March 12: The Day Crypto Market Structure Broke (Part 1)
Editor's Note: This is Part 1 of a two-part series. To read Part 2, about the solutions to the problems discussed below, click here.
There were two large downward moves in the crypto markets on March 12th, about 13 hours apart. The first was early in the morning, and the second in the evening (in the US). The first move down—a ~25% move—was fast and relatively orderly given the size of the move; however, during the second leg down, the market structure broke. Unfortunately, this led to Bitcoin crashing below $4,000 for a few minutes, marking the worst single-day price drop in seven years. Had the market structure remained intact, BTC likely would not have fallen below $4,000.
In part one of this essay, I’ll explain:
- What happened
- Why it happened
In part two, due later this week, I’ll explain:
- Potential solutions to the underlying problems
- Why a crash like this is unlikely to happen again in the near term
- Why a crash like this is likely to happen again in the medium term
The quick summary: the Bitcoin and Ethereum networks—in their current forms—cannot operate at global scale. During times of crisis, they become so congested that arbitrageurs cannot keep prices in line across venues, causing massive dislocations on individual exchanges. Massive dislocations on a single exchange (BitMEX) caused Bitcoin to dip below $4,000 for 15-30 minutes; however, this would not have happened if the market operated correctly.
Crypto Market Structure
To understand why the market structure broke, one must understand the current market structure. Crypto market structure is not like that of equities—where the overwhelming volume for most assets takes place on a single venue. Instead, crypto markets are more comparable to foreign exchange (FX) markets.
The unique traits of the crypto market structure are:
- There are a lot of trading venues. The major venues are: BitMEX, Binance, Huobi, OKEx, FTX, Deribit, Coinbase, Bitfinex, LMAX, and Kraken. There are also a long tail of smaller exchanges (many of which traded more than $1B on March 12) like Bybit, BitMax, BitForex, itBit, and Bitstamp. And then there is the DeFi ecosystem: Maker, Compound, Lendf.me, Synthetix, Uniswap, IDEX, dYdX, and more.
- Derivatives exchanges offer up to 125x leverage—that is instantly available—via perpetual swap contracts (aka ‘perps’). While very few participants use this much leverage, many trade with 10-50x leverage. Leverage enters and exits the system extremely quickly, and liquidations are common.
- Market mechanics are not uniform across venues. Each venue offers different kinds of products (spot, futures, perps, options); they accept different types of collateral for different products; liquidation methodology/parameters are unique; and liquidity for each product varies widely across exchanges. For example: 1)Binance dominates spot trading, is a top 5 player in derivatives, and offers perpetual contracts on a few dozen assets. Binance also supports a few different forms of collateral for BTC futures and perps, but only USDT for ETH. 2) BitMEX trades 8 assets via futures and perps, but only accepts BTC as collateral. 3) FTX offers the most diverse set of products (spot, futures, perps, options) and collateral options. 4) Deribit dominates options trading, and is growing in futures and perps.
- Traders cannot cross-margin positions across trading venues, and there are not yet any prime brokers with enough capital across all the venues to provide this service. This adversely impacts capital efficiency, and therefore increases cost of capital throughout the ecosystem. Moreover, most of the exchanges don’t yet net out long positions and short positions, compounding this problem even further.
- Some market participants denominate their wealth in different currencies, and therefore think about risk and trading differently than one would expect. While most denominate in USD, some denominate in BTC, and some in ETH.
- In traditional markets, most liquid assets are accepted as collateral (e.g. most stocks that trade on the NYSE). However, in crypto, lenders generally only accept USD, BTC, and ETH as collateral. This creates the opportunity for basis risk, as borrowers can be solvent but unable to meet collateral requirements without liquidating positions.
- Exchanges do not credit accounts when users deposit funds instantly. While policies vary by exchange and by asset, it generally takes at least 10 minutes after a block containing the deposit transaction is confirmed, and can take up to 60 minutes. During periods of high activity such as March 12th, a trader may have to wait several blocks to get their transaction confirmed.
- Virtually all price discovery happens on traditional exchange venues as opposed to using DeFi protocols, with only a few exceptions. As a result, DeFi prices lag centralized venues. Arbitrageurs generate meaningful profits keeping these venues in-line.
- Many traders only use one, or a small number of exchanges. For example: 1) Some US-based funds only trade on Coinbase, Kraken, CME, and Bakkt. 2) Most Chinese retail investors trade primarily on Huobi. 3) Some people refuse to KYC, and therefore only trade on BitMEX. 4) Many traders are not sophisticated enough, or just don’t care enough to try to get best execution by tapping into liquidity across the various venues.
Given all of the above, it’s clear why prices deviate between trading venues. Normally, prices don’t deviate significantly beyond maker/taker fees and a small spread (often just a few basis points). When they do, liquidity providers and arbitrageurs usually have enough capital at the major venues to profit from the arbitrage opportunities and collapse the spread.
But when volatility picks up, a few things happen concurrently: 1. Liquidations accelerate. While some traders are levered up to 125x, most on-exchange leverage is 25-30x. For context with 25x leverage, a ~3% move will liquidate a position because the exchange has to include a buffer to account for potential slippage during a liquidation. Otherwise, the trader has to post more collateral. 2. Liquidation thresholds are not uniform across venues, so prices deviate between exchanges as liquidations cascade. 3. Arbitrageurs don’t have enough capital on each venue to collapse the arbitrage, so they begin shuttling assets—primarily BTC, ETH, and USDT, and to a lesser extent USDC— between the exchanges to arbitrage the price discrepancies.
When this happens, demand for blockspace on Bitcoin and gas on Ethereum explodes upwards within minutes. Gas prices skyrocket, and many transactions don’t get included in a block for minutes, or even hours.
When this happens as prices collapse, miners start turning off their machines because mining revenues (which are denominated in crypto) fall below the cost of electricity. This in turn slows the rate at which new blocks are produced, which increases latency, and decreases aggregate throughput.
What Happened On March 12
The first leg down was most likely caused by traders de-risking as global equity markets were selling off. The second leg down was likely triggered by a lender that was liquidating collateral. The collateral had belonged to borrowers who became insolvent as a result of the first leg down. Some miners shut down their rigs after the first leg down. A lot more did during the second leg down.
And then the market structure broke.
Because BitMEX only accepts BTC as collateral, all the BTC-perp longs on BitMEX are by definition forced to take on leverage. To understand why this is the case, consider this: If a trader goes long BTC-USD using perps on BitMEX, as the value of BTC drops, the trader loses money on the trade, and the value of the BTC collateral decreases. This intrinsically creates risk for liquidity providers on BitMEX. When the market moves more than 30% in a day, even traders with relatively mild amounts of leverage start getting liquidated. Market makers understand this, and thus provide less liquidity than they otherwise would, accelerating the downward cascade.
As spot prices fell as a result of collateral liquidations, derivatives followed. BitMEX started liquidating levered longs. Those liquidations started cascading. Given the magnitude of the first leg down, many market makers simply stopped providing liquidity until the market settled down. With wild spreads—which at times exceeded $500—between BitMEX and Coinbase, liquidity providers did not want to go margin long on BitMEX to try to stop a series of cascading liquidations.
Moreover, the Bitcoin blockchain was congested. This was compounded by the fact that block production slowed as miners turned their machines off. As the prices between exchanges deviated, arbitrageurs literally could not deposit BTC on BitMEX to bring prices in line even if they wanted to try to catch the falling knife.
At one point, there were only ~$20M of bids left on the entire BitMEX order book and over $200M of long positions to liquidate. This means the price of BTC could have briefly crashed to $0 had BitMEX not gone down for “maintenance.” Given BitMEX’s central position in the crypto market structure, this price move could have propagated through to all the other BTC trading venues.
While BitMEX broke (partially because arbitrageurs couldn’t send capital to BitMEX, and partially because of BitMEX’s incompetence), DeFi outright failed.
Maker is the largest DeFi protocol, and the foundation on top of which much of the rest of DeFi is built. Maker nearly imploded, and it could have if a few things did happen that were supposed to. Maker experienced two major failures, only one of which appears to be receiving public attention. As a result of the first leg down, some Maker vaults (formerly known as CDPs) became under collateralized. Keepers—who run the open source Keeper software written by the Maker team—tried to liquidate the under-collateralized vaults. But the Keeper software was not configured to dynamically adjust gas prices in response to network congestion, so miners were not including Keeper liquidation transactions in blocks.
Therefore, as far as the Maker protocol was concerned, no one was bidding in the collateral auctions.
Someone realized that by just increasing gas prices, she would be the only auction participant, and could therefore bid as low as $0. So she bid $0 for $8M of ETH collateral and got it.
The owners of those vaults have no legal recourse. Ouch.
Additionally, because the Keeper didn’t contribute back to the collateral pool, the Maker system became under-capitalized, though it was still over-collateralized. This means that Maker had enough ETH and BAT collateral to redeem all outstanding DAI, but was below the target risk thresholds. But Maker’s second failure was far more dangerous, and is far less discussed.
Like the Keeper software, the software that powers the Maker oracles weren’t configured to run in a highly-congested network either. So many of the oracles just stopped sending prices to the Maker contracts.
MKR holders vote on risk parameters of the system, and oracles enforce those rules by reporting prices. Oracles, by failing to report prices correctly, forced MKR holders to take on undue risk.
As ETH dipped down to ~$88 around 7pm central time on the 12th, a ton of vaults should have been liquidated because a lot of vaults were set to be liquidated at around $100 (traders intentionally created vaults with $100 liquidation threshold, knowing that it’s a powerful psychological number that other traders are likely to support). But the Maker oracles just didn’t update the ETH-USD price at all as ETH-USD dipped below $100.
It’s unclear if the oracles didn’t raise their gas fees, or if they deliberately chose not to update prices in an attempt to save the system. Had the oracles reported prices correctly, all of those vaults would have been liquidated, which would have caused the ETH price to cascade down even further. As ETH dipped below $100, many market makers simply stopped providing liquidity since the market was going haywire. A few million dollars of market selling could have driven ETH down to $50 or even lower.
This failure could have cascaded much further:
- Maker could have become insolvent, which could have caused DAI to fall below the peg.
- DAI is used as collateral for loans in other DeFi protocols like Compound and Lendf.me. If the price of DAI fell, those borrowers could have gotten liquidated.
- In addition to DAI, many ERC-20 tokens are posted as collateral on protocols like Compound and Lendf.me. ERC-20s are generally correlated with ETH, and as ETH fell, they likely would have fallen as well. This would have caused even more liquidations in other lending protocols, cascading through the long tail of DeFi.
- The price of MKR could have collapsed to the point such that the upcoming MKR auction would fail to adequately recapitalize the system.
Today, DeFi activity is about 1% of that in crypto CeFi activity. And crypto is a rounding error in global markets. Just imagine how much worse this situation would have if DeFi had more capital, more people, and more trades going through.
While we are excited about the potential of DeFi as we wrote about in our Crypto Mega Theses, DeFi has a long way to go. This is both a sobering reminder of how early we are, and an indication of how many more investable opportunities there will be.
The biggest take away from this is that crypto market infrastructure is still immature. There is a lot of room to improve along many dimensions, and therefore a lot of investable opportunities.
While venues like BitMEX clearly struggled, they can improve their liquidation engine among other technical matters. However, even if BitMEX were to operate flawlessly on a go-forward basis, the market still would have struggled. The Bitcoin and Ethereum networks as they stand today simply cannot support global-scale capital markets activity.
In part two of this essay, I’ll discuss potential remedies.
Disclosure: Multicoin Capital holds BTC, ETH, DAI, USDT, and USDC tokens.