Discover more from Dirt Roads
# 13 | Primitive Obsession (II): I Have Been Trying to Understand Goldfinch and Uncollateralised Lending
Integrating vs. Outsourcing Trust In Lending Protocols
This is another issue of Dirt Roads. Those are not recaps of the most recent news, nor investment advice, just deep reflections on the important stuff happening at the back end of banking. The time we are sharing through DR is precious to me, I won’t make abuse of it.
Platform Plays: Coordinating Information
Everything that amplifies scalability is, under a certain perspective, a platform.
Being the base layer of something is not enough to be considered a platform. Base layers can also be connectors, i.e. bits of infrastructure that make distribution through the system possible. Rather than being accretive, connectors are in the best case frictionless - enabling transmission and transformation without loss. The unfortunate truth is actually that, most of the times, connectors are aware of the strategic position they occupy in complex systems and consequently suck out much more value than what they bring in. We call this behaviour rent seeking.
Platforms, on the other hand, are by definition accretive, to the point that they end up being worth less than what’s built on top of them.
We tend to have an easier understanding of a specific type of platforms: those that improve the information flow or, in other words, that reduce the information loss and the inherent uncertainty of a system. Let’s call them information platforms. Paraphrasing Shannon’s language, information platforms are marginally efficient encoders-decoders with the ability of increasing the capacity of a channel - i.e. the bandwidth of high-quality information a channel can reliably transmit.
Modern political constitutions have been among the most powerful information platforms in human history: if you augment with them thousands of people randomly operating a diverse set of tasks, the uncertainty of the outcome gets drastically reduced. Other great information platforms in history include language, mathematics, roads, the press, binary coding, TCP/IP protocols, cloud computing, and Ethereum - or more specifically decentralised general purpose open-source blockchain technologies (!).
Information systems, however, weren’t all made equal. More specifically, the costs of a divergence (vs. the expected outcome) is not the same for all systems. We call the platforms that operate in information systems where the cost of divergence is very harsh trust platforms. Creating and improving trust platforms is very difficult exactly because the cost of experimenting is so high.
Goldfinch: Uncollateralised Lending
Banks are good trust platforms, as good as they are cumbersome. With the many actors involved, the lack of transparency, and the scarce competition - as a consequence of the insufficient experimentation, banks have lost their mojo and progressively morphed from platforms into connectors or, in extreme cases, bottlenecks.
The Collateral Conundrum
DeFi, supercharged by the trustless nature of blockchain protocols, has been the first credible attempt at inventing a more efficient version of core banking. Even the most successful DeFi lending projects - readers of DR might know by now that I’m a big fan of MakerDAO, however, have seen their activities drastically constrained by the need of incorporating some strong form of trust in the absence of any governmental stop-loss or legal enforceability against bad behaviours. MakerDAO lending vaults, for example, need today to be collateralised well beyond the levels required in traditional finance, with collaterals acting as a valuable guarantee in the hands of lenders in the case borrowers wouldn’t honour their repayment duties. Unfortunately, there is not enough (well-diversified) collateral out there - we already wrote about it here on DR.
There are only two routes available to solve the collateralised lending conundrum:
Reduce or eliminate the need for any collateral
Expand the collateral pool
Goldfinch decided to go for Option 1, and completely got rid of external collaterals from the picture. The idea, trust through consensus, is simple and resonates with today’s DeFi jargon, but its execution, as described in the whitepaper, is laborious. In June of this year, the firm announced a16z and others were backing them with USD 11m in funding, and recently went live with the first borrowing pool, backing Almavest at 15% APY - not cheap.
But how does trust through consensus work?
The protocol has four key participants - plus Goldfinch itself, all with competing incentives: borrowers, backers, liquidity providers, and auditors:
Borrowers want money at the best possible terms
Backers want to lend with satisfying risk-adjusted returns
Liquidity providers want passive returns with minimum risk and effort
Auditors want to deliver a service and being retributed for it
Goldfinch, ultimately, wants the protocol, TVL and token price, to grow
A refresher on how lending works in traditional banking. In TradFi, banks underwrite a loan to a borrower trying to secure the safest collateral package possible - real estate, corporate fixed assets, invoices, cash, etc. Only in rare cases they provide unsecured loans, but also in those cases loans are somehow backed by an individual’s aversion to worsening his/ her credit history, and in extreme cases by legal enforcement. In other words, at the centre of everything there is always a collateral with some degree of enforceability. The safer the collateral, typically, the lower the interest rate. Banks tend to over-collateralise their loans in the same way MakerDAO builds margins of safety in its vaults. Where the magic of leverage happens is on the liability side: banks are able to lend out much more than what they own thanks to the benefits of diversification and, most importantly, to the existence of the ultimate backstop provided in several forms by the central bank.
Back to Goldfinch. We identify four different (stylised) phases the protocol tries to facilitate:
Phase A: borrowing and lending
Phase B: accruing interests
Phase C: repaying loans
Phase D: growing the total value locked
Phase A -> Borrowing and Lending
(1) Borrowers propose a term sheet. Borrowers seek financing from the protocol by proposing terms to potential backers for a specific pool. This step resembles that of proposing a term sheet to investors. In addition, in order to create such a pool, the same borrowers need to stake an amount of GFI (Goldfinch’s governance token) doubling the cost of an auditor approval - see more below.
(2) Auditors need to approve borrowers. Nine (as of today) randomly selected (although proportional to staking) auditors vote on any borrower’s legitimacy - not creditworthiness. Although voting happens on chain, auditors can do whatever they want, holistically, to evaluate a borrower. Auditors stake GFI in order to be selected for votes, and receive staking returns based on their ex post alignment with consensus. Borrowers’ staked GFIs are used to compensate auditors, and this amount gets slashed in case two auditors give negative approval.
(3) Backers evaluate the proposed conditions. Backers look at the borrower’s term sheets and evaluate all available information - both on and off-chain, holistically, and ultimately decide if and how much capital they want to supply to the pool. This capital represents the junior exposure of the pool.
(4) Liquidity providers fill the gap. An algorithmic leverage model allows for additional senior capital (subject to approval) to be provided on top of the junior tranche - such capital is provided blindly by liquidity providers that decide to act as senior lenders, agnostically and parametrically, across all pools.
Phase B -> Accruing Interest
Borrowers owe an interest rate to the pool as stated in the approved terms. This interest is proportionally allocated between senior and junior lenders, with an adjustment mechanism that (a) favours junior lenders for their higher risk, and (b) allocates part of the stream to a rainy-day fund under the control of GFI.
NAV accrues in the form of a growing exchange rate between FIDU (ERC20 token provided to liquidity provider the moment USDCs are supplied) and USDC, net of a 0.5% withdrawal fee.
Phase C -> Repaying Loans
Both backers (junior creditors) and liquidity providers (senior) receive identification NFTs when they supply capital, in order to track what they have supplied and how much has been redeemed.
Senior creditors can exchange their FIDU at any point in time, assuming there is liquidity available in the pool. Junior backers, instead, should wait for the seniors to get fully repaid first - although this isn’t explicitly stated in the whitepaper. New entrants in the pool can offer additional liquidity to redeeming investors.
Phase D -> Growing the Total Value Locked
As usual, Goldfinch has a set of incentives in place to favour virtuous activities from all parties, while getting everybody invested in the sustainable growth of the protocol. We have stated a few below:
Borrowers need to stake GFI in order to start a pool request
Part of the interest paid accrues in a general provision, under GFI control
Backers can back others with GFI, augmenting their ability to access senior capital and acting as an additional line of defence in case of default
Early and good backers get rewarded with GFI
Auditors need to stake GFI to get selected, and are compensated in GFI
GFI has both customary governance and profit distribution rights
I tried to summarise the most important interactions in a chart.
As you can see, there is a lot going on. In the absence of hard collateral, developing the checks and balances needed to avoid all possible fraudulent behaviours is complicated. Very complicated, and that’s exactly the main use of a collateral: by representing some form of social value - tangible or intangible, a collateral bypasses the construction of ultra-complex, and hence prone to error, systems.
What Are Goldfinch’s Primitives?
I tried to make some mental order and distill what are the primitive functions that Goldfinch tries to aggregate in their protocol:
Borrower legitimacy authentication: auditors authenticate potential borrowers by putting a measure of their own wealth at stake
Backer rating: backers are rated based on their perceived ability to profitably lend - their rating impact their ability to receive senior capital leverage
Lending marketplace: the protocol allows backers and borrowers to meet and potentially find each other at certain specific terms
Backer staking: liquidity providers stake backers, depending on their rating, and receive senior cash flow distributions in exchange
Cash flow distribution: the protocol enforces the trustless distribution of repayments in accordance to specific levels of seniority and set of parameters
There is something else happening off-chain that might also support some of those activities - e.g. contractual agreements between backers and borrowers or customary due diligence activities, and I am sure I have missed other elementary on-chain functions performed by the protocol.
The key question for me, however, remains whether all those primitives need to live together within the same financial building block in order to work appropriately. In the case of uncollateralised lending I guess they should, but in a system where hard collateral is in the picture a lot of those primitive functions are performed by the collateral itself. Collaterals ensure some level of borrower’s legitimacy, facilitate interactions between borrowers and lenders - given available price discovery, offer a good guarantee to senior backers, and sustain the cash flow distribution in bankruptcy remote set-ups - such as modern SPVs. Collaterals are fungible, portable, measurable, in a way that intangible consensus is not, and that makes building a scalable and composable trust platform easier and more robust.
Goldfinch is not the only project trying to follow the collateral-free route. Around the same time (late 2020) TrustToken launched TrueFi, a platform for uncollateralised TUSD-denominated loans. Although the mechanisms implemented are different, the underlying ambitions and challenges are not.
The efforts to drastically innovate trust platforms the way Goldfinch is doing are insanely ambitious and laudable, but I must admit I remain a fan of Option 2 to try and solve the collateral conundrum. I believe modern tech could allow us to dramatically expand the collateral pool by representing digitally a much larger set of stuff that have some recognised value, and that can be used as a mean to do pretty much everything. It feels neater, and more aligned with the idea of pure primitive composability of theoretical DeFi. But Goldfinch is at the very beginning of its journey - none of the backed pools has reached its maturity date yet. There is still a long time needed to assess whether the idea of innovating trust platforms via Option 1, i.e. purely through closed and complex mechanisms of competing incentives, will be successful and resilient. Considering how ambitious is the task at stake, I will be enthusiastic of being proven wrong.
Innovating is a communal effort. If you have great ideas you want to explore together, great companies that should be on Dirt Roads radar, or topics you would like to co-author on DR, please feel free to reply to this email or contact me on Twitter.