DeFi’s Reckoning with Concentration Risk
When small moves make big differences
“How did you go bankrupt?” Bill asked.
“Two ways,” Mike said. “Gradually and then suddenly.”
This laconic bit from Hemmingway captures the recent anxiety in Ethereum-based “decentralized finance.” The tight-knit sector of on-chain financial services has gotten itself into a fragile predicament where a small push in the wrong direction would send key institutions into a tail spin. If prices on a select few assets move down, the domino-effect of defaulting borrowers could cause large-scale destabilization of the markets.
The gist of the situation is that over $110M of debt on the borrow/lend company Aave is backed by an illiquid asset CRV (the self-described “bribe” token of the Curve.fi exchange). Curve is a big deal in DeFi, processing around 10% of total daily volume for on-chain trades. The trouble is twofold:
- Supply. Over 50% of the circulating supply of the CRV coin is pledged as collateral, securing outstanding loans.
- Demand. What remains of the supply only sees an average of $30M in daily trade volume.
If CRV depreciates, the debt left on Aave (among other borrow/lend venues) will be subject to forced close-out, which results in the CRV collateral getting sold on the open market. But the market has neither the volume nor depth to absorb the flood of over $100M in CRV, and so the forced selling will cause the price to drop further, which will make subsequent liquidations less profitable, which will cause Aave to be saddled with massive bad debt, which will cause lenders to withdraw liquidity unless Aave can fill the shortfall with their reserves, which will deplete said reserves, which will cause an even greater fear among lenders, which will . . . you get the picture.
Claims have been made that this fiasco is a symptom of tectonic, structural problems within the DeFi ecosystem. The reports on DeFi’s death have been greatly exaggerated, though the sentiment rings true. What we are observing with Aave (along with other lending venues), is a failure of basic risk management that, in times of emergency, gets replaced by ad hoc human intervention. Deals are being made in the shadows between well-capitalized individuals to shore up balance sheets and defend (i.e. manipulate) prices. Without this human cooperation and game theory, the lowering tide will leave many ships stranded on the sand. Could this have been averted?
Humans pick up where models and algorithms leave off. The better the model, the less human cooperation needed to step in and fix the mess. The true problem bracing Aave (and others) is much more narrow, all too common, and thankfully treatable with proper algorithms. Aave is currently experiencing the acute phenomenon of too much riding on one single asset. This is called “concentration risk.”
A position becomes overly concentrated once its sheer size starts to matter. Where this matters most saliently is in relation the overall liquidity of the asset. If you were forced to sell it (during a liquidation) or buy it (to repay a debt), how quickly could you close this trade without pushing the price around? The answer to that question about speed takes into account the liquidity of the markets. At any point in time there are only so many buyers & sellers, and if you drain the market depth too fast, prices start to move.
The solution to this problem is to measure the limits to liquidity, and then charge additional margin for holding large positions. The sad fact is that most on-chain DeFi risk programs stop short of this solution. Instead of measuring actual size, they only require that borrowers maintain a certain ratio of loan-to-value. But every market has its tipping point where an asset’s liquidity gets pushed to its limits. And so it follows that each asset type has a specific size beyond which positions become increasingly riskier.
Double-edged leverage
This risk of heavy concentration cuts both ways, whether prices for go up or down.
Archimedes once quipped: “Give me a firm place to stand, and with a long enough lever I can move the world”. Once that lever gets sufficiently large, moving one end of it merely the length of your arm translates into global catastrophe. In the present case of DeFi, a rather banal depreciation in CRV results in outsized effects.
But before discussing the present-day issues CRV falling through the floor, let’s first take a look at a famous case in DeFi where concentration risk was ignored with respect to prices moving up.
The Big Pump
In November 2022, the DeFi sector was given an instructive lesson for a tuition of $120M. The on-chain exchange Mango Markets operated a margining system that allowed traders borrow cash in order to swap on leverage. Their industry-standard margining system worked as follows: deposit collateral in your account, and you can borrow any other asset, so long as the dollar value of your collateral exceeds the dollar value of your debts by a healthy threshold.
In effect, for $1 of collateral, you could borrow roughly $0.90 of anything else. Your loan-to-value ratio just needs to stay under that magic 90%. The problem is that this magic ratio pays no heed of your actual size. You get $0.90 of credit on the dollar no matter whether you have $100 of collateral or $100 million. It would be theoretically possible to deposit sufficient collateral in order to borrow everything else from the escrow vaults — but who has that kind of money?
Avi Eisenberg noticed the possibility for a profitable strategy on Mango Markets, where he could spend less on the collateral in his margin account than the money that he could borrow. Because of this asymmetry, he could borrow with abandon and allow his account to go into default since the collateral he stands to lose would be worth less to him than the loans he gained. Here is how to recreate it:
- Pick the most illiquid collateral type from the ones accepted by the program.
- Slowly acquire that asset on the open market, filling up your collateral balance.
- Now the attack. Buy what little remains of the circulating asset, jacking its price skywards. The relative appreciation of its price in the open market results in a massive boost in the absolute margin afforded by the collateral.
- Borrow everything you can. According to the program’s rulebook, you can borrow money straight into your pocket so long as the total value of your loans is less than the total value of the (now mooning) collateral you have pledged.
- Walk away. When the collateral value comes crashing back down from its artificial heights, you will have insufficient margin to cover your debts, and you will get liquidated — but you do not care. The money you wasted pumping the collateral pales in comparison to the cash you extracted in outstanding loans.
The strategy was successful. Avi performed a malicious series of trades that will go down in DeFi history not so much for being sinister, but so utterly simple. Everything worked according to the rules of the Mango Markets margining system. His small market buys caused his large collateral position to appreciate in value, and he availed himself of the expanded credit to borrow everything from the pooled lenders. You can make any fruit low-hanging by cutting down the tree.
But simple problems tend to have simple solutions. If a position is overly concentrated, just ask for more margin. One example of a risk engine that includes a concentration charge: FTX. This event on Mango Markets took place before the spectacular collapse of FTX, but that failed company at least had this part of the risk model right, and were quick to point it out. As Sam Bankman-Fried explained on X (née Twitter), under FTX’s risk model, a position as large as Avi’s would be charged extra margin (ie, it affords lower leverage). A position that size would not extract $0.90 on the dollar, but $0.20. That’s 78% less borrow power. Maybe still not enough of a haircut to make the market manipulation strategy un-profitable, but it’s a step in the right direction.
Aave & CRV: Round 1
Now back to Aave — but with one last detour starring Avi.
The saga of Avi’s campaign against weak DeFi risk models did not end with his success at Mango Markets. Another soft-spot came under his cross-hairs: Aave’s decision to use CRV as collateral. Unlike Mango Markets, for Avi this time the plan was to push prices down. The strategy has more moving parts than the first, but amounts to the same moral: a risk model that ignores size is destined for death spirals.
The idea was to short CRV, heavily. But how to ensure the big short ends up highly profitable? The Achilles-heel Avi correctly identified is that CRV is overly-concentrated as collateral on Aave and — as with Mango Markets — Aave does not scale margin requirements based on position size. (They do place global caps on borrows, but up until that hard ceiling, the size does not factor into their margin requirements for borrowers).
His goal was to set in motion a cascade of liquidations on the defaulting loans backed by CRV. The situation was like a uranium bomb primed to explode. There is a critical mass of collateral beyond which an uncontrollable chain reaction of liquidations takes over: each subsequent liquidation further depresses the price of CRV, causing more loans to get liquidated, at higher & higher price impact as CRV liquidity drains out of the market. If the chain-reaction were to succeed, there would be little hope to sell off sufficient CRV positions on Aave without suffering massive price impact. And with CRV dropping to the ground, Avi could close his shorts with a tidy profit.
There is more to the fascinating story (spoiler: Avi’s campaign became a quagmire, where everyone got hurt, including Aave). But we’ll leave it off here, just to demonstrate the point that risk models ought to account for size.
Concordia’s answer
Aave is once again facing trouble from too much CRV on the precipice of triggering too much stress on liquidity. The situation is well covered in-depth elsewhere. Simply put, lending brokers allowed for too much leverage on CRV considering its liquidity in the marketplace.
Enough with describing the problem. There are, of course, ways to mitigate this risk.
Here are three pertinent examples from Concordia’s treatment of concentration risk:
1. Apply a concentration charge to jumbo positions.
There is a critical region beyond which a position’s size itself becomes a source of risk. By measuring an aggregate of indicators (such as daily volume, open interest, and floating market capitalization) our risk engine determines size thresholds for every instrument (collaterals & liabilities alike). A borrower is free to breach these critical thresholds, but the extra charge to their margin adds up. Along industry standard heuristics, the charge scales at the square-root of the size.
Given the low liquidity and daily volume for CRV, our risk model would start increasing margin requirements for collateral/debt in excess of ~$10M. This soft ceiling allows CRV investors to collateralize their large holdings, but with gradually diminishing returns on their credit worthiness.
2. Price bands and circuit-breakers
It is impossible to decide with perfect accuracy whether a sudden price jump is bonafide, or the result of a temporary market manipulation. Maybe Oprah just announced her bullishness on a small-cap asset, and the sudden 300% increase in market price is going to stick. The only recourse is to wait and see. If the price snaps back like a rubber band, then any new positions opened at the tip of the temporary cusp would stress the solvency of the broker. See the Mango Markets incident above.
Concordia’s solution is to pause certain kinds of transactions when prices move abnormally. We measure the variance of every asset’s historical returns, and set appropriate price bands several standard deviations away from the moving average. When the observed price crosses the band in either direction (high or low), the system will temporarily suspend a set of transaction types until the moving average “catches up” with the observed price, bringing it back inside the bands.
- If prices shoot upward, Concordia clamps the value **of collateral with respect to opening new borrows. Otherwise, a market manipulator could take out loans on unnaturally inflated collateral values, and leave the lenders with bad debt once the collateral comes crashing back down to earth. Clamping the value of collateral to a band on the high-side protects new loans being taken out on unfairly priced collateral.
- From the other direction, if the price moves suddenly downward, Concordia clamps the value of the borrowed asset respect to new borrows. Suppose some cartel performs a coordinated short, pushing an asset’s price down, and all at the same time borrowing more of that asset to continue the short. If the low price is not sustainable, then the rebound back upwards would result in a sudden ballooning of debt, which could run the risk of creating bad debt.
3. Diversification Benefit
Margin requirements are reduced for diversified portfolios, incentivizing borrowers to spread out their collateral holdings among different instruments. In our case, “diversification” is objectively measured based on the correlation of historical returns for assets. If all the returns of all the collaterals in a portfolio are highly correlated, then there is little diversification, and no reduction to the overall margin requirements for the portfolio. Correlated collaterals are likely to move as one group in the future, and so pose a concentration risk when taken together. The opposite is true for uncorrelated holdings.
The guiding principle is to incentivize portfolios that do not cause exaggerated effects to a balance sheet upon ordinary changes in price. Charging more margin for large (or correlated) positions, while sanitizing price data in relation to each asset’s nominal volatility, would have reduced the damage done to Mango Markets, Aave, and many, many more loan brokers.
The fact of concentration risk is well-known and comes with plenty of mitigating strategies on offer for institutions to adopt. But synthesizing them into a coherent & generalizable model is not easy, both from the standpoint of theory and from daily operations. Moreover, it is very difficult for DeFi protocols to alter the substance of their risk engines while already servicing customers.
Concordia’s purpose is to provide risk-as-a-service for digital asset brokers. We specialize on the hard work of quantifying portfolio risk, and provide brokerages with streaming, real-time access to the our models. Both on-chain and off-chain lenders can plug into our system to manage margin requirements and collateral escrow, relieving them of the difficult burden to institute & run a risk department all by themselves.
