PaperlayerView exact revision on arXiv

Source-linked AI summary

An Empirical Study of DeFi Liquidations: Incentives, Risks, and Instabilities

Kaihua Qin, Liyi Zhou, Pablo Gamito, Philipp Jovanovic, Arthur Gervais

arXiv:2106.06389v2q-fin.GNcs.CR

TL;DR

DeFi lending markets have grown rapidly, but their liquidation mechanisms have limited quantitative evaluation despite exposing borrowers and liquidators to substantial risks. The paper analyzes four major Ethereum lending protocols over two years, systematizes and compares their mechanisms, and finds that fixed-spread designs favor liquidators while an optimal strategy can further increase borrower losses.

  • Problem

    Existing DeFi liquidation mechanisms lacked broad quantitative study, despite growing leveraged lending and the risks of selling collateral at discounts.

  • Method

    The paper analyzes two years of liquidation data from MakerDAO, Aave, Compound, and dYdX and develops a methodology for objectively comparing liquidation mechanisms.

  • Results

    The study finds that fixed-spread mechanisms favor liquidators and reports 807.46M USD in cumulative liquidation proceeds across 28,138 events.

  • Takeaways & Limitations

    Existing designs sell excessive discounted collateral at borrowers’ expense, while a two-liquidation strategy can increase liquidator profit and aggravate borrower loss.

  • Takeaways & Limitations

    The paper leaves objective comparison of alternative auction designs for future work.

Abstract

from arXiv · show

Financial speculators often seek to increase their potential gains with leverage. Debt is a popular form of leverage, and with over 39.88B USD of total value locked (TVL), the Decentralized Finance (DeFi) lending markets are thriving. Debts, however, entail the risks of liquidation, the process of selling the debt collateral at a discount to liquidators. Nevertheless, few quantitative insights are known about the existing liquidation mechanisms. In this paper, to the best of our knowledge, we are the first to study the breadth of the borrowing and lending markets of the Ethereum DeFi ecosystem. We focus on Aave, Compound, MakerDAO, and dYdX, which collectively represent over 85% of the lending market on Ethereum. Given extensive liquidation data measurements and insights, we systematize the prevalent liquidation mechanisms and are the first to provide a methodology to compare them objectively. We find that the existing liquidation designs well incentivize liquidators but sell excessive amounts of discounted collateral at the borrowers' expenses. We measure various risks that liquidation participants are exposed to and quantify the instabilities of existing lending protocols. Moreover, we propose an optimal strategy that allows liquidators to increase their liquidation profit, which may aggravate the loss of borrowers.

1 INTRODUCTION

The paper studies liquidation in Ethereum DeFi lending, where growing leveraged borrowing exposes users to collateral-sale risks. Across four major protocols, it combines longitudinal measurements with mechanism comparison and proposes a strategy that can increase liquidator profit while worsening borrower losses.

  • Empirical scope and findings: 28,138 liquidation events generated 807.46M USD in cumulative proceeds across four protocols, involving 2,011 liquidator addresses.The study covers MakerDAO, Aave, Compound, and dYdX over two years, representing over 85% of Ethereum’s lending market.
  • Empirical scope and findings: A 43% ETH price decline would create up to 1.07B USD of liquidatable collateral on MakerDAO.This scenario mirrors the ETH price decline on March 13, 2020.
  • Empirical scope and findings: Liquidation activity is competitive and can leave bad debt: 73.97% of liquidations paid above-average transaction fees, while Aave V2 accumulated 87.4K USD of bad debt.The measurements also identify 641 unprofitable auction liquidations for liquidators.
  • Mechanism comparison: Fixed-spread liquidation mechanisms favor liquidators because they allow more borrower collateral than necessary to be liquidated.The paper introduces a quantitative methodology for comparing whether mechanisms favor borrowers or liquidators.
  • Optimal liquidation strategy: An optimal fixed-spread strategy can increase liquidation profit by 53.96K USD, or 1.36%, while aggravating borrower losses.The strategy lifts the close-factor restriction through two successive liquidations.

2 LENDING ON THE BLOCKCHAIN

This section introduces permissionless blockchains, smart contracts, DeFi lending, and the terminology used to describe collateralized borrowing and liquidation. It explains how on-chain execution, fees, oracles, and collateral thresholds shape lending operations.

  • 2.2 Decentralized Finance (DeFi): DeFi builds on permissionless blockchains and uses smart contracts to provide publicly verifiable, non-custodial financial services.Ethereum is described as the dominant permissionless blockchain hosting DeFi at the paper’s time of writing.
  • 2.1 Blockchain & Smart Contract: Smart contracts execute immutable on-chain programs, while transaction fees depend on computational gas and the sender’s gas price.Ethereum provides the execution environment through the EVM, and limited block capacity affects transaction inclusion.
  • 2.2 Decentralized Finance (DeFi): Liquidation systems rely on collateral prices supplied by oracles, but on-chain price oracles are vulnerable to manipulation.Prices may come from on-chain exchanges or off-chain oracle services.
  • 2.2 Decentralized Finance (DeFi): Flash loans let users borrow and repay assets within one atomic transaction and are widely used in liquidations.Failure to repay reverses the entire transaction without changing blockchain state.
  • 2.3 Terminology: Blockchain lending uses collateral-backed, typically over-collateralized loans because defaults lack compulsory enforcement on-chain.Borrowers deposit collateral, borrow cryptocurrency, and generally receive debt worth less than the collateral.
  • 2.3 Terminology: A position becomes eligible for liquidation when its health factor falls below 1, allowing anyone to repay debt and claim collateral.The health factor is borrowing capacity divided by outstanding debt.
  • 2.3 Terminology: Liquidation spread is the bonus or discount paid to liquidators, while close factor limits the debt proportion repaid in one liquidation.These parameters govern liquidator incentives and the amount of debt addressed per event.

3 SYSTEMATIZATION OF LENDING AND LIQUIDATION PROTOCOLS

The paper systematizes blockchain lending and liquidation around non-atomic auctions and atomic fixed spreads, then describes their operation across four major Ethereum protocols.

  • 3.1 Borrowing and Lending System Model: Liquidators monitor unhealthy positions and compete to repay debt for discounted collateral, using either atomic fixed spreads or multi-transaction auctions.Fixed-spread liquidations settle in one transaction, whereas auctions require multiple interactions with the lending pool.
  • 3.2.1 Auction Liquidation: MakerDAO uses a two-phase tend-dent auction in which bidders first increase debt repayment for all collateral, then reduce collateral received for full debt repayment.In the dent phase, the remaining collateral C−c_win returns to the borrower.
  • 3.2.1 Auction Liquidation: MakerDAO auctions terminate after configured auction-length or bid-duration conditions, after which the winning liquidator finalizes the liquidation and claims collateral.The auction may also terminate during the tend phase.
  • 3.2.2 Fixed Spread Liquidation: Fixed-spread protocols instantly let liquidators purchase collateral at a predetermined discount after a position becomes liquidatable.Aave permits purchases at up to a 15% discount, while the example liquidation yields a 420 USD liquidator profit.
  • 3.1 Borrowing and Lending System Model: Four studied protocols represent the largest lending platforms by TVL: MakerDAO, Aave, Compound, and dYdX.The reported TVLs are 12.49B USD, 11.20B USD, 10.15B USD, and 247.6M USD, respectively.

4 LIQUIDATION INSIGHTS

Using on-chain events and historical states, the paper measures liquidation activity, profits, participation, fees, and auction behavior across four Ethereum lending protocols.

  • 4.2 Overall Statistics: 807.46M USD of collateral was liquidated across Aave, Compound, dYdX, and MakerDAO over the measured period.The study observes 28,138 successful liquidations from protocol inception through April 2021 and normalizes values using on-chain price oracles.
  • 4.3.1 Liquidator Profit & Loss: Liquidators earned 63.59M USD across 28,138 events, while 641 MakerDAO liquidations lost 467.44K USD because collateral prices changed during auctions.MakerDAO’s March 2020 monthly profit reached 13.13M USD during congestion and a 43% ETH price collapse.
  • 4.2 Overall Statistics: 2,011 unique liquidators participated, with an average profit of 31.62K USD each; the most profitable earned 5.84M USD in 112 liquidations.The most active liquidator performed 2,482 liquidations and earned 741.75K USD.
  • 4.3.2 Fixed Spread Liquidations: 73.97% of liquidations paid above-average transaction fees, indicating competitive liquidator participation.Figure 6 compares each fixed-spread liquidation’s gas price with a 6000-block moving average of block-median gas prices.
  • 4.3.3 Auction Liquidations: MakerDAO recorded 6,762 liquidations, averaging 1.99 bidders and 2.63 ± 1.96 bids per auction.A total of 3,377 auctions ended in the tend phase and 3,385 in the dent phase.
  • 4.3.3 Auction Liquidations: MakerDAO auctions lasted 2.06 ± 6.43 hours on average, and some exceeded configured durations because liquidators failed to finalize them.The longest auction lasted 346.67 hours, with its last bid placed 344.60 hours before termination.

4.4 Risks

The study finds that liquidation mechanisms expose borrowers, lenders, and liquidators to distinct risks. Fixed-spread designs can over-liquidate collateral, while transaction costs and delays create unprofitable opportunities and bad debt.

  • 4.4.1 The Problem of Over-Liquidation: Up to 50% of collateral can be liquidated on Aave and Compound, while dYdX permits liquidation of 100%, potentially exceeding what is needed to restore solvency.The authors identify close-factor selection as a trade-off between limiting borrower losses and reducing liquidation events and transactions.
  • 4.4.1 The Problem of Over-Liquidation: Auction liquidators face collateral-price risk during liquidation, and network congestion can expose borrowers to excessive losses when bots fail to act promptly.The paper notes that alternative auction designs may mitigate over-liquidation, but comparing them is left for future work.
  • 4.4.2 Bad Debts: At a 100 USD repayment cost, 351 Type I and 3,525 Type II bad debts were identified, reducing Aave V2 liquidity by 87.4K USD.Type I bad debt reflects under-collateralization, whereas Type II arises when transaction fees exceed the recoverable excess collateral.
  • 4.4.3 Unprofitable Liquidations: At least 59.1% of Aave liquidation opportunities are unprofitable at a 100 USD process cost, while Compound had 350 such opportunities worth 125,722 USD of collateral.Rational liquidators avoid these positions, so continued health-factor deterioration can turn them into Type I bad debts.
  • 4.4.4 Flash Loans: Fixed-spread liquidators can use flash loans to avoid holding repayment assets, with 623 liquidation flash loans totaling 483.83M USD observed.The flash-loan sequence repays debt, receives discounted collateral, exchanges collateral for the loan currency, and repays principal plus interest.

4.5 Instabilities

The study measures lending-platform instability under cryptocurrency price declines and stablecoin borrowing. All four platforms are sensitive to ETH price declines, while stablecoin strategies reduce but do not eliminate liquidation risk.

  • 4.5.1 Liquidation Sensitivity: The study evaluates liquidation sensitivity by testing whether each debt becomes liquidatable as the price of a given cryptocurrency declines by up to 100%.The measurement uses a snapshot of Aave V2, Compound, MakerDAO, and dYdX at block 12344944.
  • 4.5.1 Liquidation Sensitivity: All four studied platforms are sensitive to ETH price declines, with Aave V2 more stable than Compound despite similar liquidation mechanisms and TVL.Figure 8 measures the collateral amount that would be liquidated as collateral prices decline by up to 100%.
  • 4.5 Remarks: Fixed-spread liquidation rewards liquidators but can impose unnecessary borrower losses through over-liquidation, while excessive fees and overdue liquidations contribute to bad debt.The paper also reports that fixed-spread liquidators can use flash loans to eliminate the risk of holding a specific asset.
  • 4.5.2 Stability of Stablecoins: Stablecoin borrowing mitigates liquidation risk for most of the time, but a maximum 11.1% price difference between USDC and DAI shows that risk remains.The analysis covers DAI, USDC, and USDT prices from May 2020 through April 2021.

5 TOWARDS BETTER LIQUIDATION MECHANISMS

The paper compares liquidation mechanisms using a profit-volume ratio and develops a two-liquidation strategy that bypasses the close-factor restriction. The strategy can increase liquidator profit but may intensify borrower losses.

  • 5.1 Objectively Comparing Liquidation Mechanisms: The profit-volume ratio compares monthly liquidation profit with average collateral volume, with lower values indicating better outcomes for borrowers.The analysis uses DAI-repaid, ETH-collateralized liquidations from November 2019 through April 2021 to reduce asset-price bias.
  • 5.1 Objectively Comparing Liquidation Mechanisms: dYdX has a higher profit-volume ratio than the other platforms, while MakerDAO generally has a smaller ratio than Compound, indicating more borrower-favorable outcomes for auctions.The paper cautions that Aave’s comparatively low ratio may be unrepresentative because DAI/ETH liquidation events are rare.
  • 5.2 Optimal Fixed Spread Liquidation Strategy: The optimal strategy performs two successive liquidations: it leaves the position unhealthy after the first, then liquidates the remaining collateral up to the close factor.Because a position remains liquidatable while unhealthy, this arrangement lifts the single-liquidation close-factor restriction.
  • 5.2.1 Optimality Analysis: The optimal strategy can generate more profit than the up-to-close-factor strategy, with its effectiveness increasing when the collateralization ratio is low.The comparison uses the fixed-spread liquidation profit and the alternative profit of LS·CF·D.
  • 5.2.2 Case Study: A blockchain-state execution shows the optimal strategy could add 49.26K DAI, or 53.96K USD, compared with the original liquidation.The evaluation implements the original and two alternative strategies in Solidity and executes them on the corresponding blockchain state.
  • 5.2.3 Mitigation: Allowing only one liquidation per position per block reduces the optimal strategy’s expected profit, but a rational miner would attempt it across consecutive blocks only with over 99.68% mining power.The mitigation is intended to decrease the strategy’s success probability and protect borrowers from further liquidation losses.

6 RELATED WORK

Related work covers blockchain attacks, borrowing and lending markets, and traditional-finance liquidations. The paper positions its analysis against prior studies of MakerDAO auctions and notes that blockchain liquidations differ fundamentally from TradFi mechanisms.

  • Blockchains and DeFi: Prior blockchain research studies flash-loan, sandwich, front-running, and Miner Extractable Value attacks.These works address broader DeFi and blockchain security mechanisms rather than the paper’s comparative liquidation study.
  • Blockchain Borrowing and Lending Markets: Earlier borrowing-and-lending research analyzes MakerDAO auctions, auction participation costs, and liquidation mechanisms across blockchain lending platforms.The cited work finds many auctions conclude above optimal prices but does not consider potential gas-bidding contests at auction ends.
  • Liquidations in Traditional Finance: Blockchain liquidations are fundamentally different from traditional-finance liquidations in their high-level designs and settlement mechanisms.The paper therefore treats TradFi liquidation research as related but not directly equivalent.

7 CONCLUSION

The paper studies liquidation mechanisms across major Ethereum lending platforms and finds that many sell excessive borrower collateral. It also proposes an unobserved optimal strategy that can increase liquidator profit.

  • 7 CONCLUSION: The study covers lending platforms representing 85% of the blockchain lending market and systematizes their prevalent liquidation mechanisms.Its empirical analysis spans more than two years across four lending protocols.
  • 7 CONCLUSION: Many liquidations sell excessive amounts of borrower collateral, making most liquidation systems unfavorable to borrowers.The conclusion frames this finding alongside extensive data analytics across the four protocols.
  • 7 CONCLUSION: The paper proposes an optimal liquidation strategy that increases liquidator profit and had not yet been observed in practice.The abstracted conclusion does not claim that the strategy was observed in the wild.

A POST-LIQUIDATION PRICE MOVEMENT MEASUREMENT

The study measures collateral-price movement after fixed-spread and auction liquidation events over a 1,440-block window. Only a minority of observed liquidations ended with collateral prices below the liquidation price.

  • A POST-LIQUIDATION PRICE MOVEMENT MEASUREMENT: The measurement records block-by-block oracle prices for 1,440 blocks, about six hours, after fixed-spread settlement or auction initiation.The observation classifies post-liquidation movement as horizontal, rise, fall, rise-fall, or fall-rise.
  • A POST-LIQUIDATION PRICE MOVEMENT MEASUREMENT: 19.07% of 28,138 observed liquidations had collateral prices still below the liquidation price at the end of the observation window.If those cases had used auctions, the liquidator might have suffered a loss from further collateral-price decline.

B MONTHLY DAI/ETH LIQUIDATIONS

This section presents monthly liquidation counts for DAI/ETH markets across five Ethereum lending platforms, using Table 8 and comparing them in Figure 9.

  • Table 8 reports monthly DAI/ETH liquidations across Aave V1, Aave V2, Compound, dYdX, and MakerDAO.
  • Figure 9 uses these monthly liquidation counts to compare the five platforms.

C REASONABLE FIXED SPREAD LIQUIDATION CONFIGURATIONS

This section analyzes fixed-spread liquidation configurations for liquidatable borrowing positions and derives conditions under which liquidation can increase health factor. It concludes that under-collateralized positions cannot have their health factor increased by fixed-spread liquidation, while over-collateralized positions require 1−LT(1+LS) > 0.

  • The analysis models a position as POS = ⟨C, D⟩, with collateral value C and debt value D, then considers health factor below 1.
  • A liquidator repays debt r and receives collateral worth r·(1 + LS), after which the position’s health factor is evaluated.
  • Fixed-spread liquidation never increases the health factor of an under-collateralized position.
  • For an over-collateralized liquidatable position, 1−LT(1+LS) > 0 is required for fixed-spread liquidation to increase health factor.
Loading 2106.06389v2…