Vitalik: Two Thought Experiments for Evaluating Automated Stablecoins

The recent LUNA crash, which resulted in tens of billions of dollars in losses, has unleashed a storm of criticism against the category of “algorithmic stablecoins”, with many considering them “fundamentally flawed products”. Greater scrutiny of DeFi financial mechanisms, especially those striving to optimize for “capital efficiency”, is very welcome. Greater acknowledgment that current performance is no guarantee of future returns (or even future complete collapse) is even more welcome. What is deeply wrong, however, is for people to paint all automated pure crypto stablecoins with the same brush and dismiss the entire category.

While there are many automated stablecoins with fundamental design flaws that are doomed to collapse, and many more that could theoretically survive but are highly risky, there are also many that are theoretically very healthy and have withstood the extremes of crypto market conditions in practice. So what we need is not stablecoin boosterism or stablecoin doomsdayism, but a return to principle-based thinking . So what are some good principles for evaluating whether a particular automated stablecoin is truly stable?

For me, I started by testing how stablecoins respond to two thought experiments.

What are automated stablecoins?

For the purposes of this article, an automated stablecoin is a system with the following properties:

1. It issues a stablecoin that attempts to target a specific price index. The usual target is $1, but there are other options. If the price deviates in either direction, there are targeting mechanisms that continually push the stablecoin price toward the index. This makes ETH and BTC not stablecoins (duh).

2. The target mechanism is fully decentralized and has no protocol-level dependencies on specific trusted actors. In particular, it must not rely on asset custodians. This makes USDT and USDC not stablecoins.

In practice, (2) means that the target mechanism must be some kind of smart contract that manages some reserve of crypto assets and uses these crypto assets to support prices when prices fall.

How does Terra work?

Terra-style stablecoins (which are roughly the same as seigniorage shares, although many implementation details differ) work through a dual-coin model, which we call a stablecoin and a volatility coin (in Terra’s case, UST is the stablecoin and LUNA is the volatility coin). Stablecoins use a simple mechanism to maintain stability:

1. If the price of the stablecoin exceeds the target value, the system will auction new stablecoins (and use the proceeds to burn volatility coins) until the stablecoin price returns to the target.

2. If the price of the stablecoin falls below the target value, the system will repurchase and destroy the stablecoin (issuing new volatility coins to fund the destruction) until the price of the stablecoin returns to the target value.

Now, what is the price of a vol coin? The value of a vol coin may be purely speculative, relying on the assumption that there will be greater future demand for the stablecoin (which would require burning vol coins to issue). Alternatively, the value of a vol coin may come from fees: transaction fees for stablecoin <-> vol coin transactions, or annual holding fees charged to stablecoin holders, or both. But in all cases, the price of a vol coin comes from expectations of future activity in the system.

How does RAI work?

In this post, I focus on RAI rather than DAI because RAI better embodies the pure “ideal type” of a collateralized automated stablecoin, being backed only by ETH. Whereas DAI is a hybrid system backed by both centralized and decentralized collateral, which is a reasonable choice for their product, but it does make the analysis trickier.

In RAI, there are two main types of participants (there are also holders of FLX, a speculative token, but they play a less important role):

1. RAI holders hold RAI, the stablecoin of the RAI system.

2. RAI lenders deposit some ETH into a smart contract object called a "safe". They can then withdraw RAI worth 2/3 of the ETH (e.g., if 1 ETH = 100 RAI, then if you deposit 10 ETH, you can withdraw about 667 RAI). If the lender repays their RAI debt, the ETH can be recovered in the same way.

There are two main reasons to become an RAI lender:

1. Long ETH: If you deposit 10 ETH in the example above, and withdraw 500 RAI, you will end up with a position worth 500 RAI, but with an exposure of 10 ETH, so the position will rise or fall 2% for every 1% change in the price of ETH.

2. Arbitrage: If you find an investment denominated in fiat that is rising faster than your RAI, you can borrow RAI, put the money into that investment, and profit from the difference.

If the price of ETH drops, and the safe no longer has sufficient collateral (that is, the RAI debt now exceeds 2/3 of the value of the deposited ETH), a liquidation event occurs. The safe is auctioned off and made available for others to purchase by providing more collateral.

Another major mechanism to understand is the redemption rate adjustment. In the case of RAI, the target value is not a fixed amount of dollars, instead, it moves up or down, and the rate at which it moves up or down adjusts based on market conditions:

Thought Experiment 1: Even in theory, can stablecoins safely allow all users to exit?

In the non-crypto real world, nothing lasts forever. Companies go out of business all the time, either because they couldn't find enough users in the first place, because the strong demand for their product no longer exists, or because they are replaced by a stronger competitor. Sometimes, there are partial collapses, where companies fall from mainstream to niche status (e.g. MySpace). These things have to happen so that space can be made for new products. But it's important to note that in the non-crypto world, users are usually not hurt too much when products shut down or hit a slippery slope. There are certainly a few who fall through the cracks, but by and large, shutdowns are orderly and problems are manageable.

But what about automated stablecoins? What happens if we look at stablecoins from a bold and radical perspective, that the system’s ability to avoid collapse and loss of large amounts of user funds should not depend on a constant influx of new users? Let’s take a look!

Can Terra land safely?

In the case of Terra, the price of LUNA is derived from the expectation of future fees for activity in the system. So what happens if expected future activity drops to near zero? The market cap of LUNA will drop until its market cap ratio relative to stablecoins becomes very small. At this point, the system becomes extremely fragile, and a small downward shock to the demand for stablecoins could cause the targeting mechanism to issue a large amount of LUNA, leading to hyperinflation of LUNA, at which point the stablecoins also lose value.

The collapse of the system can even become a self-fulfilling prophecy: if the system looks likely to collapse, that lowers users’ expectations of future fees, which is the basis of the value of the volatile coin, further pushing down the market value of the volatile coin, making the system more fragile and ultimately potentially triggering a crash, as we saw with Terra in May.

First, the price of the volatile coin falls, and then the stablecoin begins to shake. The system tries to support the demand for stablecoins by issuing more volatile coins. Since users have low confidence in the system, there are few buyers, so the price of the volatile coin falls rapidly. Finally, once the price of the volatile coin approaches zero, the stablecoin will also collapse.

In principle, if demand falls very slowly, the expected future fees of the volatile coin and its market capitalization equivalent to the stablecoin may still be large, so the system will continue to be stable at each step of its decline. But such a successful slow decline is unlikely, and the more likely scenario is that user interest falls rapidly, followed by a loud bang.

Safe Descent: At each step, there is enough expected future revenue to justify the volatility coin’s market cap to justify the safe divestment of the stablecoin at its current level.

Unsafe Descent: At some point, there is not enough expected future revenue to justify sufficient volatile coin market cap to secure the stablecoin, so a crash is possible.

Can the RAI land safely?

The security of RAI depends on assets external to the RAI system (ETH), so it is easier for RAI to land safely. If the demand decline is unbalanced (so either holding demand declines faster or lending demand declines faster), the redemption rate will adjust to balance the two. Lenders hold leveraged positions in ETH, not FLX, so there is no risk of a positive feedback loop where falling confidence in RAI causes lending demand to also decline.

If, in an extreme case, all demand to hold RAI disappears simultaneously, leaving only one holder, redemption rates will soar until eventually every lender's safe is liquidated. The remaining single holder will be able to purchase the safe in a liquidation auction, use their RAI to immediately liquidate their debt, and withdraw ETH. This gives them the opportunity to receive a fair price for RAI and get paid for the ETH in the safe.

Another extreme case worth investigating is that RAI becomes the dominant application on Ethereum. In this case, the expected reduction in demand for RAI in the future will drive the price of ETH down. In an extreme case, a cascade of liquidations could occur, leading to a chaotic collapse of the system. But RAI is much more resilient to this possibility than a Terra-style system.

Thought Experiment 2: What would happen if you tried to offer 20% interest per year on a stablecoin?

Currently, stablecoins tend to be pegged to the dollar, with RAI being a minor exception in that its peg adjusts up and down based on the redemption rate, and the peg starts at $3.14 instead of $1 (the exact starting value is a friendly concession, since a true math nerd would have chosen tau = 6.28 USD). But they don’t have to be, you can peg a stablecoin to a basket of assets, or the consumer price index, or some arbitrarily complex formula (“enough to buy {the global average CO2 concentration minus 375} hectares of forest land in Yakutia”). As long as you can find an oracle to prove the index, and get people involved in all aspects of the market, you can make such a stablecoin work.

As a thought experiment to assess sustainability, let’s imagine a stablecoin with a specific exponent: the amount of USD that grows by 20% per year. In mathematical terms, the exponent is

USD, where t is the current time in years and t0 is the time the system was started. A more interesting option is

USD, so it starts out like a USD-denominated stablecoin, but the USD-denominated returns are growing at 4% per year.

Obviously, no real investment can return anything close to 20% per year, and definitely no real investment can return 4% per year forever. But what would happen if you tried?

I would claim that there are basically two approaches to stablecoins that attempt to track such an index:

1. It charges holders a sort of negative interest rate to offset the dollar-denominated growth rate built into the index.

2. It turned into a Ponzi scheme, giving stablecoin holders amazing returns for a while, until one day it suddenly collapsed.

It should be easy to see why RAI is (1) and LUNA is (2), so RAI is better than LUNA. But this also shows a deeper and more important truth about stablecoins: in order for a collateralized automated stablecoin to be sustainable, it must somehow control the possibility of implementing negative interest rates. A version of RAI that programmatically prevents the implementation of negative interest rates (which is essentially what early single-collateral DAI did) would also become a Ponzi scheme if pegged to a rapidly appreciating price index.

Even outside of the crazy hypothetical scenario where you build a stablecoin to track a Ponzi index, the stablecoin must be able to somehow cope with the situation where demand for holding exceeds demand for borrowing even at zero interest rates. If you don’t do this, the price rises above the peg, and the stablecoin becomes vulnerable to price movements in both directions, which are highly unpredictable.

Negative interest rates can be achieved in two ways:

1. RAI style, with a floating target that would decline over time if redemption rates were negative;

2. Actual balances will decrease over time;

Option (1) has a UX flaw, where the stablecoin no longer clearly tracks “$1”, and option (2) has a developer UX flaw, but choosing one of these two seems inevitable — unless you go the MakerDAO route and become a hybrid stablecoin that uses both pure crypto assets and centralized assets like USDC as collateral.

What can we learn?

In general, the crypto industry needs to move away from the attitude that security depends on endless growth. It is surely unacceptable to maintain this attitude by saying “the fiat world works the same way” because the fiat world is not trying to provide anyone with returns faster than normal economic growth (except in isolated cases that deserve the same fierce criticism).

Instead, we should assess the safety of a system by looking at its steady state, even its pessimistic state under extreme conditions, and whether it can ultimately be shut down safely. If a system passes this test, that doesn't mean it's safe. It could still be vulnerable for other reasons (such as insufficient collateral ratios), or have code vulnerabilities or governance vulnerabilities. But robustness in steady state and extreme cases should always be the first thing we check.

Note: The original author is Ethereum co-founder Vitalik Buterin.

Special thanks to Dan Robinson, Hayden Adams, and Dankrad Feist for their feedback and reviews.