Authored by Hamlin Au and Krys Ziemski
For any crypto-backed lending product, the most significant intellectual and operational challenge is developing a sophisticated underwriting framework that can accurately price risk in a volatile and novel asset class. Traditional underwriting models are a necessary but fundamentally insufficient foundation. A modern, defensible approach requires augmenting these models with crypto-native data, real-time monitoring, and advanced predictive analytics.
The traditional "3 C's" of credit—Character, Capacity, and Collateral—are often ill-suited for the typical crypto user. Many early adopters and digitally native individuals lack extensive credit files, making "Character" difficult to assess through FICO scores. Their income may be non-traditional (e.g., from staking rewards, airdrops, or freelance work), making "Capacity" difficult to verify through traditional pay stubs. This often forces an over-reliance on "Collateral," which in the crypto world is subject to extreme price volatility, a risk factor traditional models are not designed to handle.
The key risks extend beyond simple borrower default to include platform risk (the risk of the exchange being hacked or becoming insolvent), counterparty risk (the risk of a lender rehypothecating collateral), and the unique risk of the collateral asset itself becoming illiquid or worthless—a "coin death" event.
The solution is to construct a Hybrid Crypto Underwriting Model that layers crypto-native risk management on top of a traditional foundation. This framework leverages new data sources and technologies, including cash-flow underwriting, on-chain analytics, and AI/ML-powered predictive models.
Layer 1: Real-Time Collateral Management (The Foundation)
This is the first and most important line of defense against market volatility.
Collateral Quality Scoring: An internal, dynamic scoring system must be developed to assess the quality of acceptable collateral. This is not a static list. The score for each asset should be based on quantifiable metrics such as market capitalization, 24-hour trading volume (liquidity), the number of reputable exchange listings, and historical volatility. Only assets that meet a minimum quality score (e.g., BTC, ETH, and fully-backed, regulated stablecoins) should be accepted as collateral.
Dynamic Loan-to-Value (LTV) Ratios: A tiered LTV system must be implemented, assigning a maximum loan amount as a percentage of the collateral's value based on its quality score. For example, a USD-backed stablecoin might receive a 90% LTV, while Bitcoin receives a 50% LTV and a more volatile altcoin receives a 30% LTV. These LTVs should be dynamically adjusted based on real-time market conditions.
Automated Margin Call & Liquidation Engine: This is the critical, automated risk mitigation mechanism. Using smart contracts and reliable price oracles, the system must continuously monitor the value of the collateral in real-time. If a market downturn causes the LTV to breach a predetermined threshold (e.g., a loan with a 50% initial LTV might have a margin call threshold of 70%), an automated notification is sent to the borrower. If the borrower fails to post additional collateral or repay a portion of the loan within a short, clearly defined grace period, the system must be empowered to automatically liquidate a portion of the collateral on the open market to bring the LTV back to a safe level.
Layer 2: Augmented Borrower Assessment (The Enhancement)
This layer enhances the traditional "Character" and "Capacity" pillars with new data sources.
Layer 3: Predictive Analytics & AI (The Future-Proofing)
This is the most advanced layer, using machine learning to forecast risk.
AI-Powered Probability of Default (POD) Model: Train machine learning models (such as CART decision-trees or neural networks) on the rich datasets collected from Layers 1 and 2. By combining on-chain behavioral data, off-chain cash flow data, and collateral characteristics, these models can generate a highly accurate POD score for each borrower, allowing for more precise risk-based pricing.
"Coin Death" Probability Model: A unique risk in crypto-lending is that the collateral itself could fail. The underwriting model must incorporate a sub-model to assess this risk. This can be based on academic models like the Zero-Price-Probability (ZPP) model, which uses time-series analysis to forecast the probability of an asset's price going to zero. The output of this model should be a key input into the Collateral Quality Score in Layer 1, ensuring that riskier, less established assets are automatically assigned lower LTVs or deemed ineligible.
Implementing this Hybrid Crypto Underwriting Model is not merely a technical exercise; it is a strategic imperative for any Web3 platform seeking to offer compliant, resilient, and profitable consumer credit products in the evolving digital asset landscape. It allows for a more comprehensive and accurate assessment of risk, enabling broader access to credit for crypto users while safeguarding the lender against the inherent volatilities of the market.
Works Cited