Undercollateralized Lending in DeFi: Better Yields with Better Risk Management
Credit is the crucial ingredient necessary to ignite economic growth — bringing tomorrow’s money forward to today to finance productive investments. Similarly, a DeFi credit boom is the missing piece that is needed for crypto and DeFi to finally grow into the multi-trillion dollar digital economy that builders envisioned. Established on-chain lending protocols follow the overcollateralized framework, a consequence of crypto’s anonymity — but this is unfit for the purpose of facilitating productive investments. What DeFi needs to bring in mass real-world adoption and catalyze massive growth in the sector is undercollateralized lending,
In this piece, we take a deep dive into undercollateralized lending in DeFi — the various types, the benefits and drawbacks, comparing it to both existing overcollateralized lending protocols and the trad-fi lending system. The main players in the undercollateralized lending market today, with their TVLs as of 12/19, are Goldfinch ($100m), Centrifuge ($80m), Maple ($70m), TrueFi ($29m), Ribbon Lend ($23m) and Clearpool ($11m).
What is Undercollateralized Lending?
When a borrower takes out a loan, lenders usually require the borrower to pledge some of his assets as collateral, which can be sold to recover the principal amount of loan in the event of default. A loan is undercollateralized if the market value of the assets is lower than the principal loan amount. This is desirable because it increases capital efficiency — the borrower loses access to fewer assets during the loan. In the real world, ‘reputational collateral’ and legal systems allow this to happen — defaulting on a loan may be beneficial to the borrower in the short-term but will harm his reputation and access to credit in the long-term, and he would be legally obligated to use future revenues to repay the loan. This requires borrowers to be identifiable real-world entities.
However, none of this is possible in crypto due to the anonymity of blockchain addresses. A Sybil borrower can take out an undercollateralized loan and default on it with one account, and then use another account for future activities.
Since identity and social reputation are difficult concepts to replicate on blockchains, initial lending solutions revolve around overcollateralization. Overcollateralized protocols e.g. Aave, Compound, MakerDAO — have a combined $11bn in TVL, one-fifth of the $51bn TVL of Defi (as of 12/19).
However, overcollateralized lending only supports speculative, not productive, use cases, such as trading and exposure to other tokenized assets through leverage. The demand for such loans, and hence yield, are heavily dependent on market conditions, hence making the DeFi market very procyclical and volatile. When market outlook is weak, the demand for loans, and yields, fall — now, supply rate for USDC on Aave and Compound is 1% (very unattractive compared to TradFi risk-free rate of 4.25%) and the utilization of the USDC pools has been ~40% for the past few months. In contrast, undercollateralized lending would be used to finance productive investments, leading to more stable returns, with yields being affected by macroeconomic conditions rather than volatile asset prices and speculative activity.
Furthermore, overcollateralized lending is useless for helping firms finance productive investments, which is the foundational purpose of finance in the first place. By requiring overcollateralization with liquid collateral, this reduces assets available for use, which is the opposite of what the firm intends. In TradFi, firms would like to collateralize their loans with their inventories or factories that still remain productive during the loan period. Additionally, most crypto use cases have thus far been speculative or financial. For DeFi to achieve real-world adoption, it is crucial to bring DeFi into real-world businesses. This step is symbiotic — helping real-world enterprises realize the efficiency and transparency and access the liquidity of DeFi, while giving DeFi stable, sustainable sources of yield and fostering mainstream acceptance and adoption of blockchain protocols.
Bridging Defi Liquidity and Real-World Investments
Undercollateralized on-chain lending brings together DeFi liquidity and real-world enterprises facing unmet credit demand.
Many parts of the developing world are underserved by financial services, with healthy businesses finding it prohibitively difficult to access the capital they need for growth. 22% of the global adult population is unbanked, mostly living in developing economies, and an estimated 35% of economic activity in developing markets happens in the informal economy. This is due to lack of trustworthy credit bureaus, competition for capital from infrastructure and government bond investments, and lack of enforcement of rule of law.
Hence, on-chain DeFi lending solutions that increase access to credit worldwide are a strong product-market fit that will drive real-world DeFi adoption and bring sustainable sources of yield to DeFi.
Goldfinch
Goldfinch is a pool-based lending protocol built to achieve DeFi’s goal of expanding financial access. It supplies undercollateralized loans to real-world businesses, and has targeted borrowers who would benefit from crypto loans the most — emerging market businesses. Goldfinch’s USDC capital has reached more than 1 million people and businesses across 20+ countries, from providing accessible debt to early-stage sustainability-focused agrotech companies in Mexico, to financing small business ownership in Kenya and beyond (footnote 99).
Goldfinch hence brings the attractive yields of emerging market private debt markets to DeFi. The private debt market is poised to become a mainstream asset class in the near future. Furthermore, private debt fundraising in emerging markets is growing rapidly — tripling from $2.4 to $8bn from 2009–19. Emerging markets are poised to show much higher GDP growth than developed economies in the coming decades, showing how lucrative investing in emerging market direct lending will be. Goldfinch’s Senior Pool is open to any KYC’ed individual to provide capital into (KYC is Know-Your-Customer, which is often a requirement for real-world companies to know who they are borrowing from and giving capital to), and is earning a base APY of 7.81% and total APY of 13.95% — which are among the highest in DeFi for USDC.
Other protocols that involve lending to real-world businesses include TrueFi. Other undercollateralized lending protocols Maple, Clearpool and Ribbon Lend have crypto-native firms such as market makers and quantitative trading firms as borrowers.
The major benefits of on-chain lending compared to lending in TradFi capital markets are efficiency and transparency.
On-chain lending brings huge efficiency gains — lending pools are the most cost-effective way to pool and distribute capital between a vast group of lenders and borrowers, and automated movement of funds between entities through smart contracts eliminates the time and cost of intermediaries. From the IMF’s 2022 Global Financial Stability Report, DeFi can provide up to 12% in annual savings to emerging market businesses in financing costs — due to the absence of labor and operational costs. And a report from the Bank of England claims that financial institutions spend $20b yearly in trade processing, which with blockchains, can be cut by 80%. On-chain lending protocols hence can charge a lower interest rate spread, benefitting borrowers in terms of easier access to credit and lenders in terms of higher yields.
Effectively Mitigating Risks of Undercollateralization
Undercollateralized lending protocols definitely carry more financial risk than their overcollateralized counterparts, but the question is whether such protocols are designed to mitigate excessive risk, be transparent about the risk that is present, and offer superior returns adjusting for this risk. Analyzing existing protocols, we find that some do better than others — transparency and incentive design have thus far effectively mitigated the risks, delivering superior risk-adjusted returns to date even amidst a tumultuous year for crypto.
In CeFi and TradFi, lending firms’ balance sheets and loan books are often opaque, with summary information reported quarterly — relatively infrequently. Without visibility into individual loans, it is difficult to understand the riskiness of loans and liquidity of the firm, especially as firms can hide looming issues from reported numbers with accounting tricks. When lenders are unaware of the risk of their borrowers, this leads to mispricing of risk, which can result in catastrophic contagion when market conditions deteriorate. The current situation with Genesis and FTX is a prime example — it initially insisted that its lending business was unaffected by the FTX collapse, but then halted withdrawals two days later and is teetering on bankruptcy due to liquidity issues.
In Maple, Pool Delegates are entities who underwrite and issue loans. They have full discretion over how to invest their customers’ deposited funds in order to achieve their target returns. Hence, individual lenders/liquidity providers trust Pool Delegates to do proper credit underwriting. Pool Delegates operate under full transparency — reporting all the loans, loaned entities, interest rates and maturities that are on their loan book. In Maple’s case, Pool Delegates are required to publish a monthly pool report documenting the active loans and the performance of their borrowing companies.
In Goldfinch, any individual who passes a Unique entity Test (KYC) can be a Backer — effectively a loan underwriter — gaining access to financial information and loans proposed by prospective lenders and being able to communicate with them. Backers supply first-loss capital in Borrower Pools.
In TrueFi, there are permissionless lending pools (called DAO pools) and permissioned pools managed by portfolio managers (called TrueFi Capital Markets). In DAO pools, TrustToken (the parent company) conducts a credit due diligence analysis, with the information kept confidential to TrustToken. Stakers of the TRU token (stkTRU) then vote on loan applications to decide which lending pools are funded. This offers a lower level of transparency than Maple and Goldfinch — hence this shows that it is not a given that all undercollateralized lending solutions have greater transparency (they only have the potential to — due to everything being done on-chain)
The protocols also use incentive design to coordinate loan underwriters’ incentives to select high-performing loans — though the extent to which it successfully acts as an incentive is debatable.
In Maple, Pool Delegates must provide first-loss capital, when they establish the pool. The intent is for this to disincentivize Delegates from making poor underwriting decisions as they have skin in the game. The recent Maple 2.0 update has changed it such that only Pool Delegates provide first-loss capital, which is now denominated in the pool asset, better aligning incentives.
Goldfinch adopts a tranching system, where Backers provide first-loss capital into the junior tranche or the Borrower Pool, while Lenders in the Senior Pool are paid out first in the event of a default. Another interesting aspect is the Approval Votes that Auditors do when verifying potential borrowers: if the majority votes Yes, those who vote No are slashed, and vice versa, hence encouraging Auditors to make carefully-considered decisions about Borrowers.
Some protocols use default insurance as a risk-management feature instead of first-loss capital. Here, a portion of protocol interest revenues is directed to an insurance fund which would help to cover losses from default. For Clearpool and Ribbon Lend, the percentage is 5% (Ribbon Lend is a fork off Clearpool). However, this only provides minuscule protection — for Clearpool, the highest principal loan coverage resulting from the insurance fund is 0.25%, and it is concerning to note that this is Ribbon Lend and Clearpool’s only source of protection for regular lenders (there is no first-loss capital) — hence all lenders are potentially exposed to big losses in the event of default.
One potential risk of undercollateralized lending protocols is exit liquidity, a consideration that is extremely important during times of distress and panic. As we saw with the ongoing FTX saga, it matters greatly in which order users are able to withdraw when funds are limited due to default. Sudden upsurges in demand for exit liquidity can also lead to illiquidity in the protocol.
Maple Finance handles these risks excellently — it institutes a lockup period (flexible pool parameter currently set to 90 days). Additionally, one needs to trigger a withdrawal, which leads to a 10 day cooldown period, after which the lender has 2 days to withdraw funds. Hence, Pool Delegates have visibility up to 10 days in the future about impending liquidity obligations which gives time to raise cash in a mass withdrawal situation. Additionally, this eliminates any race conditions (i.e. first-come-first-serve withdrawals when borrower repays) associated with no lock-ups which would massively disadvantage regular users compared to sophisticated MEV bots.
Pool Delegates also structure pools to have loan maturations evenly distributed across time to ensure capital efficiency in normal times — the fact that ~20% of their loans mature in 10–20 days, hence this is comforting as it shows that some care is taken to ensure constant stream of incoming cash flows to meet withdrawal liquidity.
That said, there have been times where exceptionally high withdrawal demand led to extended cooldown periods as long as 75 days — however, Pool Delegates and Maple proactively communicate in such situations and this provides more confidence. Furthermore, liquidity crunches are part and parcel of every undercollateralized lending enterprise, which even banks are not immune from. Banks have built-up equity buffers to protect against illiquidity, and so we argue that with time to build confidence and capital buffers, DeFi lending businesses will grow to offer customers effective risk mitigation strategies both protecting against losses and ensuring exit liquidity.
In contrast, Ribbon’s design lends itself to exit liquidity risks. They have marketed their no-lockup, instant withdrawal design as a boon for users, however in times of illiquidity, this can leave regular users stuck, as withdrawals are processed on a first-come-first-serve basis, hence any newly-returned liquidity from borrowers would instantly get removed by MEV bots.
Additionally, without a lock-up or cool-down period, there is huge uncertainty on the borrower’s side about how much liquidity it will potentially have access to in the future. At any moment a sudden mass of withdrawals could push the deposit rate to 99% and force a borrower to repay their loan or face default or uncomfortably high interest rates (due to the utilization curve). This would severely impact their operations especially in times of heightened volatility and uncertainty.
Finally, undercollateralized lending protocols have had an impressive track record event amidst the tumultuous year for crypto, which include a bear market and Cefi contagion.
Default rates to date are low: Maple: 2.3%, TrueFi 0.20%, Clearpool 0%, Goldfinch 0%. Maple recently suffered $36m of default due to the insolvency of borrowers Auros Global and Orthogonal Trading in the wake of FTX’s collapse, and TrueFi has had $4m of defaults. One can even argue that the history of these protocols is unfavorably biased as it has not included a crypto bull market.
Bond pricing theory suggests that S = p(1-R), where S is the credit spread (interest rate differential between undercollateralized loans vs the risk-free rate — i.e. overcollateralized platforms like Aave and Compound), p being the probability of default, and R is the expected recovery amount (which we assume to be 41%, the equivalent percentage for corporate bonds according to Moody’s data from 1920–2009). Taking Maple, with the highest default percentage of 2.3%, we would expect the credit spread to be only 1.5%, far lower than the observed ~7% credit spread we see (0.75% Aave USDC APY vs Maple’s 8% APY before rewards).
On a call with Maven11, pool delegate for Maple, they mentioned that they underwrite loans with the assumption of a low default rate of 1–3%. Even with a default rate of 3%, the expected credit spread would be ~2%.
Comparing this with TradFi, the credit spread is 7%, whereas the High Yield Corporate Bond credit spread is 4.8%, having hit a high of 5.9% in July, and the historical non-recessionary default rate for High Yield bonds is 2.2% from Fitch.
Hence, we can see from the short history of undercollateralized lending protocols that they offer extremely attractive risk-adjusted returns when compared to credit protocols as well as TradFi bonds. This remains true despite the recent FTX crisis. Nevertheless, there could be an increased risk premium stemming from smart contract or protocol risk as these undercollateralized lending protocols are relatively newer and unproven compared to well-known lending protocols Aave and Compound.
Innovation in On-chain Infrastructure that Supports Undercollateralized Lending
In order to achieve effective on-chain undercollateralized lending, there are several pieces of infrastructure that are key — on-chain credit scoring and on-chain identity. Several promising initial solutions have been innovated thus far.
Firstly, on-chain credit scoring has the potential to transform the world’s credit assessment infrastructure. The current system is opaque and run by a handful of companies, and credit scoring agencies may not even be accessible in developing countries. As a result, entire demographic categories have been excluded from credit.
Since the entire blockchain transaction history is immutable and freely available, and going by the thesis that an increasing proportion of people’s real-world assets will be tokenized and represented on-chain, on-chain credit scoring will prove to be a more transparent, superior solution. Some examples of on-chain credit scoring protocols are Cred, Spectra and Credora — which all use a combination of payment history, liquidation history, debt, repayment history, assets, credit utilization and other factors to assess creditworthiness.
Credora is the established credit-scoring solution for some of the major undercollateralized lending protocols. Maple’s pool delegates like Maven use Credora for live monitoring of borrowers’ on-chain assets and liabilities — which is important as borrowers are crypto-native market-makers and hence have the bulk of their assets on-chain. Credora aggregates transactions and positions from most major exchanges and is confident in its ability to track at least 85–90% of all borrower’s trading positions. Credora, like other on-chain credit scorers, utilize zero-knowledge proofs to allow loan underwriters to gain insights into borrowers’ balance sheets without giving up sensitive details such as specific positions on exchanges. For Ribbon Finance, Credora’s credit assessment takes a 50% weighting in the credit evaluation of a potential borrower, alongside traditional due diligence processes.
While Credora currently has an established position, its creditworthiness algorithms are proprietary and opaque, and furthermore, the centralization of on-chain risk-scoring on Credora’s algorithm could be a risk vector. One can think of further decentralizing credit-scoring algorithms. Spectral pioneers a distributed credit risk modelling system where credit risk modeller can submit their models to the marketplace to become part of the Multi-Asset Credit Risk Oracle (MACRO) score. Having raised $32m in Series B funding, Spectral launched its open beta in September. Though it has not announced any partnerships with lending protocols yet, it is hoping to target consumer loans and solutions like this will definitely have a crucial role to play in the undercollateralized lending of tomorrow.
Secondly, Soul-Bound Tokens are a recent attempt to solve the identity problem in crypto. Soul-bound tokens are non-transferrable — hence bound to one address, and publicly-verifiable. Hence, this information about an individual, (or “soul”) such as credentials, membership affiliations, and credit history, are stored on the blockchain — representing an individual’s reputation and identity.
Hence, the concept of SBT lends itself well to Know-Your-Customer (KYC). In TradFi, KYC checks are protective measures to mitigate fraud and identity theft, as they prove that an on-chain identity is mapped to a real-world individual or entity. Several centralized exchanges, such as Binance, which are required to uphold KYC regulations, have already experimented with this concept. Binance launched the ‘Binance Account Bound (BAB)’ token which acts as a certification of a wallet owner’s verified user status on Binance.
Goldfinch is an example of an undercollateralized lending protocol using SBT and KYC in its business. Goldfinch relies on a decentralized group of auditors to approve borrowers for the protocol to consider — at every vote, a random group of 9 auditors are selected. Hence, it is imperative that all auditors are provably unique individuals to prevent Sybil attacks — and Goldfinch hence uses KYC checks to provide each auditor with a Unique Identity (UID) SBT. Goldfinch partners with identity management services Persona to manage KYC processes and data, and Parallel Markets to manage KYB and accreditation processes. Goldfinch SBTs can be used by other protocols, hence another other protocol can build on top of this without building KYC flow or handling personal data — this is the composability benefit to DeFi that SBTs hope to unlock.
One potential downside to the composability of SBTs is the centralization of trust in a single centralized verification process. For example, all lending protocols could trust Goldfinch’s KYC processes and integrate their KYC checks instead of creating their own workflows. If lapses are somehow found in Goldfinch and their vendors’ KYC processes, this could put all other lending protocols at risk.
Another argument against SBTs is that there is still a risk that a malicious actor can default on loans, create a new account and start afresh. SBTs are a new area of innovation and I believe that with proper integration with KYC processes and design of protocols, SBT misuse can be made financially prohibitive, if not impossible.
Despite concerns that KYC and on-chain identity could disrupt the privacy and trustless ideals of DeFi, I argue that KYC requirements are complementary to undercollateralized lending protocols and the sacrificed privacy and anonymity are worth the immense benefits of transparency and efficiency that on-chain lending can provide. The future of DeFi is likely to be permissioned DeFi — with on-chain identities linking to real-world entities. Large-scale adoption of DeFi can only happen with the participation of large real-world businesses and institutions. Currently, institutional players are interested in deploying large capital to the ecosystem but are deterred as they need to follow strict compliance processes. Additionally, they are deterred by regulatory uncertainty — greater crypto regulation is impending in EU and US and there is a lack of standardization of security and identity processes among DeFi protocols.
Lending for Crypto-Native Entities (DAOs)
While lending protocols built on KYC work well for real-world businesses, DAOs are an increasingly important demand driver for lending that require an alternative solution.
DAOs are organizations with no centralized leadership, where participants interact with each other as partners to achieve common goals and decisions are made by governance votes. There are myriad use cases for DAOs: many top DeFi protocols e.g. lending, market makers and exchanges, are becoming decentralized and collectively managed by DAOs. DAOs can be used for crowdfunding, investing in projects, and potentially even providing any specialized service such as legal, data analytics, engineering work or market research.
DAOs can essentially be thought of as the businesses of the web3 world. Naturally, DAOs have revenue streams and expansion plans and many would like to access credit to accelerate their investments and business growth. However, DAO members often remain anonymous, and hence compliance with KYC/AML policies is impossible — putting TradFi lending and the aforementioned KYC-based undercollateralized lending protocols out of their reach.
Currently, DAOs only raise capital through token distribution, which is akin to equity issuance in TradFi. In contrast, most TradFi firms incorporate debt into their capital structure, with the average company’s capital structure being 70% equity and 30% debt. Debt issuance can be a more effective method of raising capital, especially during bear markets because it would be undesirable to sell governance tokens at low prices and dilute existing token-holders. Additionally, during the bear market, DAOs have been looking to make use of their treasuries — MakerDAO recently approved a proposal to move $1.6m of its USDC into Coinbase Prime.
Since DAOs are on-chain, digitally-native entities with assets and liabilities on-chain, one can naturally conceive of on-chain DAO-to-DAO lending. However, infrastructure for DAO-to-DAO lending remains undeveloped.
DebtDAO is aiming to build new debt smart contract primitives enabling collateralization with future revenue streams. Through DebtDAO’s Line of Credit contract, DAOs and Defi protocols with on-chain revenue can finance operations and other dynamic funding needs through loans backed by their revenue. DebtDAO’s Spigot contract enables secure loans to be automatically repaid by borrowers in a trustless manner.
Porter Finance (now Arbor Finance) is another protocol which aims to help DAOs raise debt in stablecoins while using their native tokens as collateral. In June it issued the first DAO-to-DAO bond with Ribbon DAO for $32m USDC.
Credit experts such as Ben Forman from ParaFi (Bell Curve podcast) sees much room for interesting innovations in on-chain debt contracts, such as customized loan covenants (e.g. triggers default if an on-chain metric such as EBITDA or leverage ratio crosses a certain threshold) and repayment mechanisms.
Hence, as we see more economic activity move on-chain and more DAOs become viable enterprises, I believe that DAO lending, supported by a new class of on-chain debt primitives.
Conclusion
In conclusion, undercollateralized lending offers a viable path forward for crypto and DeFi to gain real-world adoption. By enabling DeFi liquidity to finance productive real-world investments, especially in historically unbanked populations, the increased transparency and efficiency of blockchain-based economic activity will bring many benefits to the world. While the quality of risk mitigation strategies varies among different protocols, this is natural for such a nascent industry, and we are confident that the market will quickly converge on best practices. The gradual movement of economic activity and assets onto the blockchain will further increase the importance of on-chain lending, especially supported by innovations in on-chain identity, credit scoring, and lending between on-chain entities.
This research piece was done during my time in the FranklinDAO Research Team. Thanks to Erik Zhang and Cindy Jiang, my amazing team leads, for the guidance throughout the research process. For more works by my teammates, check out our Substack. If you liked this article or have any comments, do follow my Twitter and LinkedIn and reach out to my personally!
