The Undercollateralization Problem
Crypto, Credit, and the Future of DeFi Lending
DeFi lending has done something impressive: it has built a multi-hundred-billion dollar credit market that almost nobody actually needs. I’ve been trying to understand why on-chain lending looks the way it does, and the more I read, the more I think overcollateralized lending is a solved problem and also probably the wrong problem. Below is my attempt to work through why the current design feels exhausted, why undercollateralized credit seems like the real prize, and what might be done about the fact that every attempt at it so far has ended badly. I’m still learning this space, so treat this as thinking out loud rather than a settled view.
Overcollateralization Is a Weird Product
Start with first principles. Collateral makes sense when the asset you are borrowing against is illiquid - mortgages, artwork, private stock, your car. You can’t easily sell 3% of your house to cover this month’s expenses, so you pledge the whole thing and borrow against it. Fine. Rational.
Overcollateralizing liquid assets is much stranger. If I post X of $ABC to borrow
Y < X of $DEF, why didn’t I just swap Y worth of $ABC into $DEF and keep the rest of my $ABC? When I actually sat with this question, the honest answer seems to be that most on-chain borrow-lend activity is a synthetic way to short one thing and go long another (usually going levered long your own collateral), or a way to access cash without triggering a taxable sale. That’s a real market, and Aave and others earn real fees serving it. But it’s fundamentally a trading product for people who already have money, dressed up in the language of credit.
Now look at what credit means for normal people. Of the roughly $4.5T in outstanding non-housing consumer debt in the US, roughly $3T - about two-thirds is revolving credit (mostly unsecured credit cards) or student loans. Strip out the obviously collateralized stuff like housing and auto, and the majority of consumer credit in the real economy is un- or undercollateralized. The entire point of a credit card is that you get purchasing power before you have the assets. Overcollateralized DeFi inverts this: you only get purchasing power if you already have more assets than you need.
Why It Keeps Failing
The core problem is simple: there is nothing stopping a borrower from taking your money, sending it to a fresh wallet, and declaring insolvency.
Traditional finance solves this with two mechanisms that crypto structurally lacks:
Persistent identity. I can default on Chase and walk into Wells Fargo, but Wells can pull my credit history using my SSN and price me accordingly. My financial reputation follows my legal person around whether I like it or not. On-chain, abandoning my identity and generating a new one costs nothing.
Legal recourse. If I default hard enough, someone with a court order eventually shows up. The state is the ultimate liquidation engine.
The 2021-22 cohort of undercollateralized protocols (Maple, TrueFi, Goldfinch) tried to bootstrap this by importing off-chain trust: whitelist institutional borrowers, run manual underwriting, rely on real-world legal agreements for recourse. This was a reasonable design, and it also mostly blew up when it turned out that “reputable trading firm” was doing a lot of unsecured load-bearing work during the Three Arrows era. Even where it worked, it wasn’t really on-chain credit. It was traditional private credit with a blockchain settlement layer, inheriting all the customer acquisition costs, underwriting costs, and legal overhead of the incumbents it was supposed to replace. The blockchain was a cool technical feature that didn’t change the key driver of the business.
So What Would Actually Work?
Three broad directions I’ve come across, in ascending order of how interesting they seem to me.
1. Wallet reputation / on-chain credit history. The idea: the older and more active your wallet, the better your rates, so nuking a wallet with years of accumulated history becomes genuinely costly. This feels directionally right because it converts identity from free to expensive, but it also seems gameable. Reputation farming is just sybil farming with extra steps, and the maximum loan size the system can safely extend is bounded by the cost of manufacturing the reputation. That might still support a real market in small-dollar, short-duration credit where the value of the reputation exceeds the loan, but it probably caps out well below credit-card economics.
2. Off-chain identity bridges. KYC the borrower, bind the wallet to a legal person, pursue defaults in court. This works, but at that point you’ve rebuilt a fintech lender with worse operations. The interesting question is whether the cost of legal recourse can be driven down enough that the on-chain rails actually matter.
3. ZK credit attestations. This is the one I find most exciting, though I’m still wrapping my head around the technology. Quick explanation for anyone new to it like I was: a zero-knowledge proof (ZKP) is a cryptographic technique that lets you prove a statement is true without revealing the underlying data behind it. The classic example is proving you’re over 21 without showing your ID - the bouncer learns the one fact they need (yes, over 21) and nothing else (not your name, address, or exact birthday). The number guarantees you couldn’t have produced the proof unless the statement was actually true.
Applied to credit, the primitive would be: a proof that I control X of assets across wallets, exchanges, and bank accounts, or that I’ve repaid Y loans on time, or that my income is at least Z - without revealing which wallets, which accounts, or who I am. There’s a newer technique called zkTLS that lets you generate proofs about your regular web2 accounts (bank balances, payroll deposits, brokerage holdings), so a lender could verify “this person’s checking account has averaged $5k for a year” without ever seeing the account. Pair that with something called a nullifier - basically a cryptographic tag that stops the same underlying person from minting multiple fresh credit profiles - and you start to get something that rhymes with a credit bureau, except the borrower holds their own data and the lender only sees the proof.
The privacy part is nice, but the efficiency part might be the actual business: one proof, verified cheaply by every lender in the market, instead of every lender independently re-underwriting the same borrower. A credit bureau is, if you squint, a very inefficient version of exactly this, plus a lobbying arm and periodic data breaches.
The open problems are real, though. Sybil resistance ultimately still needs some scarce identity anchor (biometrics, government ID, a web of trust). Attestation sources can be spoofed or revoked. And someone still eats the loss on early defaults while the system calibrates - a proof tells you the borrower was good for it in the past, not that they will be in the future.
In Conclusion
Overcollateralized lending is a fine business and a boring frontier. The unsecured two-thirds of consumer credit is the actual market, and the blocker is that crypto has no native answer to the questions “who are you, and what happens if you don’t pay me back?” My tentative view, and I hold it loosely, is that the answer gets assembled from ZK attestations over off-chain financial data plus some scarce identity anchor, and that the eventual winner looks less like a lending protocol and more like a privacy-preserving credit bureau that lending protocols plug into.
Credit for people who don’t already have the money is the oldest and largest financial product on earth. Whoever brings it on-chain in a way that actually protects lenders will have built something enormous.