DeFi Lending's Risk-Reward Ratio Sparks Debate Between Researchers and Curators

Overcollateralized lending has emerged as one of DeFi's most durable primitives.

Morpho alone holds roughly $7 billion in TVL, according to DeFiLlama, with distribution via Coinbase, Kraken, and other front ends. Apollo Global Management has committed to acquiring up to 9% of MORPHO's token supply over four years, and the Ethereum Foundation has deployed nearly $19 million into the protocol's vaults.

But a quantitative analysis published Sunday by Dirt Roads, a DeFi research publication authored by Luca Prosperi, has sparked a debate over whether the depositors fueling that growth are being systematically undercompensated, or whether the lending primitive is working exactly as it should.

Bear Case: Depositors Are Selling Puts They Don't Understand

Prosperi's analysis adapts the Black-Cox first-passage framework – a refinement of Merton's 1974 structural credit model – to DeFi collateralized debt positions. In this context, depositing $USDC into a Morpho vault backed by $ETH collateral is equivalent to holding a risk-free bond and simultaneously selling a put option on that collateral, with the liquidation loan-to-value (LLTV) acting as the strike price.

Calibrated to $ETH's approximately 75% annualized realized volatility, jump intensity of 1.5 events per year with a mean jump size of -8.3%, and an LLTV of 86% against a 70% starting LTV, the model shows that the appropriate credit spread ranges from 250 to 400 basis points above the risk-free rate, in this case the Fed’s Secured Overnight Financing Rate (SOFR).

Observed depositor rates in flagship Morpho $USDC markets are roughly 2-4% APY – thin margins above SOFR, which currently stands at 3.65%.

Crypto investor Santiago Roel endorsed the findings, arguing that $11.7 billion in Morpho vaults is retail capital funding crypto-collateralized lending "thinking it's a savings account." No institution, he says, would accept near risk-free rates to come on-chain. He pointed to a structural shift from early DeFi — when triple-digit APYs at least compensated for risk — to a present where vaults with completely different risk profiles present the same thin yields, and depositors simply pick the highest number.

"Last cycle we saw a lot of retail pour savings into algo stablecoins promising 'risk-free' yield," Roel wrote. "This cycle vaults have a lot of demand but they are mispriced for the level of risk."

Bull Case: It's a Repo, Not a Put Option

The pushback came swiftly from practitioners with skin in the game and challenged not just the model's inputs but its foundational analogy.

Steakhouse Financial's adcv, whose firm curates the primary Morpho vaults that Coinbase routes retail deposits through, argues that on-chain lending is structurally closer to a repurchase agreement than a put option sale.

In a repo, one party temporarily exchanges an asset for cash with a commitment to repurchase and, critically, the lender holds the collateral outright throughout the transaction. On Morpho, collateral is locked in smart contracts and can be seized and liquidated atomically if value declines toward the LLTV threshold. The lender's exposure is bounded not by the theoretical option payoff on the collateral's full volatility distribution, but by the narrow residual risk that liquidation mechanics fail to make the lender whole.

This reframing leads to adcv's central empirical objection: the loss-given-default (LGD) parameter. Prosperi's model sets LGD at approximately 5%, derived from Morpho's formulaic liquidation incentive. But the liquidation penalty is a cost borne by borrowers — not a loss absorbed by lenders. For liquid crypto-native collateral on prime markets, on-chain liquidation has historically resulted in near-zero bad debt for depositors because the overcollateralization buffer, continuous oracle monitoring, and open liquidator competition work as designed.

Steakhouse's own data supports the claim. During the sharp selloff in late January and early February, when BTC fell 17% and $ETH dropped 26% in a single week, Morpho processed approximately $238 million in liquidations. Users of Steakhouse's vaults absorbed zero bad debt and maintained full withdrawal liquidity throughout.

"If you set the LGD parameter to a few basis points over 0% rather than approximately 5%, the model outputs fall exactly in line with observed rates at around 3-30 basis points," adcv wrote.

Hasu, a strategy lead at Flashbots, made the same point more bluntly.

"Great model, but bad data in, bad data out," he wrote. “If you use the historically observed level of bad debt on Morpho prime markets, even with a big safety buffer, the result changes: Now, depositors should demand an excess return of only 3-30bps, which is in line with rates observed in the wild."

The Real Risk Is Fundamental, Not Market

MonetSupply, a contributor at Spark, offered a third perspective that aligns broadly with the curators' position but redirects the risk conversation entirely. The bulk of the risk in on-chain prime repo, he argued, is not from market price-jump risk – the variable that Prosperi's model centers on – but from fundamental and technical risks embedded in collateral assets and oracle mechanisms.

Most blue-chip collateral in Ethereum DeFi consists of tokenized Bitcoin (WBTC, cbBTC) or liquid staking tokens (wstETH, weETH). These issuers have long track records, but remain subject to custody and key management failures, smart contract vulnerabilities, and business continuity risks. Oracle providers introduce an additional dependency layer. The probability of incidents across these vectors is low, MonetSupply argues, but losses in a failure case can reach 100% of exposure – a fat-tailed distribution that Merton-style market risk models do not capture.

He pointed to the most recent major DeFi loss events – the Resolv exploit and the Drift Protocol vault drain – as evidence. Both were driven by fundamental risk factors, not market volatility. "As a DeFi lender, the primary driver of risk is these fundamental factors rather than jump risk," he wrote.

MonetSupply also offered the most rigorous version of the structural premium argument, framing it through the lens of liquidity premia and convenience yield. For traditional finance investors, prime money market funds and T-bills are the benchmark liquid assets, and they would never accept sub-SOFR yields. But for crypto-native actors, the relevant measure of liquidity is not speed-to-bank-account but speed-to-on-chain-execution. A directional crypto fund facing even a one-hour delay between requesting redemption of a money market fund and receiving a wire to their exchange account could miss a 5-10% move in a volatile asset, he argued, wiping out years of excess risk-adjusted return over on-chain repo.

Convenience yield — the implied return on holding inventory close at hand — provides the same logic from a different angle. If on-chain actors derive meaningful benefit from having capital instantly deployable within the crypto ecosystem, even if that benefit is realized infrequently, it can be entirely rational to accept risk-adjusted returns below SOFR on prime repo.

Spark's own USDT savings vault, MonetSupply noted, maintains over $700 million in available withdrawal capacity against $885 million in total deposits, far exceeding those of typical on-chain lending markets, which already offer a significant liquidity advantage over off-chain cash equivalents.

DeFi's Structural Advantages

A separate thread in the debate argues that the risk-free rate comparison itself is flawed on even simpler grounds.

Pseudonymous trader MilliΞ contends that DeFi yield carries structural properties traditional fixed income does not: composability that enables permissionless derivative applications, censorship-resistant access without custodians who can "play silly games with you," and instant withdrawals versus the 30-day redemption windows typical of money market instruments.

"This may not matter to most of us first-worlders," they wrote, "but it sure matters to the remainder of the planet."

Where Both Sides Agree

Nobody disputes that the vast majority of retail depositors flowing into Morpho through exchange front-ends do not understand the credit exposure they are taking, and that vault risk profiles vary dramatically even when headline yields look similar.

Similarly, no one disputes that the track record supporting the curators' optimistic loss assumptions is short and tested only in broadly favorable conditions; a point underscored by the Resolv exploit that cascaded across fifteen Morpho vaults in March, and the Stream Finance collapse that hit lending markets in November 2025. Steakhouse's own vaults avoided those losses, but other curators' depositors were not as fortunate.

Prosperi's analysis also flags concerns outside the LGD debate. Leverage looping strategies, such as recursive wstETH/WETH or sUSDe loops at 7-10x effective leverage, behave not as credit products but as leveraged carry trades on mean-reverting basis spreads, where a 5% depeg at 10x leverage triggers liquidation. And the growing push to onboard non-crypto-native collateral breaks every assumption in the framework simultaneously: unobservable volatility, discrete oracle monitoring, multi-week liquidation delays, and jurisdictional enforcement risk.

The Real Test

The core disagreement is over which measure of risk matters: the structural exposure embedded in the position, or the empirical loss history of the platform. Prosperi and Roel argue the former; Hasu, adcv, MonetSupply, and the curator ecosystem argue the latter – while adding that the model is looking at the wrong risk entirely, and that rational actors may have good reasons to accept thin or even negative spreads over SOFR.

Structural models can overstate market risk by assuming passive borrower behavior and ignoring the efficiency of on-chain liquidation mechanics, which have performed as advertised even under severe conditions. But they may understate the fundamental risks that MonetSupply identifies, which lie entirely outside the analysis framework. Meanwhile, empirical models can understate risk by extrapolating from a short, favorable sample.

As institutional allocators expand on-chain credit exposure, the question may ultimately be settled not by models but by the next sustained drawdown, or the next fundamental failure.

"The mispricing will become visible when the market turns," Prosperi wrote. The curators are betting it won't, and vault depositors agree with them, at least for now.

This article was written with the assistance of AI workflows. All our stories are curated, edited and fact-checked by a human.