On Forking DeFi Protocols

As a suite of new layer 1 blockchains are launching, I’ve been thinking about Ethereum’s network effects, and the defensibility of the DeFi protocols built on top of Ethereum.

A couple of years ago, I wrote about the network effects of non-sovereign Layer 1 monies like Bitcoin and Ethereum. Since then, the DeFi ecosystem on top of Ethereum has blossomed: utilizing a couple dozen DeFi protocols, users have withdrawn a few hundred million dollars of debt fully collateralized against an even larger pool of capital.

Now that these protocols are collectively facilitating a few hundred million dollars of economic activity, it’s possible to reason about defensibility. One way to do this is to quantify and compare network effects. However, it’s very difficult to quantify network effects with precision since the underlying dynamics of each protocol are unique, making it challenging to compare on an apples-to-apples basis.

In this essay I’ll consider the 1) effort, and 2) capital required to fork each of the major DeFi protocols. Then I’ll rank the relative strengths of these network effects, and conclude with a discussion of Ethereum’s ecosystem level defensibility.

This essay assumes working knowledge of each protocol.

Synthetic Stablecoin - Maker

About a year ago, I wrote about how Layer 2 assets such as MKR (the “equity” of the MakerDAO system) can capture value in permissionless, open settings. In that essay, I specifically identified the presence of “unforkable state” as the key to value capture. In the context of Maker, the obvious unforkable state is the collateral (primarily ETH) backing the DAI loans.

However, it’s now clear that this framing is incomplete. To understand why, let’s assume that the only source of network effect for Maker is the collateral. A wealthy 3rd party could fork all of Maker’s contracts, create an alt-Maker ecosystem, and deposit tens of millions of dollars of collateral to bootstrap liquidity. But what then? The alt-Maker system is useless if no one wants to buy or interact with alt-DAI.

Maker’s most potent source of defensibility is not MKR, or the collateral backing the DAI loans, but the liquidity and usability of DAI. DAI must be liquid in order for Maker to be usable. If someone withdraws DAI debt against ETH collateral, the DAI is useless if there’s no liquidity. But usability is a superset of liquidity. DAI’s usability is clear acceptance by merchants, its use in other protocols like Augur, and its use as collateral in lending protocols like Compound and Lendf.me. DAI is plugged into all kinds of 3rd party apps, services, and infrastructure, and that makes it more useful and usable.

The combination of DAI’s liquidity and usability is a powerful moat. A well capitalized alt-Maker team could offer a higher Dai Savings Rate (DSR), and pay 3rd parties to integrate alt-DAI, but it’s unclear if this would gain meaningful traction.

Fiat Collateralized Stablecoins - USDT

While Tether is not a pure DeFi protocol, with an outstanding market cap of over $5B, I included it because it’s such an integral part of the crypto ecosystem.

The source of the defensibility for USDT is clear: it’s the most liquid asset in the crypto ecosystem alongside BTC. It is available on all major non-US exchanges, serves as collateral for many derivatives exchanges, and is used to settle a huge percentage of OTC trades.

Despite fierce competition from USDC, PAX, TUSD, GUSD, DAI, and others, USDT still commands >80% of the stablecoin market (measured using market cap). This is the ultimate testament to the defensibility of USDT.

There are a few teams that are working on stablecoin clearinghouses, including DeFi protocols such as StableCoinSwap, and Shell; and centralized clearing houses like Stablehouse. If these are successful - and therefore reduce the friction associated with trading stablecoins - USDT may be negatively impacted.

For example, if these protocols/companies provide strong guarantees that large quantities of stablecoins can be swapped with minimal slippage, derivatives exchanges may begin to accept other stablecoins as collateral. Today, FTX offers this service natively; however, the presence of liquid stablecoin clearinghouses may accelerate this trend, which is likely bad for USDT.

Collateralized Money Market - Compound / Lendf.me

The unforkable state in Compound is the collateral.

Therefore, the defensibility of Compound can be understood as follows: as the value of the collateral pool increases, borrowers can borrow more capital, which draws more lenders, etc.

How difficult is it for someone to fork Compound and bootstrap liquidity in alt-Compound?

There are a few ways to do this. The alt-Compound team can:

Support assets that Compound does not support (e.g. USDT). Introduce more favorable collateralization ratios and liquidation penalties. Lend their own assets in the alt-Compound pool at a competitive or even discounted rate. Subsidize 3rd party lenders to undercut Compound’s rates.

Today, Compound has less than $100M of collateral. If the creators of alt-Compound undercut Compound’s rates by subsidizing users - for example on the order of 100 basis points per year - the annualized opportunity cost of bootstrapping liquidity would be less than $1M. This level of scale is venture fundable.

However, in addition to Compound’s internal liquidity (in the form of lending and borrowing rates), Compound is also subject to a couple of unique forms of external liquidity that may provide additional defensibility.

First, there are 3rd-party aggregators such as Instadapp, Zerion, RAY, idle.finance, Aave, etc. These systems route deposits to Compound, which in turn lowers borrowing rates, which then attracts more borrowers. While organic capital flow is certainly good, it’s not clear that it matters on the margin (since an alt-Compound team can subsidize rates to bootstrap liquidity).

Interestingly, the presence of aggregators can actually backfire because the aggregators are incentivized to send user assets to the highest-yielding lending pools. Assuming similarly trusted contracts, governance and oracle mechanics, aggregators will not be loyal to Compound at the expense of their users, and so an alt-Compound can easily win over aggregators with subsidies. Moreover, a sufficiently large aggregator can siphon liquidity away from Compound into its own pool or alt-Compound fork. While this hasn’t happened yet, I expect it will in the coming years.

Overall, it’s unclear if 3rd-party aggregators will act as a substantial source of defensibility.

Second, let’s consider cTokens. cTokens are somewhat-analogous to DAI. If third party apps integrate cTokens (for example, for use as collateral), that makes cTokens more usable outside of the core Compound protocol. That makes it difficult for lenders (cToken holders) to move from Compound to alt-Compound.

While the Maker/DAI - Compound/cToken analogy is good, it’s not perfect: the only reason to create DAI is to sell it for something else (e.g. more ETH). Therefore alt-Maker is useless unless there is a market for alt-DAI. However, this is not true for Compound. Compound is still useful even if 3rd-party apps do not utilize cTokens.

Empirically, this is playing out as one should expect. The China-based dForce community forked the Compound codebase and launched a collateralized money market protocol called lendf.me, and they’ve already bootstrapped ~$20M of collateral into the system in just a few months. They accomplished this by

1. Offering products that Compound does not support (notably USDT, imBTC, and HBTC)
2. Localizing the service with 3rd party integrations for Chinese users.

It does not appear that the dForce community had to subsidize rates on the lendf.me product to accomplish this.

Maker is more defensible than Compound. With a subsidy budget, anyone can fork Compound and bootstrap liquidity internal to the lending/borrowing market. But successfully forking Maker requires more than a subsidy budget: it requires liquidity and usability of DAI external to the protocol.

Generalized Synthetic Asset Protocol - Synthetix

Synthetix is a specific type of exchange focused on trading synthetic assets. The defensibility of an exchange is generally understood to be a function of liquidity. However, Synthetix is not a traditional exchange because it does not offer a central limit order book (CLOB) like virtually all major exchanges across traditional markets and crypto markets (e.g. NYSE, CME, Coinbase, Binance).

One of the defining features of Synthetix is that takers do not incur any slippage when trading synths against the collateral pool; however, liquidity is limited based on the amount of collateral in the system. This means that liquidity - and therefore defensibility - is primarily a function of available collateral.

Interestingly, the growth of the Synthetix exchange is actually hampered by the need for takers to onboard into the Synthetix ecosystem by trading real assets (non-Synths such as ETH) for Synths (such as sETH). Today, most users onboard into the Synthetix ecosystem via Uniswap, and the largest liquidity pool on Uniswap is sETH-ETH. So while the need for a liquidity bridge is a constraint to growth, it’s also conversely a moat: if someone forks the Synthetix system to create alt-Synthetix, she will need to bootstrap an analogous liquidity bridge.

How do the network effects of Synthetix compare to Maker and Compound?

First, let’s consider collateral in the protocol. Like in the cases of alt-Maker and alt-Compound, anyone who forks Synthetix can capitalize the collateral pool themselves, or subsidize others for doing so. Therefore, the collateral base is unlikely to provide defensibility.

Next let’s consider exogenous assets: DAI, cTokens, and Synths. Unlike Maker’s DAI, Synths do not require liquidity external to the protocol... by design! Instead, Synths are more comparable to Compound’s cTokens: like cTokens, Synths can be used as collateral in third-party apps, but don’t need to be in order for the protocol to function. While this can become a source of defensibility, it hasn’t yet.

The last major form of defensibility is the real asset <--> Synth bridge. While Synthetix leverages Uniswap for this today, an alt-Synthetix team could easily provide their own real asset <--> alt-Synth bridge using Uniswap, Kyber, or other freely available DeFi tools.

Automated Market Makers - Uniswap, StableCoinSwap, Shell, Bancor, FutureSwap, Kyber

Compound is an automated market maker (AMM), albeit for borrowing/lending instead of trading. As such, the defensibility of most of the trading-focused AMMs can be understood to be comparable to that of Compound, excluding cTokens.

Empirically, this seems to be the case. While not all of these AMMs are directly competitive because of different product focuses (e.g. StableCoinSwap and Shell are focused on stablecoin trading, while FutureSwap is focused on futures), the defensibility of each is primarily a function of the size of each protocol’s liquidity pool. Whereas larger liquidity pools in Compound allow for tighter lending/borrow rates, larger liquidity pools in trading-focused AMMs offer lower slippage for takers.

Kyber has become the most liquid AMM over the last 12 months largely by 1) tapping into other AMM liquidity pools such as Uniswap, and 2) because of third-party integrations that route taker order flow. It’s clear that all the AMMs are going to tap into one another’s liquidity pools as they continue to improve over time.

Paradoxically, once all the AMMs within a given vertical (e.g. stablecoin swaps) tap into one another’s liquidity pools, all of those AMMs become perfect substitutes. None of the AMMs will be able to compete on distribution. The ultimate winner from this end-state of perfect competition will be takers, who will always receive best execution.

Non-Custodial Central Limit Order Book Exchanges - dYdX, IDEX, Nuo, 0x

The defensibility of these protocols are comparable to those of centralized exchanges, albeit with a few disadvantages.

First, all of these protocols are subject to the constraints of the underlying blockchain which ultimately settles trades, including non-deterministic order execution, high latency, and miner front running. All of these constraints deter liquidity providers, and therefore increase slippage.

Second, in general none of the decentralized exchanges support cross-margining or position netting. While I hope to eventually see this develop in the DeFi ecosystem, it’s clear that this is years away. Meanwhile, centralized exchanges like FTX and Binance offer cross-margining today, and are rapidly expanding their product offerings to maximize capital efficiency for traders.

Mixer - Tornado.cash

Tornado Cash is unique among the other DeFi protocols above. While the others are focused on borrowing/lending and trading, Tornado is focused on mixing funds to maximize user privacy.

Today, Tornado Cash does not support private payments in a pool. Rather, it can just be used to anonymize funds. The source of defensibility in Tornado is the size of the anonymity pool. Since funds cycle through the Tornado pool relatively quickly (the entire asset base turns over every 1-2 weeks), the network effects are fleeting. Moreover, beyond a certain point, a marginally larger anonymity pool doesn’t really matter. For example, as the anonymity set grows from 500 to 1,000 addresses, it’s not clear that the next marginal user cares. Who is the marginal user who believes 1/500 isn’t good enough, but that 1/1,000 is? Thus, in its current form, Tornado Cash is not that defensible.

However, in a future version of the service, Tornado Cash aims to support privacy-enabled asset transfers inside the privacy pool (rather than just anonymizing funds, which is what’s available now). In this model, capital will be stickier as it won’t leave the system so quickly. This will allow the anonymity pool to grow much larger, making it more useful for larger amounts of capital.

The notion that large amounts of capital will only enter a large privacy pool is unique relative to the other services above. For example, if the entire privacy pool is just 1,000 ETH, that pool may not be useful for someone wishing to anonymize 9,000 ETH, and in fact be harmful for first 1,000 ETH in the pool, as the owners of the first 1,000 ETH may not want a 90% probability of being associated with the other 9,000 ETH.

For a user who wants to anonymize 10,000 ETH, they may require a pool of 90,000 ETH. This model, while not yet available, is clearly more defensible than the status quo because it enables the wealthiest people to use the service, and the wealthiest people are the people with the largest incentive to hide their wealth.

Ranking

After considering the hypothetical difficulty of forking these protocols and the empirical evidence we have so far (which protocols have and have not been forked), I’ve ranked the defensibility of these protocols from strongest to weakest. Note that this ranking is necessarily conjecture, as it’s impossible to quantify and therefore rank on a purely objective basis:

1. Fiat collateralized stablecoin - USDT
2. Synthetic stablecoin - Maker
3. Mixers - Tornado.cash
4. Generalized synthetic asset protocol - Synthetix
5. Collateralized money market - Compound / Lendf.me
6. Automated Market Makers - Uniswap, StableCoinSwap, Shell, Bancor, Futureswap
7. Non-Custodial Central Limit Order Book Exchanges - dYdX, IDEX, Nuo, 0x

I ranked USDT at the top because it faces the most competition, and is still 5 - 10x larger than its largest competitor, USDC. While USDT is controversial, it’s clearly extremely defensible. Coinbase is one of the best capitalized companies in the space, and has been unable to meaningfully displace USDT after 18 months. While it’s possible that stablecoin clearinghouses may change these dynamics, it’s too early to know.

Based on the commentary above, it should be clear why Maker is next. The Maker protocol does not function if DAI is not liquid and usable external to the core protocol. Both of these traits are not easily forkable or subsidize-able.

I ranked Tornado third, above the lending and trading protocols, because wealthy users - who will provide the vast majority of capital in these protocols - require the presence of other wealthy users. And because wealth is not distributed evenly (relatively few wealthy users), I expect that the market may only support 1 - 2 privacy marketplaces, rather than the 10+ that are available in trading and lending.

While I noted above that Synthetix and Compound are similar, I ranked Synthetix above Compound because of the real asset - Synth bridge.

One of the common traits of the protocols in the bottom half of the list is heightened competition. This is clear empirically: entrepreneurs and venture investors are betting that these markets are not that defensible. Furthermore, as discussed above, competitors can bootstrap liquidity in most of these markets fairly easily by subsidizing liquidity.

Ecosystem-Level Network Effects

I’ll conclude this essay by considering the implications of everything discussed above on Ethereum’s defensibility.

In short, Ethereum’s defensibility - at least as it pertains to DeFi protocols - is materially stronger than that of any individual DeFi protocol. The primary source of Ethereum’s defensibility is not capital, or liquidity, but the composability and interoperability of these protocols as a whole.

It’s truly amazing that someone can use ETH as collateral, withdraw DAI against it, lend out the DAI on Compound or Lendf.me, and use that DAI as collateral to borrow ZRX, and sell the ZRX for ETH…. in a single transaction, in a single moment in time.

The ultimate testament to the power of the Ethereum-level network effects around DeFi was the first bZx attack. The attacker’s transaction was likely the single most complex transaction ever processed by the Ethereum network, chaining together five sophisticated protocols. Recreating this level of interoperable infrastructure in any other ecosystem is going to take years (just as it took Ethereum years). As such, I recommend that most new layer 1 teams focus on other use cases beyond DeFi, at least until they bootstrap their respective ecosystems.

Bravo to the Ethereum community for pioneering DeFi!

Thanks to Haseeb Qureshi, Alex Pruden, Ali Yahya, and Michael Anderson for providing feedback on this essay.