contagion event that governance could partially contain with a parameter change, while Solana’s Cashio experienced an instantaneous, unrecoverable exploit with no peg-defense mechanism (PSM or otherwise) that could have helped once the fake collateral was already minted against. This suggests the relevant comparison for peg safety is not “which chain has a PSM” so much as “does the specific protocol have (a) a fee-bounded arbitrage tool with a circuit breaker, and (b) collateral-validation code correct enough that (a) is even the right line of defense.”

9. Reference doc

Section titled “9. Reference doc”

The reference

Section titled “The reference”

LitePSM — Sky Protocol Docs, fetched 2026-08-29. developers.skyeco.com/guides/psm/litepsm

Summary of the reference

Section titled “Summary of the reference”

This page documents Sky’s gas-optimized Peg Stability Module, the on-chain tool this page’s peg-arbitrage-bound discussion (§3) is built around. It states the core purpose plainly: LitePSMs “stabilize the USDS and DAI stablecoins against their pegged value by enabling users to trade USDS or DAI with other stablecoins like USDC (referred to as ‘gems’) at fixed exchange ratios and fees,” with minimizing gas cost as the design’s specific improvement over the original PSM contract.

The mechanism is two functions: sellGem (“sell USDC and receive USDS,” or DAI in the DAI-paired instance) and buyGem (“buy USDC with USDS”). Fees are charged asymmetrically by direction: tin on every sellGem call and tout on every buyGem call, both set by Sky Ecosystem governance and expressed in WAD (10^18) fixed-point, so that “a tin value of 0.01, represented as WAD/100, would assess a fee of 1% (100 basis points).” The page documents a governance-controlled circuit breaker built into the same fee parameters: setting tin or tout to the maximum possible uint256 value halts sellGem or buyGem transactions respectively — the exact lever Sky used during the March 2023 USDC depeg (§6) to slow USDC inflows into DAI’s collateral base. Finally, the page documents a buf parameter that “dictates” the maximum DAI balance the LitePSM-DAI-USDC instance can hold at any time, bounding how much of DAI’s supply can be concentrated through this single, fast-arbitrage channel.

Read together, the page describes a peg-defense tool that is simultaneously DAI/USDS’s tightest, fastest arbitrage bound against USDC and its single largest concentration-risk channel — the same mechanism that makes the peg precise in calm markets is what transmitted USDC’s own crisis directly into DAI’s price in March 2023.

Key quotes

Section titled “Key quotes”

“LitePSMs stabilize the USDS and DAI stablecoins against their pegged value by enabling users to trade USDS or DAI with other stablecoins like USDC (referred to as ‘gems’) at fixed exchange ratios and fees.” (Core Purpose)

“The tin fee is assessed on all sellGem transactions, and the tout fee is assessed on all buyGem transactions.” (Fee Structure)

“Governance can set either tin or tout to the value of type(uint256).max, halting sellGem and buyGem transactions respectively.” (Fee Structure)

“The maximum DAI balance held by LitePSM-DAI-USDC at any time is dictated by the buf value set by Sky Ecosystem governance.” (Debt Ceiling)

How to read the original

Section titled “How to read the original”

Background needed: the general no-arbitrage peg logic in §3 above, and what “gem” means in Sky’s terminology (any external collateral asset the PSM accepts, here specifically USDC). Skip the WAD fixed-point encoding details on a first pass — the concept (fees as fractions of 10^18) matters more than the literal scaling arithmetic. The hardest part to read correctly is that the tin/tout circuit breaker and the everyday fee band are the same parameter used two different ways: normally a small tin (e.g., a few basis points) just narrows the arbitrage band, but the same variable set to its maximum value becomes an emergency stop — there is no separate “pause” function distinct from setting the fee absurdly high.

What changed since

Section titled “What changed since”
  • This LitePSM design is itself a gas-optimized successor to Maker’s original dss-psm contract; both remain deployed, with LitePSM the newer, cheaper path.
  • The March 2023 USDC depeg (§6) is the clearest real-world demonstration of this page’s fee-parameter circuit breaker in actual use, and directly motivated the parameter changes Sky made in the days following.
  • Sky’s broader collateral diversification since 2023 (adding more real-world-asset and Treasury-backed collateral alongside USDC) is a longer-term response to the concentration risk this page’s mechanism creates, though the specifics of that diversification are outside this page’s scope.

Secondary references

Section titled “Secondary references”
  1. Rate Mechanism (Dai Savings Rate) — Sky Protocol Docs — read for the demand-side peg lever that complements the PSM’s supply-side arbitrage bound; see /stablecoins/cdp-stablecoins/.
  2. Cointelegraph, “Circle’s USDC instability causes domino effect on DAI, USDD stablecoins” (2023-03) — read for the fullest contemporaneous account of the pass-through this page’s §4 reconstructs.
  3. /stablecoins/fiat-backed-and-cctp/ — read for the off-chain issuer-redemption arbitrage that PSMs are a faster, on-chain complement to.

The reference

Section titled “The reference”

Stablecoin “Interchange” — And Why It Doesn’t Work — Alex Grieve (Paradigm), 2026-01-07. paradigm.xyz/writing/stablecoin-interchange-and-why-it-doesnt-work

Summary of the reference

Section titled “Summary of the reference”

This piece argues against a proposed U.S. market-structure amendment that would restrict stablecoin issuer/affiliate reward-sharing to merchant-facing transactions only, modeled on credit-card interchange. The author frames this as a category error: traditional card rewards are “funded primarily by interchange fees” collected from merchants per transaction, whereas “stablecoins work fundamentally differently. They are debit products, not credit products. The revenue that funds the ecosystem doesn’t come from merchant ‘swipes’; it comes from the yield on the reserve assets (like U.S. Treasuries) backing the tokens” — revenue generated continuously by holding the token, not by spending it.

The piece works through the arithmetic of the proposed rule: a user holding $10,000 in USDC at a 4.5% Treasury yield generates $450/year in value for the issuer regardless of whether it’s ever spent, but under a transaction-only reward rule that same user “would have to spend $22,500” annually at a 2% retail reward rate to recapture the same $450 — a stark illustration of what the author calls a “Holding Tax” on anyone using stablecoins for savings, B2B settlement, or remittances rather than retail spending. The argument connects directly to why holders are willing to hold (and thus help sustain demand for, and liquidity behind, redemption arbitrage of) a stablecoin at all: if reserve-yield sharing is legally blocked outside retail transactions, the author warns liquidity and Treasury demand migrate to “offshore, unregulated tokens” without the restriction — an indirect argument that reserve-yield economics, not transaction fees, are what keeps large, redemption-ready balances (the ones that make peg arbitrage liquid and cheap) parked in a given stablecoin.

Key quotes

Section titled “Key quotes”

“Stablecoins work fundamentally differently. They are debit products, not credit products. The revenue that funds the ecosystem doesn’t come from merchant ‘swipes’; it comes from the yield on the reserve assets.” (The Debit Reality vs. The Credit Myth)

“In a stablecoin model, value is a function of AUM (Assets Under Management), not transaction frequency.” (The Debit Reality vs. The Credit Myth)

“By mandating that reward payments can only be triggered by a purchase, Congress would be creating a ‘Holding Tax’ on individuals.” (A “Holding Tax” on Consumers)

“Policy should follow the physics of the underlying technology. Stablecoins generate value by existing, not just by moving.” (Bottom Line)

How to read the original

Section titled “How to read the original”

Background needed: the basic GENIUS Act framework for reserve-backed payment stablecoins (see /stablecoins/payments/) and how credit-card interchange economics work, since the piece’s argument is built entirely on contrasting the two. Skip the political specifics of which lawmakers proposed the amendment on a first pass — the durable content is the debit-vs-credit revenue-model distinction. The hardest part to connect to this page’s main topic is implicit: the piece is about rewards regulation, not peg mechanics directly, but its core claim — that stablecoin economics run on continuous reserve yield rather than per-transaction fees — explains why issuers and large holders have a standing financial interest in redemption-ready liquidity, the same liquidity that keeps peg arbitrage (§3) cheap and fast.

What changed since

Section titled “What changed since”
  • This piece post-dates the GENIUS Act’s 2025-07-18 signing and argues about a specific market-structure amendment being negotiated afterward; the amendment’s outcome is not resolved within the piece itself.
  • The USDC/SVB incident (§6) predates this piece by nearly three years and is not discussed in it, but illustrates the flip side of the same reserve-yield model: the same Treasury-backed reserves that fund holder rewards are also the assets whose custodial bank can fail.

Secondary references

Section titled “Secondary references”
  1. Paradigm, “The GENIUS Act Passed – Now the Real Work Begins” — read for the regulatory framework this rewards debate sits inside; see /stablecoins/payments/.
  2. LitePSM (above) — read for the on-chain mechanics of how reserve-backed value actually gets arbitraged into a tight peg.

10. Sources

Section titled “10. Sources”