Decentralized finance (DeFi) borrowings are inherently a lot more fluid than traditional borrowings. Multiple loans may be made with no fixed repayment schedule, all of which can then be repaid either at once or through multiple repayments of varying amounts over an extended period.
Borrowings are heavily collateralized (e.g., up to 150 percent) and the collateral can be liquidated should the borrowing increase or collateral values fall such that the collateral ratio falls below the prespecified threshold. Liquidation proceeds are used to repay the loans, with any post-liquidation balance remaining with the borrower. As a result, borrowers are extremely conservative in their borrowings.
As it stands currently, there is no accepted credit-risk assessment methodology for DeFi lending. Little is known about the borrowers outside of their crypto wallets. As a result, where credit is extended the loans are heavily collateralized; a situation that is arguably highly capital-inefficient. Traditional credit risk assessments are usually centered around default events (Probability of Default/Exposure at Default/Loss Given Default). We propose that liquidations can be used as a proxy for default in the traditional sense to enable a better assessment of DeFi borrowers and lower collateral requirements.
Ideally, one should use some form of a risk assessment methodology to reduce the collateral requirements, just as credit scoring is used for mainstream retail credit to provide reputational capital.
In traditional finance, default occurs when the terms of a credit agreement are breached. For bonds, it is any missed payment; in other instances, there is some latitude. For most retail lending, default occurs once the days-past-due status exceeds 90 days or any other event that suggests elevated risk levels, the potential for loss, and a need for mitigating actions.
The accepted wisdom is to focus on borrowers’ behavior and ignore lenders’ actions, including collection activities and liquidation. Lenders’ main focus is to limit losses, but losses (bankruptcies, distressed liquidations, write-offs) may be rare, and extreme actions may have been taken to limit them.
Hence, the focus on defaults increases the number of cases for analysis and eliminates the potential noise of collections activities.
Where liquidations do occur, they may be voluntary or distressed. Voluntary—when shareholders sell assets for various reasons, possibly due to a stressed situation, but without pressure from external parties. Distressed—when pressures arise from outside. Missed payments to creditors are a typical reason, but other risk factors may also be at play (e.g. having negative equity). Distressed liquidations go hand-in-hand with bankruptcy proceedings. For individuals, bankruptcy precedes distressed liquidation. If the realized value is insufficient or there is no collateral (unsecured lending) borrowers may still be liable for the balance—with the potential seizure of other personal assets.
It has been argued that credit scoring cannot be applied to DeFi because there is no default event, but it may be possible to use liquidation as a proxy. Liquidation is the equivalent of a write-off in traditional lending, excepting the collateral value is realized then and there. If the traditional PD/EAD/LGD framework is converted into PL/EAL/LGL the loss given liquidation (LGL) is almost always zero due to the high collateralization levels.
Therefore, despite the concept of default being absent in DeFi, we can potentially go a step higher in the severity chain. Defaults are a precipitating event for bankruptcy and distressed liquidations; most credit scoring is based on defaults due to their greater numbers and lack of noise arising from mitigating actions. That said, Edward Altman developed his Z-Score models for bankruptcies, and distressed liquidations have been used to validate scoring models where credit data is sparse [Anderson 2019].
Although it can be argued that liquidation events in DeFi are imperative given the inherently volatile nature of the crypto markets (so far, at least!), however, we believe that they hold sufficient value to be used for risk modeling. Liquidation events imply that the borrower i) was too aggressive in their borrowings, ii) had insufficient headroom to handle market fluctuations and/or iii) did not take the necessary actions to prevent liquidation when the event was imminent.
As a closing point, we do not consider liquidation events in isolation when predicting the probability of liquidation, and thereby its MACRO Score. Instead, we also consider specific additional data points as part of the overall MACRO Score methodology, resulting in a more robust credit scoring model with a higher discriminatory power.
Credit Intelligence and Modelling: Many Paths through the Forest, Raymond Anderson 2019