An on-chain loan generally requires posting more value than is borrowed. The excess is not conservatism for its own sake but the only substitute available for identity and enforcement.
There is no borrower to pursue
A traditional lender extends credit against a legal claim on a known person or business, backed by courts, credit records and the ability to garnish future income.
A protocol has none of that. It sees an address, no name, no jurisdiction and no way to recover anything the address does not already hold.
Excess collateral converts a credit decision into a mechanical one. The loan is safe because the asset is already in the contract, not because the borrower is trustworthy.
The cushion is sized by volatility
Each accepted asset gets a ratio determining how much can be borrowed against it, and the ratio tightens as the asset's price moves become larger and faster.
Stable, deep assets support borrowing close to their value. Thin, volatile tokens support far less, because their price can travel a long way before anything can respond.
The buffer must cover the worst plausible move during the time it takes to notice and act, which is why the sizing question is really about speed.
Price feeds add their own delay
Protocols read prices from oracles that update on an interval or when a movement threshold is crossed. Between updates, the contract is working from a stale number.
That lag has to be priced into the cushion, since a position can be underwater in the market while still appearing healthy to the contract.
Feeds that update more aggressively allow tighter ratios, which is one reason major assets with reliable pricing get more favorable terms than obscure ones.
Unwinding costs eat into the margin
When a position breaches its threshold, someone must buy the collateral and repay the debt, and they will only do so if there is a discount worth their trouble.
That incentive is paid out of the borrower's collateral, so the buffer must be large enough to cover both the price gap and the liquidator's cut.
Market depth matters here too. Selling a large amount of a thin asset moves its price, so the required cushion grows with position size relative to available liquidity.
Attempts to lend on less
Undercollateralized lending on chain exists but reintroduces the missing ingredient in some form, whether reputation systems, whitelisted institutional borrowers or off-chain legal agreements.
Each of those trades openness for enforceability, which is the same trade traditional finance made long ago rather than an innovation specific to this market.
The persistence of the overcollateralized model reflects that constraint. Where nothing can be pursued afterward, everything has to be held in advance.