This time isn’t different

I think I’ve seen this film before, and I didn’t like the ending

Trevor Chow

🛑 Epistemic Warning: I know precious little about tradfi, and certainly even less about defi!

June 20, 2007.

That’s the day Bear Stearns bailed out two of its hedge funds. It marked the first sign that the subprime mortgage crisis had spread beyond just the housing markets, past the lenders and securitisers, all the way to one of the big Wall Street investment banks.

It would take 9 more months before Bear Stearns faced bankruptcy, and another year after that before the Dow Jones hit its lowest point in March 6, 2009. However gloomy financial markets might have looked in mid-2007, they got a lot worse before they got better.

It’s been exactly 15 years since that day, and storm clouds are now gathering in the world of cryptocurrencies and decentralised finance. In that spirit, let’s revisit the Global Financial Crisis, and see if defi has anything learn from the most severe crisis in tradfi since 1929.

Safety via diversification

The Global Financial Crisis marked the end of the Great Moderation. During the preceding three decades, there was an unparalleled bull market and a 4x rise in housing prices. Across the board, people were irrationally exuberant and assumed that housing could only appreciate. Thus banks began offering subprime mortgages and lending to people with bad credit, on the assumption that in the worst case, you could just take the house as collateral. Since it would be worth more, the risk seemed pretty small.

Simultaneously, there was a global savings glut as investors looked for places to put their money. To meet this demand, banks decided to manufacture more safe assets for investors to buy, packaging mortgages into mortgage-backed securities. These were financial instruments which provided payments depending on the underlying mortgages. Investors might not have been willing to buy individual mortgages, but MBSs seemed like a much more attractive proposition. Instead of needing to worry about the risk of any particular mortgage, they could just buy a pool of many different mortgages. One homeowner might default, but when in the world would everyone default at the same time? It was “diversified” - in other words, it was safe.

How about the rare cases where the mortgages were of such poor quality that even MBSs didn’t seem safe? Well, banks would just repackage them into a collateralised debt obligation, bundling enough together till even a pool of bad MBSs seemed “diversified” enough. It wasn’t just mortgages, however - other loans were being packaged into asset-backed securities. For example, car companies would make loans to help people finance their car purchases, and bundle these loans into ABSs, which were in turn put into CDOs too.

Once these ABSs and CDOs were seen as safe assets, they were used everywhere, beyond just the balance sheets of investment banks. To fund their operations, ordinary companies often issued asset-backed commercial paper, which were essentially short-term loans. These ABCPs were backed by the ABSs and CDOs. For example, companies like General Electric would use the CDOs derived from their car loans to collateralise these ABCPs, which they used to pay for day-to-day costs.

Money market mutual funds, which aimed to keep the value of every dollar of investment at a dollar, bought up these ABCPs in droves, since they were backed by seemingly risk-free assets. Even insurance companies like AIG got a piece of the action, offering insurance on ABSs and CDOs for investment banks which wanted to offload their risk.

Everywhere you looked in the financial system, there was a massive web of obligations connected by these so-called “safe assets”.

Housing theory of everything

Then all of a sudden, the housing bubble popped.

The immediate effect was that subprime mortgages fell in value. That meant real estate investors got screwed. For example, New Century was the second largest subprime mortgage issuer in the world. With their assets disappearing in value, they filed for Chapter 11 bankruptcy on April 2, 2007.

Soon enough, investment banks which owned ABSs and CDOs saw their balance sheets plunge into the red. Bear Stearns would liquidate their aforementioned hedge funds in July, due to their exposure to these assets. By August, BNP Paribas considered it impossible to even value these financial instruments, preventing any investor withdrawals from its own hedge funds. In September, Northern Rock, a British bank which had bought up huge swathes of these assets, faced a run from depositors. By early 2008, Northern Rock was nationalised by the British government, while Bear Stearns was bought out by JPMorgan Chase alongside the Federal Reserve.

Meanwhile, the government-sponsored enterprises of Fannie Mae and Freddie Mac began to buy up subprime mortgages from the banks, in an attempt to take some of the risk off their books. None of this was enough to stem the bleeding, and by September 7, Fannie and Freddie themselves had to be taken over by the Federal Reserve. On September 15, Lehman Brothers went bankrupt and Merrill Lynch was purchased by Bank of America. Just like that, three of the five largest investment banks (Bear Stearns, Lehman Brothers and Merrill Lynch) had disappeared.

The next day, the huge insurance payouts from ABSs and CDOs collapsing meant AIG was taken over by the Fed too, while the Reserve Primary Fund, the oldest MMMF, lost the one-to-one parity which every MMMF is expected to maintain. In other words, it had “broken the buck” due to its exposure to ABCPs which were becoming as worthless as their underlying assets. Unsurprisingly, investors everywhere pulled $144 billion out of US MMMFs the following day, causing companies to face issues in covering their costs as they struggled to roll over their ABCP.

With the interbank lending market, the ABCP lending market and the MMMF lending market under pressure, there was only one remaining source of short-term funding: the secured repo lending market. Unlike the other three, borrowing on the repo market required handing over collateral. But as investors sold their other assets to compensate for ABS and CDO losses, even those who weren’t exposed to those assets saw their balance sheets plumment and their collateral lose value, putting pressure on the repo market.

If it was clear whose balance sheets were genuinely worthless and insolvent and who was simply temporarily illiquid, perhaps they could have been isolated and contained. But with the fog of war making it impossible to figure that out, the costs of borrowing skyrocketed across the board. Banks stopped lending and credit dried up, bringing the real economy to a standstill.

Thus the housing crisis became a global financial crisis.

Fiscal theory of the price level

That’s the story of the GFC.

I’m now going to tell a different story. At least, it starts as a different story. Whether or not it ends that way? We’ll see.

We begin with Terra, which was a cryptocurrency ecosystem with two key securities: Luna and TerraUSD (UST). Luna was the cryptocurrency for transactions made on the Terra protocol. You can think of it as equity with no fixed value: its price reflected the extent to which people were building useful things on Terra’s blockchain.

Meanwhile, UST was a stablecoin, designed to trade at par with the US dollar. You can think of it as debt: a fixed income asset where you expect to be able to sell it for the amount you paid plus interest on top. Unlike many other stablecoins which had fiat assets backing its value, UST was an algorithmic stablecoin. That meant its peg to the US dollar was defended by an algorithm, which would ensure 1 UST could always be traded for $1 worth of Luna. If UST was worth too much, the Terra protocol would induce inflation selling UST to the market, and induce deflation by selling Luna if it was worth too little.

Algorithmic stablecoins can seem weird, so let’s map it to a more obvious concept from traditional finance.

Consider the US dollar.

One way of thinking about the US dollar is to conceive of it as an algorithmic stablecoin of sorts. Much like UST, it is an asset where you expect to be able to redeem it for the value you paid plus interest, though here the interest rate is negative thanks to inflation. Much like UST, its purchasing power is meant to follow a stable path, though here it is meant to fall by 2% a year. Much like UST, inflationary policy involves issuing more of the stablecoin via monetary expansion. Much like UST, deflationary policy involves swapping the market’s stablecoins for an asset which represents a share of the economic output of the system i.e. government surpluses or deficits. The only difference is that with the US dollar, this can be achieved forcibly, by raising taxes to soak up US dollars, rather than relying on consensual market transactions.

This should make you wonder: if the Terra protocol can’t demand taxes, what defends the value of Luna?

Mundell-Fleming trilemma

The next tradfi analogy is how countries establish and defend currency pegs.

The Mundell-Fleming trilemma tells us that a country can only pick two out of the following three: free capital mobility, fixed exchange rates and central bank independence. When places like Hong Kong decide to peg their currency to the US dollar, they do two things. Firstly, they ensure they have a large foreign exchange reserve which they can use to defend the peg. Secondly, they accept that if they want free capital mobility, they’ve outsourced monetary policy to the central bank of whatever currency they’re pegged to e.g. the Federal Reserve.

Terra did things a bit differently. They did have a version of forex reserves, the Luna Foundation Guard, which held billions worth of other cryptocurrencies. However, they tried to have their cake and eat it too, refusing to accept the trilemma. In order to lure people into holding UST, they set up the Anchor protocol, which paid sky high yields if people bought UST and lent it to Anchor. In other words, they wanted UST to have a fixed exchange rate with the US dollar, refused to accept any barriers to capital flow and wanted to set their own interest rates.

What happened when they started to lower the unsustainably high interest rate paid on Anchor? Almost immediately, there was a huge exodus of UST deposits. There are two ways to sell UST. The first is to rely on the stablecoin’s protocol of giving you Luna in return for UST. The second is to sell it for other stablecoins on a decentralised exchange (DEX) known as Curve, which offers pools of liquidity between various pairs of cryptocurrencies.

So when people started heading for the doors, the first mechanism meant that the supply of Luna massively increased. This meant Luna’s price fell, incentivising people to sell Luna for something else. In turn, even more people tried to get out of UST and Luna, causing a death spiral. After all, there was nothing keeping up the value of Luna.

As for the second mechanism, the flood of sales meant that there was more UST than other stablecoins on the exchange. To rebalance this, the exchange began to offer UST at a discount, but no one was biting. With the imbalance growing, Curve had to keep on offering steeper and steeper discounts. Thus both of these mechanisms destabilised the peg.

It is at this point that LFG deployed its billions worth of cryptocurrency reserves. It is possible to defend pegs this way. During the 1997 Asian Financial Crisis, the Hong Kong Monetary Authority bought up HK dollars by selling foreign currencies, fending off pressure from George Soros and other speculators who were trying to push down the price of the HK dollar. Rather than mirroring the HKMA however, the LFG turned out to be closer to the Bank of Thailand, which had to accept the depegging of the Thai baht from the US dollar in 1997, after exhausting all of its forex.

It’s yield farming all the way down

While LFG’s defense wasn’t able to stop the UST from collapsing to nothing, what it was able to do was push down the prices of the many cryptocurrencies it was selling. Of course, it wasn’t just them either: everyone else who was overexposed to the collapse of UST and Luna had started to sell their other cryptocurrencies.

In particular, there was another asset which got pushed off its “peg”. This is staked ether (stETH). Ethereum is currently preparing for an upgrade to Ethereum 2.0, and as part of that, people can stake its token, ether (ETH). This means locking up the token on the yet-to-be-unveiled Ethereum 2.0 network, such that it cannot be spent or sold until the new network comes online at which point each ETH staked can be redeemed 1:1 for ETH. Of course, they get cryptocurrency rewards in return.

For those who want to maintain liquidity, Lido Finance protocol offers a token known as staked ether. In return for giving Lido one ETH, you get back one stETH. This is a token, so it can be used like any other cryptocurrency. One popular trade has been to deposit ETH with Lido, get stETH, use it as collateral to borrow ETH on a lending protocol like Aave, and then rinse and repeat.

This recursive yield farming can seem pretty safe. Most of the time, stETH trades 1:1 for ETH. However, this isn’t an actual peg, and until Ethereum 2.0 arrives, there’s no reason why it should be exactly 1:1. In particular, once you hand over your ETH, Lido will have staked it for you, so there are no takebacks. The only way to unwind the trade is by selling stETH on secondary markets, and if there is more demand for liquidity in the market or if people believe that the new network is being delayed, it is quite likely to trade below parity.

The UST depeg led to stETH dropping below parity for two reasons. Firstly, the broad crypto downturn and asset firesales meant liquidity in the market was drying up in general. Secondly, many people had been engaging in another yield farming trade. This involved an asset called bonded ETH (bETH), which was just a representation of stETH on the Terra blockchain. By exchanging ETH for stETH with Lido and then converting that into bETH, they could lend it and earn yield on the Anchor protocol. This had led to a significant demand for stETH. When UST collapsed, people pulled out their money, converting it back and raising the supply of stETH.

What Agatha tells Anderton

The chaos in UST and stETH began to cause systemic contagion, hitting key financial institutions in crypto.

One of the biggest dominoes to fall was Celsius.

Celsius is one of the largest centralised cryptocurrency financial service providers. You can trade crypto with it, you can deposit crypto with it to yield interest and you can borrow from it. Although it tends not to describe itself as such, you can think of its business model as a bank. It has a retail-facing commercial banking arm and a retail-and-wholesale-facing investment banking arm, borrowing in the former in order to lend in the latter.

Unlike most banks in tradfi, which earned the difference between what it lent at and what it borrowed at, Celsius stood out by offering wildly above-market yields to depositors as well as the low interest loans to borrowers. How could it possible provide both of these simultaneously?

The trivial answer is that they were making a loss, and returns were partly being paid for by getting in new depositors. however, the more complete answer is rehypothecation and leverage.

Like basically every financial institution, Celsius lends out its depositor’s money. What rehypothecation refers to is that Celsius also lends out the collateral it gets from borrowers. In particular, it had been lending UST on Anchor, yield farming ETH/stETH with Lido and Aave as well as doing the stETH/bETH with Lido and Anchor. This meant that it was exposed to both UST and stETH.

So when UST depegged, they took quite a hit. Alone, this likely wouldn’t have taken down Celsius. However, it had also lost a few hundred million in various hacking incidents during the previous few months, and worst of all, the depositors were starting to withdraw their money in a panic.

This coincided with the stETH break in parity, which caused a huge maturity mismatch. Celsius’s ETH-denominated liabilities had to be paid out immediately, but its ETH-denominated assets were incredibly illiquid. Mostly they were either in stETH or directly locked up in Ethereum 2.0. The problem was that everyone was deleveraging at the same time, so the secondary markets for selling stETH were drying up.

Centralised exchanges (CEX) didn’t have deep enough liquidity, and neither did DEXs like Curve. Liquidity providers (LP) didn’t want to be left trapped and holding a bag full of stETH, so they exited; the stETH/ETH trading pool on Curve had become 80% stETH, meaning there wasn’t enough ETH for Celsius to sell all of its stETH, even if it was willing to do so at a loss. The only way out was with over-the-counter (OTC) trades to brokers, but that meant accepting an even steeper discount. All of this would cause the value of Celsius’s ETH-denominated assets to fall in value when marked to market price, encouraging even more withdrawals.

In a manner not too dissimilar to tradfi bank runs, Celsius got fucked.

Liquor, ladies and leverage

Oh, and then there’s leverage.

In order to further juice up its returns, Celsius borrowed from the MakerDAO protocol. MakerDAO, much like Terra, has two crucial currencies: the equity token Maker (MKR) and the stablecoin DAI which is pegged to the US dollar. Unlike UST, DAI is an overcollateralised crypto-backed stablecoin. This means that for every $1 of DAI Celsius borrowed, it needed to put in $1.50 worth of collateral into the MakerDAO lending protocol, and if the loan-to-value ratio ever went past a certain threshold, it would get seized.

The problem with getting leverage via an overcollateralised stablecoin is that they are natural liquidity drainers. As soon as the UST contagion led to selloffs in the rest of the crypto market, the value of Celsius’s collateral began to get marked down, bringing it closer to liquidation. In the midst of a massive cash exodus, Celsius had to deploy what limited resources it had to stopping a default on its loans.

Doubly fucked.

As Celsius took blow after blow, traders could smell blood in the water. If they sold enough of the asset types which Celsius was using as collateral, they could force Celsius to be liquidated. This would let them buy up Celsius’s stuff at bargain bin prices, and with all of this being publicly viewable on the blockchain, short sellers knew exactly what to target and how much to sell.

Triply fucked.

The result? Celsius gated withdrawals, preventing depositors from taking their money out. The ensuing fear, uncertainty and doubt sent asset prices tumbling, leaving it with even less collateral. At this point, Celsius could have tried to repay its loans, or top up its collateral to prevent liquidation.

For every $1 borrowed, Celsius needed to provide $1 to repay it or $1.50 to have enough collateral. Ostensibly, the former is cheaper. Yet Celsius chose the latter. Why? Because it knew it definitely couldn’t repay its debts, but it would cost pennies on the dollar to top up enough collateral to survive a bit longer. Thus what little remained of Celsius’s liquid ETH was locked in as collateral.

Triply fucked.

In short, Celsius took their depositor’s money as well as their borrower’s collateral, levered it up and bet it all on black at the roulette table. When UST and stETH got depegged and when other assets fell alongside, this house of cards began to collapse. Rather than honouring their customer’s redemptions, Celsius locked up their money. Then they doubled down on their bets and took one last punt, hoping on the off chance that their assets rebounded in price and they could make it all back.

In tradfi, you’d at least get your money back via laws surrounding deposit insurance, but here, none of this applies. Depositors and borrowers have no protection against Celsius using their money and collateral for whatever it deems fit. In the wild west of crypto, there is only one law: if it’s not your keys, it’s not your money. Because of how large Celsius was and how many counterparties it engaged with, this triggered chaos across a whole web of transactions.

Hedge funds that don’t hedge

It wasn’t just the shadow banks that were facing funding pressures. Another victim of this vicious cycle is Three Arrows Capital (3AC), one of the largest hedge funds in the crypto space. Much like Celsius, it was exposed to UST and stETH, thus leaving it in a bit of a bind as the losses cascaded and its collateral went to zero. In fact, it was even more exposed, because it had been one of the main organisations helping to support LFG, buying up Luna in exchange for various other assets in order to capitalise Terra’s crypto reserve. That half a billion dollars went up in smoke virtually overnight.

At this point, the story should be clear. They were levered up too much and got margin called when their collateral fell in value. The big difference is that they basically ghosted everyone, and thus got liquidated by their counterparties. These were some huge counterparties, including centralised exchanges like Bitmex, FTX and Deribit; lenders like Celsius, BlockFi and Genesis; market makers and trading firms like 8BlocksCapital, as well as brokers who had engaged in OTC trades with them.

This is especially concerning for counterparty lenders if they don’t have enough capital buffers on their balance sheets to bear the losses of liquidating 3ACs portfolio. For example, 83% of BlockFi’s loans are under- or un-collateralised. Nor was 3AC just a trading partner; in many cases, it was an investor which also managed assets for its venture bets. This provided another avenue for contagion, with lending platforms like Finblox which had received capital from 3AC having to impose withdrawal limits of their own.

The end of the beginning

As for what happens now? I have no clue.

If the collapse of 3AC we’ve seen in the past couple of days is crypto’s Lehman Brothers, it’s worth remembering that the immediate aftermath of Lehman’s collapse was seen as a triumph of the accountability of free markets against moral hazard. But within a matter of days, people realised this was no controlled demolition. The worst was not over - far from it.

Lehman didn’t mark the end of the crisis back in 2008. It wasn’t even the beginning of the end. At best, it was the end of the beginning. Nor is it over now. It’s not clear what topples next, if anything. Perhaps another stablecoin, like Tether or Magic Internet Money. (Yes, it’s really called that.) Perhaps a financial service provider like Nexo or Babel Finance, the latter of which halted withdrawals much like Celsius. Perhaps one of the many exchanges which has had to liquidate counterparty positions at a loss.

Equally, perhaps we get a knight in shining armour who comes and pumps enough liquidity into the system to calm everyone down. The world of defi might not have a lender of last resort, but that was once the case for tradfi too. Back in the Panic of 1893 and 1907, there was no Federal Reserve. Instead, JP Morgan personally bailed out the US financial system. Twice!

One promising candidate for this is FTX. In part, it is large enough to plausibly have aligned incentives to monitor risk carefully: for it to make money, it needs to have solvent counterparties and it needs to itself be solvent. More than that, as a CEX, it is less affected by sharp market downturns than lenders. This is coupled with its friendly relationship with the market maker Alameda Research, which could help provide emergency liquidity given how well market makers tend to do in times of volatility.

Mamma Mia! Here we go again

By now, the two stories should sound familiar.

  1. Irrational exuberance and a savings glut fuels a demand for safe assets.
  2. Financial institutions restructuring and repackaging new yield-bearing securities until they are treated as safe.
  3. Market actors act as shadow banks, levering up till the system is so fragile a single huff from the Big Bad Wolf can bring it down.
  4. When that arrives, there’s a run on every finnancial institution that even has a whiff of this no-longer-as-safe asset.
  5. This leads to margin calls, liquidations, panic over counterparty risk and contagion, till credit trickles entirely to a close.

In 2008, we learnt a few things after all of this.

Firstly, that the financial system is one that transforms risk, but doesn’t destroy it. By creating safe assets which are information insensitive, it can help reduce idiosyncratic risks. Yet safe assets are at the heart of systemic risk, and even seemingly safe assets like money market mutual funds can go bust.

Secondly, that you shouldn’t mistake leverage for genius. Too often, leverage is used because of greed or moral hazard, rather than because there’s a genuine opportunity.

Thirdly, that counterparty risks matter. When something goes wrong, the opacity of the system determines how badly everything else is hurt. if you have a large shadow banking system, you better make sure they hold enough liquid assets.

The greatest trick

The greatest trick the centralised defi institutions ever pulled was convincing the world they weren’t tradfi. Indeed, the very term itself is a misnomer. The trend of centralisation is unsurprising. In any network, there is a tendency towards having easy-to-use onramps and offramps, especially since decentralisation is naturally subject to adverse selection and a lack of recourse. Can defi be the exception to the rule?

I think this crisis has shown a glimpse of that potential. While some major cedefi institutions might have done badly this time around, the defi protocols and critical infrastructure have weathered the storm so far. The transparency of having everything on-chain helped sound the alarm, which is why more people knew about Terra than Celsius, and more people knew about Celsius than 3AC.

As long as there are transactions off the blockchain, there will be things you cannot audit. With those known unknowns, panic and contagion is almost inevitable. But that’s the point of the entire project: to ensure trustless and transparent transactions via the blockchain and smart contracts. If everything is on-chain, you can prove, beyond a shadow of doubt, that you are or aren’t exposed to the risks people are concerned about.

Of course there will always be some centralisation in crypto. Among tradfi advocates, there are those who think that this reflects the utter folly of the defi project. That once you’ve opened Pandora’s box to fraud, deception and exploits, you can’t close it again. That the increased transparency and tradability means increased volatility. Or that without a central bank, the interconnectedness of defi means it will always be on the verge of teetering. I disagree. When when all is said and done, when everything is over and when the worst has come to pass, there’s still one thing left in Pandora’s Box: hope!