Forward the Original Title‘Lessons from 2008’

Introduction

As ETH staking yields compress to ~3%, investors have turned to tokenized pools of restaking positions known as Liquid Restaking Tokens (LRTs) for increased ETH denominated Yield. As a result, value held in LRTs has skyrocketed to $10b. A large driver of this activity has been the ~$2.3B used as collateral to take on leverage. However this is not without risk. The individual positions that make up LRTs have hard-to-model, idiosyncratic risks and little on-chain liquidity exists to support healthy liquidations in the event of a large slashing event.

There are some meaningful parallels between the current LRT landscape and the events preceding 2008 worth exploring. In 2003, the federal funds rate dipped to 1% (the lowest it had been in 50 years). In search of greater dollar-denominated returns, investors flocked to the American housing market. At the center of the 2008 crash, laid excessive leverage and insufficient liquidity on Mortgage Backed Securities (MBSs), which, like LRTs, consisted of pooled positions (mortgages) with hard-to-model, idiosyncratic risks. When bad mortgage practices led to an increase in defaults, a self-reinforcing spiral of liquidations, panic and a lack-of-liquidity lead to devastating reductions in economic output felt around the world.

Given these similarities, it’s worthwhile to ask (and try to answer): what can we learn from the past?

A Brief History of 2008

(Note : there is obviously a lot more that isn’t written about here, but to keep the message on topic, I cherry picked some of the most relevant aspects for our story).

A simplified, over causal story of what caused the 2008 recession is as follows :

Originator and Securitizer Incentives

The increased demand for MBSs naturally incentivized mortgage lending to increase supply. As a consequence, the “originate and distribute” model became increasingly popular. This entailed mortgage lenders (originators) quickly offloading the risk of default to securitizers and then to traders looking for more yield (distribution). By offloading the risk, origination was much more scalable as they could quickly originate and sell mortgage debt without needing a massive balance sheet and robust risk management practices.

Herein lies our first principle-agent problem : since the mortgage originators didn’t need to hold onto the risk from any of the loans they made, they had a means and incentive to give out more mortgages with little risk. As an anecdote on where this incentive led, a class of mortgages emerged that were so bad, they were called designed-to-default loans.

Rating Agency Incentives

However, the mortgage originators and securitizers were not alone in propping up the new (seemingly stable) yield sources. In comes the rating agencies. Based on the construction of any given MBS, rating agencies were tasked with deciding which were pristine (AAA) and which were risky (B and lower). Their involvement accelerated the oncoming crisis in two key ways :

1. Rating agencies were paid by the institutions pooling and securitizing the mortgages. This conflict of interest created competition for lowering rating standards in order to secure business. For example, one rating agency (Fitch) whose analyses were less likely to award AAA ratings practically lost all MBS rating deal flow.
2. The prevailing risk models were incorrect. Notably, they assumed default risk between mortgages were more independent than they actually were. As a result, securitizer could tranche MBSs by risk (where the riskiest tiers would see the first X% of losses from defaults) to create Collateralized Debt Obligations (CDOs). The least risky tranches would be more likely awarded a AAA rating. And the most risky tranches could be re-pooled, re-tranched, and then re-rated. The top tranches of these new CDOs, often warranting new AAA ratings (Hint : the probabilities of defaults were not independent).

Excessive Leverage

In 1988, Basel I Capital Accord was approved which set capital requirements or internationally active banks. A capital requirement is how much capital a Bank is required to hold for every “risk-weighted” dollar of risky assets - in essence, it prescribes a maximum leverage ratio (12.5 : 1). If you’re familiar with crypto lending protocols, you can think of risk-weighted capital requirements as having a similar role as the Loan-to-Value ratio for different assets. However, the “risk weightings” were not always to minimize risk - they were often used to incentivize banks to achieve alternative objectives. To encourage banks to finance residential mortgages, mortgage related securities were weighted at 50% of the risk of commercial loans, which meant you could achieve twice the leverage (25 : 1). In 2007, Basel II lowered the risk weighting even more for AAA rated MBSs to allow for a 62.5 : 1 leverage to capital ratio (Note: less leverage was allowed for worse rated MBSs) (GAO Report on Mortgage Related Assets).

Despite the capital requirements, “ratings and regulatory arbitrage” in the form of Special Investment Vehicles (SIVs) provided a paths to circumventing even more leverage. An SIV is a minimal legal entities “sponsored” by a bank, but with entirely separate balance sheets. SIVs could borrow money at good rates to buy assets despite having little-to-no credit history because it was assumed that the “sponsor” bank would backstop any losses. The banks and the off-balance sheet SIVs were practically the same organizations.

For a long time, the sponsoring banks didn’t need to meet any capital requirements in relation to the SIVs’ debt. That was until Enron propped up stock prices (and subsequently crashed) after hiding debts in specially designed off-balance sheet vehicles. In light of this, regulators revisited the topic, but didn’t make significant changes - SIVs could continue to operate with 10% of the capital requirements that their sponsor Banks had in the U.S. Put in terms of a leverage ratio, Banks could still take on 625 : 1 leverage on AAA-rated MBSs through SIVs. (Note: keep in mind this does not mean banks maximized leverage or only held MBSs, just that they could).

Unsurprisingly, SIVs became the largest mechanism for funding mortgage holdings in the global financial system (Tooze 60).

Complexity Drives Opacity

Somewhere in here there is also a lesson on complexity. Finance isn’t simple - it largely revolves around some parties being better equipped to assess and take risks than others. A government bond might be relatively easy to assess in isolation. A single mortgage, less so but still reasonable. But how about a pool of mortgages that are evaluated on a set of complex assumptions? Or tranches of those mortgages using even more assumptions? Or how about a re-pooling and then re-tranching of tranches of pooled mortgages… a headache.

At some point in all of the pooling and tranching it becomes incredibly convenient to delegate risk assessment to the rest of “the market” and not due diligence what is actually held in these derivatives.

And the incentives for complexity in the derivatives market are strong, they favor sharps over the fish. When asked who would be buying their synthetic CDOs, financial engineer and Goldman Sach’s employee Fabrice “fabulous Fab” Tourre, responded “Belgian widows and orphans” (Blinder 78).

Simple… right ?

But the “Wall Street is Greedy!” narrative is reductionist. When all was said and done, the losses from AAA rated bonds issued from 2004 to 2007 (peak mania) weren’t even that large - only 17 bps by 2011 - and yet the global markets collapsed at an unprecedented scale. This points to it being unlikely that excessive leverage on bad collateral wasn’t the only culprit.

In Credit Crises, Gorten and Ordonez argue that when revealing information about collateral’s quality has a cost, even routine market movements can trigger recessions. The model shows that as a market progresses without serious shocks, lenders pay less to produce information. As a result, borrowers with poor quality collateral that is costly to rate enter the market (e.g., subprime MBS held in SIVs). The reduction in rating leads to increased output since lending becomes less costly and borrowers can cheaply source collateral. But when a small drop in value of some risky collateral is revealed, creditors can become incentivized to pay the costs to rate again. As a result, lenders begin to avoid collateral that is costly to rate, even if it’s good quality. This credit crunch, can lead to a significant reduction in output (Gorton and Ordonez).

Parallels Between MBSs and LRTs

The motivation behind demand for safe ETH yields in crypto (or at least Ethereum) resembles the demand for safe USD denominated yields in traditional finance. Just as it did for government- issued USD yields in 2003, the size of “government-issued” ETH yields (staking ETH) is being compressed, dropping to about 3% now that about 30% of the ETH supply is staked.

Not so differently than in 2008 with MBSs, staking yield compression has incentivized the market to look towards riskier venues for increased returns. This analogy is not new. Notably, in ‘What PoS and DeFi can learn from mortgage-backed securities’ by Alex Evans and Tarun Chitra, Liquid Staking Tokens (LSTs) are likened to MBSs. Among other points, their work discusses the usefulness of LSTs to overcome competition between staking yields (needed for security) and DeFi yields for capital by allowing stakers to access both. Since then LST holders’ preferred use in DeFi has been to borrow against them to increase their leverage.

However, the relationship between MBSs and Liquid Restaking Tokens (LRTs) appears to be even more uncanny.

Whereas LSTs like stETH pool together validators with relatively homogenous risk (due to validating the same relatively stable protocol), the restaking market is entirely different. A restaking protocol facilitates the pooling of stake for various actively validated services (AVSs) simultaneously. To incentivize deposits, these AVSs pay fees to stakers and operators. In contrast to vanilla ETH staking, the number of ETH restaking opportunities are uncapped - but as a consequence can also have idiosyncratic risks (e.g., unique slashing conditions).

As a result of the higher yield, the risk-seeking crypto market has flocked to deposit, with ~$14b in TVL as of writing this. Making up a significant portion of this growth (~$10b) are Liquid Restaking Tokens, which tokenize shares of a pool of restaking positions.

On one hand, you have vanilla ETH staking yields, which feel similar to being “government issued and backed”. For example, most stakers probably assume that in the event of a major consensus bug leading to mass slashing, Ethereum would hardfork.

On the other hand, you have restaking yields, which can come from literally anywhere. They can’t count on issuing ETH in-protocol to incentivize continued security. And in the event of an implementation bug for a custom slashing condition, Ethereum hard forking would be much more controversial. If the circumstance were dire enough, perhaps we’d be able to see whether the hardfork from the DAO hack created any moral hazard in enabling “too big to fail” protocols (a timely reference to the bailout of banks, which were assumed to be “too big to fail” else they would create systemic risk for the global financial system).

The incentives of LRT issuers and ETH restakers are not so far off from the incentives of mortgage securitizers and banks looking for higher yield from increasingly more leverage. As a result, we could see the crypto-equivalence of designed-to-default loans, not only become possible, but also prevalent (if they aren’t already). One particular type of designed-to-default loans were called NINJA loans because the borrowers had No Income and No Job or Assets. Such a phenomenon in restaking would look like low-quality AVSs (e.g., No Income, No Application(s) and No TVL) getting meaningful amounts of LRT collateral staked to them for the interim yield they provide through token inflation. As we’ll discuss in the following sections, there are some meaningful risks if this occurs at scale.

The Practical Risks

The most significant financial risk is the risk of a slashing event occurring and dropping the LRT’s value below the liquidation threshold for various credit protocols. Such an event would trigger liquidations for the LRTs, and risk having significant price impact on the price of correlated assets due to the unbonding and sale of the assets in the LRTs for more stable ones. If the initial wave of liquidations is large enough, it could cascade into liquidations for other assets.

There are two plausible ways I could see this actually happening :

1. Bugs in a newly implemented slashing condition. New protocols will have new slashing conditions. This means new opportunities for bugs that affect large swaths of operators. If “designed-to-default” AVSs become prevalent enough, this outcome feels fairly likely to occur at some point. That said, the magnitude of the slashing event also matters greatly. Currently, AAVE (which has >$2.2b of LRT collateral at the time of writing this) has a liquidation threshold of 95% for borrowing ETH against weETH (the most popular LRT) - this means the exploit would need to contribute to a slashing event of >5% of collateral to initiate the first wave of liquidations.
2. Socially engineered attacks. An attacker (either a protocol or operator) could convince various LRTs to deploy capital to them. Afterwards, they take out a large short position on the LRT (and perhaps ETH and other derivatives). Since the capital is not theirs, all they won’t have much at stake outside of reputation. If the builders or operators do not care about their social reputation (perhaps because they’re pseudonymous) and the gains from the short position and attack bounty are significant enough, they should be able to walk away with a considerable amount of money.

Of course all of these are only possible if slashing is enabled - which isn’t always the case. But until slashing is enabled, restaking’s benefits to a protocol’s economic security are minimal, so we should prepare for the setting for when slashing risk is on.

Avoiding Past Mistakes

So the big question remains… what can we learn from the past ?

Incentives are Important

The current competition between liquid restaking tokenizers primarily centers around providing the largest ETH denominated yields. Similar to the increase in demand for risky mortgage origination, we will see demand for risky AVSs - this is where much of the slashing (and liquidation) risk lies in my opinion. Risky assets alone are not much of a concern, but when used to take on excessive leverage without sufficient liquidity, they become one.

To restrict dangerous amounts of leverage, lending protocols set supply caps, which determine how much of a given asset the protocol can take on as collateral. The supply cap largely depends on how much liquidity is available. If there is little liquidity, then liquidators will have a harder time trading the liquidated collateral into stable coins.

Similar to Banks taking on excessive leverage to increase the notional value of their portfolio, there could be a meaningful incentive for lending protocols to cross the line on best practices to support more leverage. And while it’d be nice to think that the market would avoid this entirely, history (2008 and otherwise) tells us that people are prone to delegate (or outright ignore) due diligence when given the promise of profit and information revelation costs are high.

Learning from past mistakes (e.g., incentives of rating agencies) tells us that it would be useful to construct a non-bias third-party to help assess and coordinate the risks for different collateral types and lending protocols - especially LRTs and the protocols they secure. And using their risk assessments to make recommendations on safe, industry-wide liquidation thresholds and supply caps. The degree to which protocols deviate from these recommendations should be publicly available for monitoring. In an ideal world, this organization wouldn’t be funded by those who stand to benefit from risky-parameter setting but instead by those looking to make (or facilitate) informed decision making. Perhaps it is a crowd-sourced initiative, an Ethereum Foundation grant, or a for-profit “come-for-the-tool, stay-for-the-network” play that serves individual lenders and borrowers.

Largely funded by the Ethereum Foundation, L2Beat has done a great jobs stewarding a similar initiative for Layer 2s. So I have some hope that something like this could work for restaking - for example, Gauntlet (funded by the Eigenlayer Foundation) seems to have already started, although there is not yet information related to leverage. Even if built however, something like this is unlikely to prevent risk altogether, but at the least it would lower the cost of keeping market participants informed.

This also brings us to a second, related point…

Insufficient Models and Liquidity Shortages

We previously discussed how rating agencies and mortgage securitizers vastly overestimated the independence of default between mortgages. What we learned as a consequence is that housing prices falling in one area of the U.S. can massively influence housing prices, not just somewhere else in the U.S., but also around the whole world.

Why ?

Because there is a small group of large players which provide most of the liquidity for the world’s economic activity - and this small group also held MBSs. When a bad mortgage practices caused MBS prices to fall, the ability of these large players to supply liquidity to the market fell with it. Since assets needed to be sold into illiquid markets to repay loans, prices everywhere (mortgage-related or not) dropped too.

A similar overestimation of independence from “sharing” liquidity can arise unintentionally with the setting of parameters in lending protocols. Supply caps are set to ensure that collateral in the protocol can be liquidated without risking insolvency. However, liquidity is a shared resource that every credit protocol relies on to ensure solvency in the face of liquidations. If one protocol sets their supply caps around the existing liquidity at a snapshot in time, another bunch of protocols can come come along one-by-one and make their own supply cap decisions, rendering each previous assumption about the available liquidity inaccurate. As a result, lending protocols should avoid make decisions independently from one another (barring they don’t have preferential access to liquidity).

Unfortunately, if the liquidity is permissionless to access at any point to anyone, it will be difficult for protocols to safely set parameters. However, a solution to this uncertainty presents itself if you can give preferential access to liquidity in certain events. For example, the spot market for an asset used as collateral could have a hook whenever a swap is called that queries the lending protocol to check if a liquidation is possible. If a liquidation is pending, then the market could only allow sales to be triggered by a message call from the lending protocol itself. This functionality could allow lending protocols to set supply caps with more confidence through partnering with exchanges.

A Case Study ?

We might already have one case study to watch as the LRT market plays out.

AAVE currently has over $2.2b of weETH collateral supplied on-chain, yet according to Gauntlet’s dashboard, there is only $37m of on-chain liquidity on the exit path to wstETH, wETH, or rETH (this doesn’t even account for slippage or USDC exits, making the liquidity even worse in practice). As other credit protocols begin to accept weETH collateral (e.g., Spark currently having >$150m weETH TVL), the competition over a small amount of liquidity will heat up.

The liquidation threshold for ETH borrows against weETH is 95%, meaning a slashing event worth >5% of the LRT collateral should be enough to trigger a first wave of liquidations. As a result, hundreds of millions (to billions) of sell pressure would flood the market. This would almost surely cascade into sell pressure on wstETH and ETH as liquidators exit into USDC, risking a subsequent wave of liquidations for ETH and correlated assets.

But as mentioned before, so long as slashing is not on, there is little risk. As a result, deposits in AAVE and other credit protocols should be safe from slashing risk for the time being.

Key Differences

It wouldn’t be proper to write a whole post about parallels between LRTs and MBSs (and today in crypto and pre-2008), without also discussing some of the key differences. While this post communicates some of the similarities between MBSs and LRTs, they are certainly not identical.

One of the most important differences is the open, over-collateralized, algorithmic, and transparent nature of on-chain leverage v. banking (and shadow banking) leverage. The capital inefficiency of over-collateralization brings with it some meaningful advantages. For example, if a borrower defaults (and there is sufficient liquidity), the lender should always expect to be repaid - this is not the case for under-collateralized loans. Their open and algorithmic nature also opens up assets for liquidation immediately and for anyone to participate in. As a result, untrustworthy custodians and conniving counterparties are unable to take harmful actions like delaying liquidations, executing it below its worth and re-hypothecating collateral without consent.

Transparency is also a huge advantage. On-chain information about a protocol’s balances and the quality of collateral is verifiable by anyone. In the context of the previously discussed work by Gorten and Ordonez, we would say that DeFi operates in an environment where it is less-costly to evaluate the quality of collateral. As a result, the costs of revealing information about the quality of collateral should be lower, allowing the market to correct more cheaply and often. In practice this means that lending protocols and users have a richer set of readily available information to base key parameters choices on. However it’s worth noting that for restaking there are still less-objective, off-chain factors such as code quality and team background that are costly to produce information about.

One anecdotal sign is that it feels like post-BlockFi, Celsius and co. collapsing, a greater proportion of lend and borrow activity is happening on-chain. Notably, we’ve seen tremendous growth in AAVE and Morpho deposits but little-to-no reemergence of off-chain lending operations the size of previous cycles. However, getting concrete data on the current size of off-chain lending markets is not an easy task - meaning there is a chance there’s been significant, but under-publicized growth. Barring no direct lending protocol hacks, and all else held equal, it should be less fragile for leverage to be achieved on-chain for the aforementioned reasons.

As slashing risk increases across LRTs, we may get another first class opportunity to see the pros (and cons) of transparent, over-collateralized, open and algorithmic lending in serious actions.

And finally perhaps the biggest difference is that we don’t have a government to bail us out if something goes haywire. There is no government backstop for lenders or Keynesian Tokenomics. There is just the code, it’s state, and how that state changes. So wherever we can, let’s try not to make avoidable mistakes.

Last Thoughts and Thanks

Big thanks to ADCV, Sam Hart, Zion, Max Einhorn, Nick Cannon, Mallesh, Gwart and others for providing useful notes and discussions.

If you enjoyed reading this, feel free to leave a comment or send a DM to continue the conversation!

Citations

Blinder, Alan S. After the Music Stopped : The Financial Crisis, the Response, and the Work Ahead. Penguin Group, 2011.

Chitra, Tarun, and Alex Evans. “What PoS and DeFi Can Learn From Mortgage-backed Securities.” Medium, 14 Dec. 2021, medium.com/gauntlet-networks/what-pos-and-defi-can-learn-from-mortgage-backed-securities-3d60dc18ee51.

Gorton, Gary, and Guillermo Ordoñez. “Collateral Crises.” American Economic Review, Feb. 2011, bpb-us-w2.wpmucdn.com/web.sas.upenn.edu/dist/7/542/files/2019/11/CC.pdf.

History of the Basel Committee. 9 Oct. 2014, www.bis.org/bcbs/history.htm.

Kashyap, Anil K., et al. “An Analysis of the Impact of ‘Substantially Heightened’ Capital Requirements on Large Financial Institutions.” Scholars at Harvard, May 2010, scholar.harvard.edu/files/stein/files/impact_of_substantially_heightened.pdf.

Mortgage Related Assets : Capital Requirements Vary Depending on Type of Asset. GAO-17-93, United States Government Accountability Office, Dec. 2016, www.gao.gov/assets/gao-17-93.pdf.

Tooze, J. Adam. Crashed: How a Decade of Financial Crises Changed the World. 2018, ci.nii.ac.jp/ncid/BB29404013.

Disclaimer：

1. This article is reprinted from [Possibility Result]. Forward the Original Title‘Lessons from 2008’. All copyrights belong to the original author [Possibility Result]. If there are objections to this reprint, please contact the Gate Learn team, and they will handle it promptly.
2. Liability Disclaimer: The views and opinions expressed in this article are solely those of the author and do not constitute any investment advice.
3. Translations of the article into other languages are done by the Gate Learn team. Unless mentioned, copying, distributing, or plagiarizing the translated articles is prohibited.