Remaking the Maker Protocol
by Ross Ulbricht
A follow-up to this article is here.
I read the Maker Protocol white paper recently. What a cool concept! The people behind Maker have created a cryptocurrency that tracks the value of the US dollar (a stable coin). You get all the benefits of a cryptocurrency without the crazy price swings. Naturally, it has become very popular, and hundreds of millions of dollars worth have been minted.
I dug further into Maker’s documentation, and read some articles about its recent collateral crisis in March 2020. There is no way I can get a full understanding of the system and what happened from within prison, but taking what I have read at face value, I believe I can contribute some ideas that will help keep this kind of crisis from happening again.
The stable coin issued by the Maker protocol is called DAI. The value of DAI is backed by collateral held in virtual vaults on the Ethereum blockchain. Vault owners are the foundation of the whole system. The value in their vaults backs the value of all DAI in circulation. If the total value in the vaults ever falls below the dollar-pegged value of all the DAI, then the system is insolvent and can be considered a failure.
I think the problem with the Maker Protocol which led to its recent crisis is a misunderstanding of the role of vault owners. In several places, I have seen them referred to as “borrowers,” which did not make sense to me at first. What are they borrowing? And then it clicked: the idea is that vault owners are borrowing DAI from the Maker Protocol itself and putting up collateral (usually ether) in their vaults to back the loan. This is like taking a mortgage out on your house with the ether being the house and the DAI being a loan from a bank. Just like with the mortgage, vault owners have to pay interest on the money they have borrowed in what is called a “stability fee.” And if the value of their collateral drops too low, their vault can be auctioned off. This is like a house going into foreclosure. Here is why this analogy does not work: the DAI being “borrowed” has no value apart from the collateral that is backing it. Its value comes from the fact that, when push comes to shove, it can be redeemed for the collateral in a liquidation auction. The Maker Protocol has assumed the role of the bank, issuing loans backed by collateral, but it seems to me that the vaults themselves are the banks, at least how banks used to be.
Before the advent of modern central banking, banks used to hold gold in their vaults, then print and lend out paper currency backed by that gold. Vault owners in the Maker Protocol are doing something similar. They are holding ether in their vaults, then lending out DAI backed by that ether. So, they would be more appropriately thought of as lenders, not borrowers. Banks were supposed to keep enough gold in their vaults to cover all outstanding currency that the gold was backing. In practice, they printed more currency than they could back and just hoped everyone didn’t show up at once for their gold. When the inevitable bank runs did occur, bankers looked to their pals in government for a bailout. Nowadays, dollars are backed by nothing and fractional reserve banking is the norm. Banks still get into trouble and expect to be bailed out. It’s a bit of a mess, really.
This is not a good model for Maker to emulate. The protocol is not a modern bank because it doesn’t have the backup of the government. Instead, vault owners should be treated like good old-fashioned banks, except they can’t cheat because all of the accounting is done publicly on the blockchain. If they try to issue more DAI than their ether can support, they run the risk of being liquidated. The fundamental difference between this and the current way the Maker Protocol is designed is that, as lenders, vault owners should collect interest, not pay it.
When I read the white paper, my first question was “why on earth would anyone put their valuable ether in a vault to back other people’s stable coins (DAI) if they are going to be charged a fee? And on top of the stabilization fee, they also risk losing 13% of their collateral to a liquidation fee if the collateral’s value drops too low. So why take all this risk without a reward?” It turns out the system rewards other behavior (that should not be rewarded), and vault owners can take advantage of it.
The protocol has a smart contract users can lock their DAI in and receive a “savings rate” while it is there. “Ok,” I thought. “If the savings rate is higher than the stability fee, then vault owners can just mint DAI backed by the ether in their vaults, put it in the savings contract and — so long as they mind their liquidation ratio — get a return on their ether (savings rate minus stability fee).” But why would we want to incentivize this? The whole point of DAI is that it can circulate as a decentralized, blockchain-based stand in for US dollars, not be locked in a smart contract.
It turns out before the crisis in March, the savings rate was around 8% and the stability fee was around 0.5%. So my suspicions were correct: vault owners could get a return (7.5%), not by selling DAI into circulation and being good stewards of their vaults, but by keeping their minted DAI for themselves and parking it in the savings contract. Apparently, a huge fraction of all DAI in existence was in the savings contract before the crisis hit. (see Figure 1.)
The purported purpose of the savings contract is so the committee overseeing the protocol (Maker DAO) can adjust the savings rate to draw DAI in and out of circulation in order to influence the price and keep it close to parity with the dollar. I really don’t think this is necessary. Market forces are what keep the price stable because there are arbitrage opportunities on both sides of parity that push the price back to equilibrium. The savings contract just feels like a tool a central banker would want so he can imagine he is “managing the economy.” This top-down impulse to control the price is not in harmony with the decentralized ethos of the protocol. Worse, it is incentivizing behavior that makes the system less, not more stable.
A fully decentralized protocol is “governed” by math and logic, leaving the human decision making to the end users. The reliance on a central committee to set important parameters within the system is a weakness, not a strength. That is not to say that the Maker Protocol could necessarily function without a governing committee. It is a complex system, and the fact it works at all is impressive. However, the committee must be seen for what it is: a stop-gap. It is not an efficient mechanism for discovering the best values for the system parameters it oversees.
Crisis
Around mid-March, the price of ether dropped suddenly by about 50%, and the price of DAI spiked by over 10% (a huge amount for something that is supposed to trade one-for-one with the dollar). Many vaults became under-collateralized as the value of their ether dropped and were forced to liquidate. That’s a risk vault owners take. It is their job to make sure ether is added to make up the shortfall or that their outstanding DAI balance is paid down. However, the crisis became so acute that some vaults were forced to liquidate at extreme discounts, losing nearly all (and in some cases all) their collateral.
The problem stemmed from the fact that the whole system is chronically under-collateralized. Not only are vaults owners penalized with a stability fee for collateralizing the system, but the DAI burden is constantly being added to by the savings rate. Then during the crisis, no one wanted to give up their stable, high-yield DAI for a crashing, negative-yield ether in the liquidation auctions.
In the end, DAI had to be backed by USDC, a stablecoin backed by cash in a bank (hardly a decentralized solution). If the Maker Protocol is to become truly decentralized and independent, it has to get the fundamental incentives right and reject the allure of top-down control.
A Modest Proposal
Obviously, I advocate abolishing the savings contract and stabilization fee, but to replace them with what? First of all, the savings contract need not be replaced. The value of DAI is its stability, not some arbitrary rate of return. With regard to the stability fee, it should be negative. That is to say, vault owners should be rewarded, not penalized for providing collateral. So, it shouldn’t be called a stability fee at all. It really should be called a savings rate because vault owners are saving their ether in a vault and earning a return on it for the service of backing DAI. But this could be confusing because that term has already been used. Instead, it could be called a “collateral rate” for the rate of return on collateral.
Whatever its name, it will draw collateral into the system rather than repelling it. Where, you may ask, will the DAI come from to pay this collateral rate? It should come from the DAI issued by the vaults providing collateral. It is the holders of DAI who are being provided a service (a decentralized stablecoin), so it is they who should pay. And how should this rate be determined? A simple fix to the current system would be for MakerDAO (who already sets the stability fee and savings rate) to reduce the savings rate to zero and the stability fee to some small negative number, perhaps -1%. This would have to be done such that DAI would simply disappear at a rate of 1% per year, leaving that much more collateral unencumbered than would be otherwise.
A better solution would be to set up a system whereby vault owners set their own rates and compete with each other for the interest from DAI holders, with the lowest rates winning. This would keep rates low generally, but when there is a serious collateral shortage, rates will automatically rise via market forces, encouraging more collateral to come in and discouraging DAI hoarding. This is exactly the opposite of what we saw during the crisis of mid-March.
With the incentives appropriately aligned, there should be a much deeper market for DAI with tighter bid/ask spreads. The reason for this is that, with a positive rate of return, vault owners will be wanting to get their newly-minted DAI out onto the market so that others will be holding the depreciating asset (the rates cancel out while they are holding their own DAI). Thus, there will be ample liquidity above parity.
Below parity, there is an arbitrage opportunity so long as DAI holders can redeem DAI for collateral. If the price is below parity, people can buy discounted DAI, redeem them for excess collateral at the oracle exchange rate from the issuing vault, and then sell the collateral on the market at full value. (The oracle exchange rate is how the Maker Protocol determines collateral value). Thus, there will be ample liquidity below parity.
These forces will stabilize the price and ensure full collateralization, especially in turbulent market conditions.
Technical Part
The mechanics of this market-based system for determining the collateral rate get a bit complicated. Just skip to the conclusion if you are not interested in the details.
As noted above, vaults should set their own rates, with low-rate vaults winning the competition. However, unless there is excess collateral in the system, there is no competition. If all collateral is backing DAI, vault owners can charge whatever they want, with redemption for collateral the only recourse left to DAI holders. We can solve this by allowing unencumbered collateral to collect the interest of another vault’s DAI if its collateral rate is lower than the other vault’s rate. Thus, without having to mint DAI and sell it on the market, a vault owner can collect interest just by charging a lower rate. This is great for keeping rates low, but now there is no incentive to actually issue and sell DAI, especially because the vaults that do risk liquidation.
A compromise must therefore be struck whereby excess collateral in a low-interest vault can capture only a fraction of the issuing vault’s interest, with the rest going to the issuing vault, but at the low-interest vault’s rate. Thus, with excess collateral in the system, there will be a natural cutoff rate above which vaults that have issued DAI are losing interest to vaults with excess collateral below the cutoff, with both earning at the lower rate.
Which vaults capture from which? The vaults above the cutoff with the lowest rates should be captured from by the vaults with the highest rates below the cutoff. This will incentivize high-rate vaults to keep their rates low.
What fraction of the issuing vault’s interest should be captured by excess collateral below the cutoff? This should vary based on how much excess collateral is in the system compared to how much total DAI there is outstanding. If all DAI is paying interest to excess collateral (i.e, vaults with only excess collateral are below the cutoff), the capture rate should be 0%. The system is awash in excess collateral. The rate should increase linearly, reaching 100% when there is no excess collateral in the vaults below the cutoff: the system is near insolvency and needs excess collateral fast.
All of this handles the incentives keeping excess collateral in the system and collateral rates low, but the reason we want excess capacity is so we can use it when needed. To do that, there needs to be a mechanism whereby DAI are automatically switched from vaults close to insolvency to ones with excess collateral. This could be done through liquidation auctions the usual way. Vaults would mint DAI with their excess collateral and use it to bid. However, it would be good if this were done automatically to speed things up in a crisis, with the assets and liabilities (collateral and DAI) simply transferred to the well-capitalized vaults. Vaults could mark a portion of their excess collateral for this purpose and set a fee, with vaults liquidating to the low-fee vaults first.
The collateral rates set by the various vaults in the system will tend to converge on a single “market rate,” which is just the cutoff rate. Vaults below the cutoff are “leaving money on the table” because — so long as they don’t go above the cutoff — they can charge more without losing to a cheaper vault. One the other hand, vaults above the cutoff will be losing out to vaults below it with excess collateral.
When collateral (or the value of collateral in the case of an exchange rate change) drops, the cutoff will automatically rise because there will be less excess collateral below the cutoff, leaving higher-rate vaults to collect their full interest. The capture rate will also increase because a smaller fraction of DAI is paying interest to excess collateral. Both of these things will encourage new vaults and new collateral to flow into the system, alleviating the shortage, all without intervention from MakerDAO.
I tried to think of a way that the capture rate too could be decentralized and governed by market forces rather than by the linear formula above, but could not. The problem is that vaults are already competing based on their collateral rate for the benefit of DAI holders, so how could they also compete based on the capture rate for the benefit of vaults issuing DAI at the same time? One could devise a two-parameter formula that weighs these two values, leading to a single case, with interest going to the vaults with the lowest score, but this seems very complicated and rife with hidden loopholes. One could also leave the collateral rate and capture rate up to the vaults and direct interest to the vaults with the lowest total income, but then vaults could use extreme setti ngs, favoring DAI holders with very low collateral rates, but capturing 100% of the interest, for example.
Conclusion
The Maker Protocol is a very cool concept, and I hope it succeeds, but I fear it will meet with the kind of crisis we saw in mid-March 2020 again if these fundamental issues are not addressed. The next crash could result in systemic insolvency and catastrophic failure. There is a lot of money and even more potential riding on this. The crisis should be a wake-up call that reform is needed or new protocols need to be tried. I just hope the ideas above can help point that reform in the right direction.
