Case Study 1: Black Thursday March 2020 — When $8M in ETH Was Liquidated for $0

Background: The World Falls Apart

On March 12, 2020, as the COVID-19 pandemic triggered a global financial panic, the cryptocurrency market experienced its most violent single-day crash since 2018. Bitcoin fell from $7,900 to $3,800 — a 52% decline in 24 hours. Ethereum fell from $194 to $95 — a 51% decline. Traditional markets were in freefall too: the S&P 500 had its worst day since 1987, and crude oil prices collapsed.

For DeFi, this was the first real stress test. The total value locked in DeFi protocols was approximately $1 billion — a tiny fraction of what it would become by 2021, but at the time, it represented the entirety of the decentralized finance experiment. The question was not academic: could algorithmic, smart-contract-based financial systems survive a real market crash?

The answer, it turned out, was "barely" — and one protocol nearly did not survive at all.

MakerDAO's Vault System

To understand what happened on Black Thursday, you need to understand MakerDAO's lending mechanics, which are closely related to (and in many ways the predecessor of) the Aave and Compound systems described in this chapter.

MakerDAO allows users to deposit ETH (and other approved collateral) into "vaults" and borrow DAI, a stablecoin. The system requires overcollateralization — at the time, the minimum collateral ratio for ETH was 150%, meaning you needed $150 in ETH to borrow 100 DAI (worth $100).

When a vault's collateral ratio fell below 150%, it was subject to liquidation. MakerDAO's liquidation mechanism was an auction system: the protocol would auction off the seized collateral to the highest bidder. Bidders (liquidation bots) would compete to offer the most DAI in exchange for the collateral, and the proceeds would cover the vault's debt. Any surplus collateral was returned to the vault owner.

This system had been tested in smaller market drops and worked as designed. Liquidation bots competed vigorously, collateral sold at or near market price, vault owners lost a small liquidation penalty, and the protocol remained solvent.

March 12 was not a smaller market drop.

The Cascade Begins

As ETH plummeted, thousands of vaults simultaneously fell below the 150% collateral ratio. The number of liquidations that needed to occur was unprecedented — the system had never processed this volume.

Simultaneously, the demand for Ethereum block space exploded. Everyone — DeFi users, traders, exchanges, liquidation bots — needed transactions to be confirmed immediately. Users wanted to add collateral to their vaults. Traders wanted to close positions. Exchanges wanted to process withdrawals. Liquidation bots wanted to execute liquidations.

Gas prices — the fee paid to have transactions included in Ethereum blocks — spiked from around 10-20 gwei to over 200 gwei. Some transactions were paying 500+ gwei. A transaction that normally cost $0.50 now cost $50-100.

This gas spike was the first domino.

The Liquidation Bots Fail

Most MakerDAO liquidation bots were configured with gas price limits. They were profitable at 20-50 gwei. At 200+ gwei, the gas cost of executing a liquidation ate into (or exceeded) the liquidation profit. Many bots simply stopped operating.

But the problem was worse than uneconomical liquidations. Ethereum's mempool (the queue of pending transactions) was so congested that even bots willing to pay high gas prices could not get their transactions included in blocks. Transactions were being dropped, stuck, or replaced. The infrastructure that the liquidation ecosystem depended on — Ethereum nodes, RPC providers, transaction relay services — was overwhelmed.

The result: as vaults fell below the 150% threshold, the liquidation auctions launched, but in many cases, there was only one bidder. Or zero bidders.

Zero-Bid Auctions

This was the catastrophic failure. MakerDAO's auction system at the time had a critical design flaw: there was no minimum bid. If only one bidder showed up to an auction, they could bid any amount — including zero DAI.

On Black Thursday, several liquidation bots submitted bids of 0 DAI for collateral lots worth hundreds of ETH. With no competition (other bots were disabled by gas costs or stuck transactions), these zero bids won. The winning bidders acquired ETH for free.

The numbers were staggering:

• Approximately 1,200 vaults were liquidated.
• Of these, a significant number were liquidated at bids far below market value.
• An estimated $8.32 million worth of ETH was seized in zero-bid or near-zero-bid auctions.
• The protocol incurred approximately $6.65 million in "bad debt" — debt that could not be recovered because the collateral was sold for far less than its value.

For vault owners, the devastation was total. Many had deposited ETH expecting a standard liquidation process — losing maybe 13% (the liquidation penalty) if prices crashed. Instead, they lost 100% of their collateral. Their ETH was taken for $0, and they still owed the DAI debt.

The Solvency Crisis

MakerDAO's surplus buffer — its reserve fund, accumulated from fees — was approximately $500,000 before the crash. The $6.65 million in bad debt wiped it out entirely and then some. For the first time, MakerDAO was insolvent: the total DAI in circulation exceeded the value of collateral backing it.

If this sounds abstract, consider the concrete implication: DAI was supposed to be worth $1, backed by overcollateralized crypto collateral. After Black Thursday, there was not enough collateral to back all outstanding DAI. The peg was at risk. If DAI holders lost confidence and tried to redeem en masse, the system could not honor all claims.

MakerDAO governance moved quickly. Within days, the community voted on an emergency response:

1. Debt auction: The protocol minted and auctioned new MKR governance tokens to raise DAI and cover the deficit. This diluted existing MKR holders — the governance token lost approximately 20% of its value — but it recapitalized the system.

2. System parameter changes: The liquidation penalty was increased, minimum bid requirements were added, and the auction mechanism was redesigned (eventually replaced by the "Liquidations 2.0" system, which used Dutch auctions with declining prices to ensure faster settlement).

3. Emergency shutdown was considered but rejected. MakerDAO has a nuclear option: emergency shutdown, which freezes the system and allows all participants to redeem their claims. This would have been the DeFi equivalent of a bank holiday. The community judged that the system could be saved without it.

The debt auction raised approximately $5.3 million in DAI through the sale of 20,980 newly minted MKR tokens. Combined with other recovery measures, MakerDAO was recapitalized within two weeks.

Root Cause Analysis

Why did this happen? The post-mortem identified several interlocking failures:

1. Auction Design Without Minimum Bids

The most direct cause. The auction system assumed competitive bidding. When competition disappeared (due to gas costs and network congestion), a single bidder could win at any price. The fix was straightforward — add minimum bids and redesign the auction mechanism — but the damage was already done.

2. Network Congestion as a Systemic Risk

MakerDAO's liquidation mechanism depended on Ethereum network availability. When the network became congested, the mechanism failed. This is a form of infrastructure risk that many DeFi protocols had not adequately considered. The protocol's security model assumed that liquidation bots would always be able to transact. March 12 proved that assumption wrong.

3. Liquidation Bot Centralization

Despite being "decentralized," the liquidation ecosystem was concentrated among a small number of sophisticated operators. When those operators pulled back (due to gas costs, infrastructure failures, or risk management), the ecosystem's diversity dropped to near zero. There was no fallback.

4. Speed of the Price Decline

The 50%+ crash in ETH happened in hours. The auction system had time delays built in (to allow bidders to participate). Those delays, designed for normal market conditions, were disastrously slow during a crash. By the time an auction concluded, the collateral had lost more value, making the bad debt worse.

5. Insufficient Surplus Buffer

$500,000 in reserves for a system with hundreds of millions in outstanding DAI was woefully inadequate. It could absorb a minor hiccup, not a systemic shock. After Black Thursday, MakerDAO governance significantly increased the surplus buffer target.

Aftermath and Lessons

For Protocol Design

Black Thursday rewrote the playbook for DeFi liquidation design:

• Dutch auctions (starting at a high price and declining) became standard, replacing English auctions (starting low and bidding up). Dutch auctions execute faster and do not require competitive bidding to reach a fair price.
• Minimum bid requirements became universal. No protocol now allows zero-bid liquidations.
• Keeper incentive redesign ensured that liquidation remained profitable even during extreme gas price environments (by increasing liquidation penalties or automating keeper gas management).
• Circuit breakers were discussed (and in some cases implemented) — mechanisms that pause liquidations during extreme volatility to prevent cascades.

For Risk Management

• Collateral ratios were reassessed. Many users had been operating at the minimum 150% — comfortable in normal markets, catastrophic in a crash. The lesson: overcollateralization ratios should be calibrated not for normal volatility but for tail events.
• The "worst-case scenario" was recalibrated. Before March 2020, many DeFi participants assumed a 30% crash was the worst case. After March 2020, 50%+ crashes in hours became the planning baseline.
• Multi-collateral diversification was accelerated. MakerDAO had already begun accepting USDC as collateral (ironic for a "decentralized" stablecoin, but pragmatically necessary). After Black Thursday, the push toward diverse collateral types intensified.

For the DeFi Ecosystem

Black Thursday was, paradoxically, both DeFi's darkest day and its proof of concept. The system nearly failed — but it did not fail. MakerDAO was recapitalized through a governance process, without any government bailout, central bank intervention, or emergency decree. A decentralized system experienced a crisis, diagnosed the problem, raised capital, and recovered — all through transparent, on-chain governance.

The total bad debt of $6.65 million, while devastating for affected vault owners, was small in absolute terms. No centralized entity made users whole — the losses were real and permanent. But the protocol survived, adapted, and became stronger. The Liquidations 2.0 mechanism that replaced the broken auction system was more robust, and MakerDAO went on to manage billions in TVL through subsequent market cycles.

Discussion Questions

1. Accountability: Several vault owners who lost their collateral in zero-bid auctions filed a class-action lawsuit against the Maker Foundation. They argued that the protocol's documentation described liquidations as resulting in a "13% penalty" — not total loss. Is a protocol development team liable for economic damages caused by a smart contract design flaw? How should liability work in decentralized systems?

2. Gas as an attack vector: Could a malicious actor intentionally congest the Ethereum network (by spamming high-gas transactions) to disable liquidation bots and profit from zero-bid auctions? Is this a practical attack, and how would you defend against it?

3. Emergency shutdown trade-offs: MakerDAO governance rejected emergency shutdown during Black Thursday. What would have happened if they had triggered it? Who would have benefited, and who would have been harmed? Under what conditions should emergency shutdown be used?

4. Moral hazard in recapitalization: MakerDAO minted new MKR tokens to cover the bad debt, diluting existing MKR holders. Did this create moral hazard — the expectation that MKR holders will always backstop the system? How does this compare to government bailouts of traditional banks?

5. Infrastructure dependency: MakerDAO's liquidation mechanism failed because Ethereum was congested. As DeFi grows, what infrastructure dependencies remain single points of failure? How can protocols design for infrastructure failures?

Timeline

Key Takeaways

1. Liquidation mechanisms are only as good as the infrastructure they run on. A perfectly designed auction system fails if bidders cannot participate due to network congestion.

2. Stress testing must include infrastructure failure scenarios, not just price scenarios. The question is not "what happens if ETH drops 50%" but "what happens if ETH drops 50% AND gas prices increase 10x AND half of liquidation bots go offline."

3. Minimum bid requirements are not optional. Any auction system that allows zero bids will eventually see zero bids.

4. Overcollateralization buffers must be sized for tail events, not normal volatility. A 150% collateral ratio provides no safety margin in a 50% crash.

5. Decentralized governance can respond to crises, but it operates on the timescale of days, not hours. Protocol design must be resilient enough to survive until governance can act.