Ration or Price: The Choice Every Credit Market Makes | HL Hunt
Ration or Price: The Choice Every Credit Market Makes
A lender facing an applicant it considers risky has two options: charge more, or refuse. Every credit market resolves that choice somewhere, and the resolution is driven far more by price limits, product conventions, and what a lender is willing to be seen charging than by anything economic. What makes the choice consequential is that the two options produce completely different kinds of harm. Expensive credit creates an identifiable person paying an identifiable amount. Refusal creates someone who didn't get a loan, whose subsequent difficulties get attributed to their circumstances, and who appears in no dataset at all. One error is measurable and the other isn't — which is why the argument about which is worse never ends.
In this report
The choice
| Rationing | Pricing | |
|---|---|---|
| Mechanism | A threshold; decline below it | Charge more as risk rises |
| Who is served | Those above the line | Almost everyone, at a price |
| Error made | Excluding people who would have repaid | Charging people who didn't need to pay that much |
| Harm is | Invisible | Visible and itemized |
| Who complains | Nobody, mostly | Borrowers, advocates, regulators |
| Appears in data as | Nothing | A price, a balance, a default |
Both are ways of handling the same fact — that some applicants cost more to serve than others. Neither eliminates the underlying risk; they allocate who bears it. Rationing places it entirely on the excluded, who bear the cost of not having credit. Pricing places it on the included, who pay more.
And the bottom row is the one that determines everything downstream. A rationing error leaves no record because the thing that would have generated the record didn't happen.
Neither approach eliminates the risk. They differ in who bears it, and in whether anyone can see that they did.
What actually decides it
Not, mostly, an assessment of which serves borrowers better.
- Price limits. Where a cap binds, applicants whose expected cost exceeds it cannot be served profitably — so a cap converts a pricing market into a rationing one at the margin, mechanically.
- Fixed costs. Per our selection analysis, small loans carry fixed costs that a permissible price may not cover regardless of risk — which produces rationing by amount rather than by risk, and is why small loans are scarce even for good credits.
- Product convention. Some products are conventionally single-price; charging a wide range is not how they're sold.
- Reputational limits. A lender may be able to charge a price it isn't willing to be seen charging.
- Operational cost of pricing. A market with many price points needs more infrastructure than one with a cutoff.
- Regulatory attention. High prices attract scrutiny; declines mostly don't — which is an incentive to ration that nobody states.
The last point is worth dwelling on because it's a direct consequence of the visibility asymmetry. A lender choosing between charging 39% and declining faces a real difference in regulatory and reputational exposure, and the difference doesn't correspond to any difference in how well borrowers are served.
Which means the ration-or-price line in any given market is substantially an artifact of what is easy to observe and object to.
The visibility asymmetry
The core of the report, and it's an instance of a structure this desk has documented repeatedly.
Our measurement analysis established that when a decision forecloses its own alternative, the foreclosed outcome generates no data and is therefore treated as costing nothing. Rationing is that structure operating at the level of an entire market rather than a single lender.
Compare what each error leaves behind:
| Someone charged too much | Someone wrongly refused | |
|---|---|---|
| Record exists? | Yes — an account, a price, a payment history | An application record, and nothing after |
| Harm quantifiable? | Yes, to the dollar | No |
| Can be interviewed? | Yes, and identified as a class | Not identifiable as a class |
| Subsequent difficulty attributed to | The loan | Their circumstances |
| Generates advocacy? | Yes | Rarely |
The fourth row is the one that does the most work and gets noticed least. When someone who was refused credit subsequently loses their housing, nobody records the refusal as a cause. The eviction is attributed to their financial situation — which is true and incomplete, and there's no mechanism that would ever connect the two.
So the two harms are not merely differently visible. One of them is systematically misattributed to something else, which means even a determined attempt to count it would find the events filed under other headings.
Our reject inference analysis and override analysis both find, from lenders' own data, that near-cutoff declines perform far better than the models predicted — which is direct evidence that rationing errors are real and substantial. That evidence exists only because a small number of lenders accidentally generated it through manual overrides.
What a cap does
Work the mechanism, because it's the concrete case where the choice is made deliberately.
A cap sets a maximum price. Its effects are two, and both are real:
It prevents the highest prices. People who would have paid above the cap and are still served pay less. That's a direct, quantifiable benefit to an identifiable group.
It converts pricing into rationing at the margin. Applicants whose expected cost — funding, losses, and the fixed costs from our verification analysis — exceeds the cap cannot be served profitably, so they're refused.
Stylized illustration of the boundary:
| Applicant | Expected cost to serve | Under a cap at 36% |
|---|---|---|
| A | 14% | Served, and cheaper if they'd been charged more |
| B | 31% | Served, near the boundary |
| C | 44% | Refused |
| D | 70% | Refused |
The question the cap debate turns on is what happens to C and D, and it's the question nobody can answer, because C and D stop appearing in the data at the moment the cap takes effect.
Note also what the fixed-cost point does here. A cap expressed as a rate binds hardest on the smallest loans, because fixed costs are a larger share of a small amount — so a rate cap rations by loan size as much as by borrower risk, and the smallest borrowers are excluded first regardless of their creditworthiness.
This report deliberately takes no position on where caps should sit. The position it does take is that the debate is conducted with one side's evidence and that this is a fixable measurement problem rather than an irreducible disagreement about values.
What people do instead
The determining question, and what little can be said about it.
Someone refused credit for a genuine need does something. The possibilities, roughly ordered by how well they're documented:
- Goes without, and absorbs the consequence — per our time preference analysis, that consequence is frequently a step-change rather than a marginal cost.
- Borrows from family, converting a commercial obligation into a relationship one.
- Uses a product outside the cap's scope — the substitution our definitional analysis describes, where the same economic transaction is restructured to fall outside a category.
- Depletes assets, including retirement funds at the cost our liquidity analysis computes.
- Uses an unregulated or illegal source.
- Doesn't need it after all, which is the outcome that would vindicate the refusal.
The third is the one that most complicates cap evaluation, because it means a cap can reduce measured high-cost lending while the underlying transactions continue in a different legal form. Our definitional analysis argues that category-based regulation generates this substitution structurally rather than occasionally.
And the honest position: the mix across these outcomes is not known, varies by market, and determines the answer. A cap whose refused applicants mostly land in the last row is a good policy; one whose refused applicants mostly land in the fourth or fifth is not. Both claims are made confidently and neither is usually supported.
Where markets do both
Real markets aren't purely one or the other, and the hybrid forms are where most consumer credit sits.
- Priced within a band, rationed outside it. Risk-based pricing operates up to a ceiling, and below that quality applicants are declined — the standard arrangement.
- Rationed on amount rather than risk, which is what line assignment does. Per our line analysis, a small line is a partial refusal that nobody records as one — the customer was approved, and the harm of an inadequate limit is as invisible as the harm of a decline.
- Rationed by requirement. Collateral, a guarantor, or a deposit converts a refusal into a conditional approval, which per our services guide is the standard pattern outside lending too.
- Rationed by friction. A process onerous enough to deter is a refusal that doesn't have to be issued or explained.
The second and fourth deserve attention because they're rationing that doesn't appear as rationing in any statistic. Approval rates look fine; the customer got a product they can't use for what they needed. Which means the measured share of applicants served overstates how much rationing a market does, and by an unknown amount.
The option neither side argues for
The framing that makes the debate look intractable is that price and refusal are the only two levers. They aren't — they're the only two available if the cost of serving is taken as fixed.
Per our distribution analysis, the cost of serving an applicant has components that do move:
- Verification cost, which source-connected data has already reduced substantially.
- Origination cost, which automation reduced.
- Collection cost, which the payroll channel eliminates entirely, per our employer analysis.
- Assessment cost, which per our verification analysis is what makes thin-file applicants expensive rather than risky.
Every reduction in serving cost moves applicants from the refused side of a cap to the served side without anyone paying more. Applicant C at 44% becomes servable at 36% if $8 of cost comes out — which is not a large amount on most loans.
Which is the constructive conclusion. The ration-or-price debate is a fight over how to allocate a fixed cost, conducted by two sides that both take the cost as given. Reducing it dissolves part of the disagreement, and it's the only intervention that improves both error types simultaneously — fewer people refused and lower prices for those served.
It's also, per our distribution analysis, where almost no effort goes, because the returns are diffuse and delayed while the returns to distribution are immediate and measurable.
The strongest objections
"This is a pro-industry argument dressed as measurement." The objection to take seriously, since "we can't measure the harm of exclusion" is convenient for anyone opposing price limits. Two responses. The evidence for rationing errors comes from lenders' own reject inference and override data, which is not advocacy. And the report's actual recommendation — reduce serving costs — is a demand on the industry rather than a defence of it, and it's the recommendation neither side of the usual debate makes.
"Some borrowers are better off refused." Genuinely true and conceded. Where a loan would worsen a situation — the compounding sequence in our time preference analysis — refusal is the right outcome, and ability-to-repay requirements exist for that reason. The argument isn't that refusal is always wrong; it's that its costs are unmeasured while the costs of the alternative are itemized, so the two can't be weighed as things stand.
"You've avoided taking a position on caps." Deliberately, and it's a limitation. The evidence needed to hold a defensible position on cap levels doesn't exist in a form this desk can assess, and asserting one anyway would be exactly the failure the report identifies in others. What we'd defend is the narrower claim: the debate is under-evidenced on one side specifically, and that's fixable.
Testable implications
- Near-cutoff declines should perform better than models predict, consistently across lenders — already evidenced in override data and testable properly by randomized approval below the cutoff.
- Cap introduction should reduce measured high-cost lending while producing substitution into adjacent products, measurable as a shift in product mix rather than a fall in total borrowing.
- Rate caps should ration by loan size as well as by risk, excluding the smallest loans first regardless of creditworthiness.
- Refused applicants should show worse subsequent outcomes than matched approved ones — the direct test, requiring follow-up nobody currently does.
- Reductions in serving cost should expand access without price increases, which distinguishes them from every other intervention.
- Approval rates should overstate access, because inadequate lines and deterring friction are rationing that no statistic records.
The fourth is the one that would settle the question and the one nobody runs. Following applicants after denial — comparing those just below a cutoff to those just above — is a design that requires no randomization and no experimental cost, only the willingness to look at what happened to people a lender declined. That the comparison is almost never made is itself the strongest evidence for this report's central claim.
The conclusion we'd hold: every credit market chooses between two error types, and only one of them generates evidence. That isn't a reason to prefer either. It's a reason to be suspicious of any confident claim about the trade-off, including the ones that sound most protective.
Frequently asked questions
Refusing to lend rather than lending at a risk-reflective price. Most markets do some of both, and where the line sits is set by price limits and convention more than by economics.
One creates a transaction and the other prevents one. Expensive credit produces an identifiable person and a quantifiable amount; refusal produces someone whose later difficulties get attributed to their circumstances.
Both, in proportions that depend on the market. They prevent the highest prices and convert pricing into rationing at the margin — and what the refused applicants do instead is the unobserved question.
Only by measuring outcomes for refused applicants — randomized approval below a cutoff, cross-jurisdiction comparison, or post-denial follow-up. All are possible and rarely done.
Key takeaways
- Rationing and pricing allocate the same risk differently; neither removes it, and they differ in whether anyone can see who bore it.
- The line between them is set largely by caps, fixed costs, convention, and regulatory attention rather than by what serves borrowers.
- Refusal harm is not just invisible but systematically misattributed to the person's circumstances.
- A rate cap rations by loan size as well as by risk, excluding the smallest loans first because fixed costs bind hardest there.
- Inadequate credit lines and deterring friction are rationing that approval-rate statistics never record.
- Reducing the cost of serving is the only intervention that improves both error types at once, and almost nobody works on it.
This report presents an analytical framework and the authors' interpretation; it is not legal, policy, or financial advice, and it deliberately takes no position on the appropriate level of any price limit. Worked figures are stylized illustrations. The outcomes of applicants refused credit are not well documented, and no claim about the net effect of rate caps in any market should be inferred from this analysis; the implications identified as testable are hypotheses.