The Asymmetry Flip: When the Lender Knows You Better Than You Do

Lending theory rests on a premise that shaped the entire field: the borrower knows more than the lender. They know their intentions, their circumstances, their private information about whether they'll repay — and because the lender can't see it, raising prices attracts the wrong borrowers and lenders ration credit instead. For commercial lending and insurance that account remains broadly right. For consumer credit, we think it has partly inverted, on the specific question that matters most. A lender holds the observed behaviour of millions of people resembling you. You hold your intentions — and intentions predict behaviour badly. On whether you'll revolve, whether you'll clear the promotional balance in time, whether you'll overdraw: the better-informed party is frequently not you.

What the classic model says

The standard account, compressed:

1. Borrowers know their own risk; lenders don't.
2. A lender raising rates to cover risk attracts borrowers most willing to accept high rates — disproportionately the riskiest.
3. So raising the price can lower expected return.
4. Lenders therefore ration credit rather than pricing freely: they decline rather than charge more.

This is elegant, empirically supported in several markets, and it underlies how the field thinks about approval cutoffs, collateral, covenants, and disclosure. Every one of those instruments exists to close a gap running from borrower to lender.

Our claim isn't that the model is wrong. It's that it describes one kind of information, and consumer credit turns increasingly on a different kind.

Kind of information	Who knows more	Example
Private facts	Borrower	An imminent job loss, a diagnosis, an intention to default
Verifiable history	Roughly equal	Payment record, balances, income
Future behaviour	Lender	Whether you'll revolve, overdraw, or miss the promotional deadline

The first row is the classic model and it's still real. The third row is where modern consumer credit economics happen, and there the arrow points the other way.

The lender doesn't need to know what you will do. They need to know what people like you did — and they have that, while you have only what you intend.

Where it inverts

The mechanism has two halves, and both have to hold.

Half one: lenders can predict behaviour at the segment level. A lender with a large book observes how borrowers with each profile actually behaved — what share revolved, what share paid the promotional balance in time, what share overdrew and how often. They don't know what you will do. They know the base rate for people indistinguishable from you, which is sufficient for pricing a portfolio.

Half two: people forecast their own behaviour optimistically and systematically. The error isn't random noise around a correct expectation — it runs one direction. People expect to pay in full and revolve. Expect to clear the balance before the promotional period ends and don't. Expect to use the membership and don't. If self-forecasting error were symmetric, the flip wouldn't produce anything; it's the directional bias that makes it economically consequential.

Put the halves together and the informational position is:

• The borrower's forecast: "I'll pay this off in three months."
• The lender's forecast: "Among borrowers with this profile taking this product, most carry a balance past twelve months."
• Which is better calibrated: the second, and it isn't close.

This is a genuine asymmetry in the technical sense — one party holds material information the other lacks, and the transaction is priced using it. It just runs the opposite direction from the one the literature was built around.

The consequence for market structure is the sharpest test. Adverse selection predicts rationing; the flip predicts the opposite. Under the classic model a lender fears that attractive terms draw bad risks. Under the flip, a lender can extend an offer confident that a predictable share of recipients will use it in the way that generates revenue — so solicitation replaces rationing, which is what the consumer credit market actually looks like.

Two-part pricing is the signature

If the flip is real, we should see product structures optimized for it. We do, and they share one shape:

An attractive headline price, plus a separate charge that applies only under conditions the customer expects to avoid.

Product	Headline	Contingent charge	Customer expects
Introductory rate card	0% for a period	Full rate after expiry	To have repaid by then
Deferred interest financing	No interest if paid in full	All accrued interest retroactively	To meet the deadline
Free checking	No monthly fee	Overdraft charges	Not to overdraw
Rewards card	Points and benefits	Interest on carried balances	To pay in full
Minimum payment	A small required amount	Extended interest cost	To pay more than the minimum

Every one is priced attractively against the customer's forecast and profitably against the lender's. That isn't a coincidence — it's what a profit-maximizing product looks like when the two forecasts differ predictably.

Deferred interest is the purest case and worth isolating. The structure charges nothing if the balance clears by a date and everything accrued since purchase if it doesn't. A customer confident of meeting the deadline sees a zero-cost product; a lender knowing the completion base rate sees a well-priced one. The entire economics sit in the gap between the two estimates, and the retroactive feature makes the payoff to the gap as large as possible.

Note also what this does to the cross-subsidy in our subsidy analysis. That report established the direction of the transfer without fully explaining why the paying group is stable rather than learning to avoid the charges. The flip supplies the answer: the payers are the people who misforecast, and misforecasting persists because it's a prediction error rather than an information gap. Knowing overdraft costs $35 doesn't help someone who expects not to overdraw.

What it explains

Several patterns this desk has documented separately fit together under the flip.

Solicitation volume. Adverse selection predicts caution about unsolicited offers. The flip predicts the opposite, and the market is characterized by heavy solicitation.

Profitability concentrated in a minority. The revolvers in our revolver analysis and the fee-payers in our subsidy analysis are the customers whose behaviour diverged from their expectation. The profitable customer isn't the risky one — it's the one who misforecast themselves, which is a different segment from the one credit risk models identify.

Behavioural data outperforming application data. Our renewal analysis found observed behaviour beats stated information at renewal. Under the flip that's exactly right — stated information is the borrower's self-forecast, and self-forecasts are the weakest input available.

Why financial education underperforms. Education addresses knowledge. The flip says the deficit is calibration about one's own future conduct, which knowing the price doesn't correct — consistent with the weak results our disclosure analysis documents.

Why credit lines get extended to people who look constrained. A lender's expected return is highest where behavioural prediction is strongest, which is not the same as where credit risk is lowest.

Why disclosure fails harder here

Our disclosure analysis identified a seven-link chain and found five links routinely broken. The flip breaks a sixth, and it's one that framework didn't isolate.

The chain assumes that a consumer who receives, reads, understands, and can compare a price will act on it correctly. That assumes the consumer can predict which prices will apply to them. Under two-part pricing they frequently can't.

Concretely: a customer told that the rate after the promotional period is 27.99% has received accurate, comparable, timely information. They will still not weight it, because they don't expect to be borrowing then. The disclosure was perfect and irrelevant.

Which yields a specific and uncomfortable conclusion: disclosure of contingent prices is close to useless where the customer misforecasts the contingency — and two-part structures are precisely the ones where that holds. The disclosure regime is weakest exactly where the pricing structure is most sophisticated.

It also explains an otherwise puzzling observation. Requirements to disclose contingent charges more prominently have generally produced modest effects. That's the predicted result, not a failure of implementation — making a price more visible doesn't change a decision made by someone who believes the price won't apply.

The remedy that follows

If the deficit is self-prediction rather than price knowledge, the intervention has to supply a different kind of information.

Personalized behavioural feedback. Not "the rate is 27.99%" but "based on your last twelve months, you would pay approximately $840 on this balance." That's a statement about the customer rather than about the product, and it targets the actual error.

The design principles:

• Use the customer's own history, which is the most credible evidence about their behaviour and which the institution already holds.
• Express it in dollars over their likely horizon, not as a rate.
• Deliver it at the decision point.
• Where individual history is thin, use the segment base rate — "most customers who take this offer still carry a balance after twelve months."

The closest existing example is the minimum payment disclosure requiring a statement of how long repayment takes at the minimum. That's a self-prediction correction rather than a price disclosure, and it's the model worth extending.

Two honest complications. It requires the institution to disclose its own prediction, which is competitively sensitive and which no firm will do voluntarily where the prediction is unflattering. And the paternalism objection has real force — telling someone what they'll do is different from telling them what something costs, and reasonable people find it intrusive. We'd hold that a factual statement about the customer's own past behaviour is the least paternalistic version available, but it is not neutral.

The alternative remedy is structural rather than informational, and follows the finding in our disclosure analysis that defaults outperform information: constrain the structure rather than explaining it. Requiring interest to accrue only prospectively, capping retroactive charges, or setting minimum payments at levels that retire balances in reasonable periods all operate without requiring anyone to forecast themselves correctly.

Where the flip doesn't apply

The argument is narrower than it may sound, and the boundaries matter.

Private facts still favour the borrower. Someone who knows they're about to lose their job, or has a diagnosis, or intends not to repay holds information no model can see. Classic adverse selection is entirely intact for that category, and it's why the exercise-cost framing in our option analysis remains necessary.

Small business lending is closer to the classic model. Owners hold substantial private information about their business, thin files limit prediction, and each business is more idiosyncratic — which is why the assessment in our valuation guide looks so different from consumer scoring.

Insurance remains classically adverse-selected, because the insured knows their risk and the insurer's population data is weaker relative to it.

Prediction is population-level, not individual. A lender knows the base rate, not this person's outcome. That's a real limitation: the flip is statistical rather than absolute, and any individual borrower may well be right about themselves. It matters for portfolio pricing and not for the individual case, which is a distinction the argument must keep.

It requires scale. A small lender lacks the observations. The flip is therefore a property of large consumer books, which is itself a scale advantage that compounds the concentration our selection analysis describes.

The strongest objections

"This is condescending about consumers." The most serious objection. Saying people misjudge themselves risks treating adults as incapable, and the evidence for optimistic self-forecasting comes from settings whose generalizability is debated. Our response: the claim is about a specific and well-documented kind of error rather than about general competence, and it applies to sophisticated people too — the finance professional who intends to pay in full and doesn't is the same case. It's a prediction problem, not an intelligence one.

"Firms aren't doing this deliberately." Largely conceded and it doesn't matter to the analysis. A product that prices well against realized behaviour survives competition whether or not anyone reasoned about self-forecasting. The structures are selected for by profitability, not designed from the theory — which makes the pattern more robust rather than less.

"Adverse selection still dominates; you've overstated the flip." A fair empirical challenge, and we'd note the flip is a claim about which asymmetry drives pricing structure rather than about which is larger overall. The test is whether two-part pricing is concentrated where behavioural prediction is strongest, which is checkable and is the implication below we'd most want examined.

Testable implications

1. Two-part pricing should be most prevalent where behavioural prediction is strongest — large books, repeat products, rich histories — and rarer in thin-data segments.
2. Profitability should concentrate in customers whose stated intent diverged from behaviour, which any lender collecting intent at application can test directly.
3. Personalized behavioural feedback should change behaviour more than price disclosure at equivalent prominence. The sharpest test of the remedy.
4. Solicitation intensity should correlate with prediction quality rather than with credit quality.
5. Products should show a bimodal outcome distribution — a group who used them as intended and paid little, and a group who didn't and paid a lot — rather than a smooth spread.
6. Structural remedies should outperform disclosure remedies on the same product, which the minimum payment case partly tests already.

The second is the one worth running and almost nobody has. Ask applicants how they expect to use a product, then compare to what they did. The gap is measurable, it's the central quantity in this report, and a lender with intent data and outcome data can compute it in a day.

The broader point is about which model the field reasons from. An industry that thinks its problem is borrowers hiding information will build verification, collateral, and disclosure. If the operative asymmetry runs the other way, those instruments address a gap that isn't the binding one — and the interventions that would matter are about helping people predict themselves, or about constraining the structures that profit from their not doing so.

Frequently asked questions

---

This report presents an analytical framework and the authors' interpretation; it is not financial or policy advice. The evidence on optimistic self-forecasting comes from a research literature whose magnitude and generalizability across settings remain debated, and the implications identified as testable should be treated as hypotheses rather than established results.