The Information Not Collected: How the Data Boundary Became the Inclusion Boundary

A household paying rent on time for eleven years, never missing a utility bill, insured continuously, has almost no credit file. A household that opened a card at twenty-two has a thick one. The difference isn't reliability — it's which of their obligations happened to be held by an industry that organized a reporting mechanism. Credit reporting records credit accounts exhaustively and records nearly nothing else, and that boundary was set by history rather than by any assessment of what predicts repayment. Which produces a circularity nobody designed: the system can assess people who already had credit, so having had credit is the qualification for being assessable for credit.

What's inside the boundary

Obligation	Positive payment recorded?	Non-payment recorded?
Loans and cards	Yes, in detail	Yes
Mortgage	Yes	Yes
Rent	Rarely	Yes, if it reaches collections or court
Utilities	Rarely	Yes, via collections
Telecom	Rarely	Yes, via collections
Insurance premiums	No	Via collections
Childcare, tuition	No	Via collections
Cash income	Not visible at all	—

Read the two columns against each other. For most of the list, reliability produces nothing and failure produces a record. The system is not neutral about these obligations — it observes them only when they go wrong.

That asymmetry is the report's central structural finding and it gets its own section below, because it isn't a gap in coverage. It's coverage of one tail only.

For most household obligations the system records failure and not success. That isn't a gap in coverage — it's coverage of one tail.

Why it fell there

The boundary looks arbitrary because it is, in the sense that it wasn't chosen. It's the residue of which industries had the conditions to organize reporting.

What those conditions were:

• Concentration. A manageable number of institutions who could agree on a format and a mechanism. Lenders had this; landlords never have.
• Mutual interest. Lenders share borrowers, so each gains from what the others know. A landlord gains little from telling a lender their tenant pays.
• Existing records. Institutions that already maintained account systems could furnish; those keeping informal records couldn't.
• Regulatory framework. Once a structure existed, rules formalized it — and the rules describe the structure that emerged rather than one that was designed.

Per our verification analysis, a credit file is portable standardized trust, valuable because many parties accept it. That value is a network property, so it accrued to whichever obligations got into the network early — and once the network existed, its shape became the definition of what counts as a payment record.

Which is worth stating plainly: nobody assessed whether rent payment predicts loan repayment and decided it didn't. The question was never reached, because the mechanism that would have carried the answer was never built.

The circularity

The consequence that makes this more than a historical curiosity.

1. The system records credit accounts.
2. So it produces thick files on people who have borrowed and near-empty ones on people who haven't.
3. Lenders assess what's recorded, so thin-file applicants are hard to assess.
4. Per our thin-file analysis, they're declined at rates exceeding what their performance justifies.
5. Being declined generates no record either.
6. So the file stays thin.

Having had credit is the main qualification for being assessable for credit. Which is a selection mechanism operating on everyone at once, produced by a data boundary rather than by any policy.

Two of this desk's findings meet here. Per our verification analysis, thin-file applicants are frequently expensive rather than risky — the barrier is the cost of establishing facts about someone with no portable record. And per our rationing analysis, the resulting exclusion generates no evidence, so the cost of the circularity is invisible to everyone including the people maintaining it.

The circularity also explains something otherwise puzzling: why the thin-file problem hasn't been competed away. A lender who could assess these applicants profitably would do so — but the assessment requires information the network doesn't carry, and building an alternative network is the coordination problem that made the original one hard.

Damage travels, reliability doesn't

The specific unfairness in the current arrangement, and it's sharper than the coverage gap.

An unpaid utility bill reaches your credit file through collections. Eleven years of paid ones reach nothing. Per our reporting analysis, that pattern holds across rent, utilities, telecom, insurance, and most other recurring obligations.

Why it happens is mechanical rather than intentional. The collections industry is inside the reporting network and the originating industries aren't — so an obligation crosses the boundary only by failing, at which point it's handled by a party that furnishes.

The consequences compound:

• A household can only lose from these obligations, never gain. There is no upside case.
• The population most affected is renters, who per our housing analysis are disproportionately the households the system already serves worst.
• It distorts what files mean. A file with a utility collection and no positive utility history reads as a person who fails at utilities, when it may be one lapse against a decade.
• It penalizes the shift away from credit. A household that pays cash and avoids borrowing accumulates no record while remaining exposed to the collections route.

The third point matters for how files are read. Because only failures are recorded, the base rate is unobservable — you can't tell whether a collection is an outlier in an otherwise perfect record, since the record doesn't exist. That's an information problem for lenders as well as an unfairness to households.

The households outside entirely

The boundary excludes not just obligations but people, and the exclusion runs deeper than thin files.

• Cash income is invisible. Someone earning reliably in cash has, from the system's perspective, no income at all — which per our irregular income analysis compounds with variability to make the household unassessable on two dimensions at once.
• Informal obligations are invisible in both directions — money lent to and borrowed from family appears nowhere.
• Household-level arrangements are invisible. The system observes individuals while households pool resources, so a person with no accounts in their name may be jointly managing substantial obligations.
• Recent arrivals have no domestic record regardless of history elsewhere, since files do not port across systems.

The last is worth noting as evidence for the report's thesis. A person with decades of documented repayment in another country arrives with an empty file — which demonstrates that the file is a record of participation in one particular network rather than a record of creditworthiness.

And connecting to our structural analysis: the system assesses individuals, but the entity actually managing obligations is the household. The unit of observation doesn't match the unit of decision, which is a measurement error that runs through everything built on top.

Why closing it is hard

The obvious remedy is to record more obligations. The obstacles are real and worth stating properly, because advocacy for alternative data frequently understates them.

Adding an obligation adds its negative tail. A household that pays rent reliably gains; one that has struggled acquires a record it didn't have. Voluntary opt-in resolves this for the consumer and creates a selection problem for the lender, since a file that includes rent only when the tenant chose to include it carries information about the choice as much as about the payments.

Coverage would be uneven. Participation would be voluntary for most providers, so file completeness would vary by who your landlord happens to be — which per our attribute analysis produces a missing-data pattern that is not missing at random and that models handle badly.

Furnishing carries obligations. Per our furnisher guide, a party that reports takes on accuracy duties, dispute handling, and correction responsibilities. For a small landlord that burden is disproportionate, which is the same fixed-cost problem our institutional analysis describes — and it's why the industries that never organized reporting still haven't.

Data quality would be worse. Lenders maintain account systems built for this; a landlord's records are not, and disputes would rise accordingly.

None of these is a reason to leave the boundary where it is. They're reasons the remedy requires infrastructure — an intermediary carrying the furnishing burden, coverage broad enough to avoid selection, and dispute handling that works — rather than an exhortation to report.

Where the argument stops

An important distinction, because this report's argument is frequently conflated with a broader one it doesn't make.

The case here is narrow: obligations of the same kind the system already records, held by people the system currently cannot see. Rent, utilities, telecom, and insurance are recurring payment commitments with due dates and amounts — structurally identical to what a credit file already holds.

That is different from arguments for collecting data of kinds the system has never held — behavioural signals, inferred characteristics, information derived from activity that isn't a payment obligation. Those raise separate questions:

• Can a person know what's recorded about them? A payment record is checkable; an inferred characteristic frequently isn't.
• Can they contest it? Per our correction guide, dispute rights depend on the record being specific enough to dispute.
• Can they act on it? Per our explainability analysis, a reason must describe something the applicant can influence.
• Does it proxy for something impermissible? A question that applies with more force the further data gets from the obligation itself.

More data is not the argument. A specific asymmetry is. The system records one category of obligation completely, records others only when they fail, and treats the resulting picture as a description of a person. Fixing that asymmetry is a different project from expanding what may be collected, and conflating them has probably slowed both.

The strongest objections

"Credit accounts predict better, which is why they're used." Partly true and it doesn't establish the conclusion. Credit history does predict credit performance well — but that's partly because it's what models were built on, and the comparison against rent or utility history has rarely been run on comparable data because the data doesn't exist. The claim that the boundary sits in the right place is untested rather than established, and the absence of the test is a consequence of the boundary.

"Voluntary reporting programs already exist." They do, and per our rent reporting guide they help the people who use them. Two limits: uptake is low relative to the affected population, and voluntary participation creates the selection problem above. A remedy available to those who find it and opt in is not a fix to a structural boundary.

"You're assuming these obligations predict repayment." Fair, and the report shouldn't be read as asserting the size of the effect. What we'd defend is that the question is open and the current arrangement forecloses it — and that an asymmetry recording only failures is hard to justify on any account of what a file is for.

Testable implications

1. Rent and utility payment history should predict credit performance among thin-file borrowers, testable wherever both series exist for the same people.
2. Households with reliable non-credit payment records should perform better than their files suggest — which is the thin-file finding, with a proposed cause.
3. The asymmetry should be visible in file composition: collections from utilities and telecom appearing far more often than positive tradelines from the same sectors.
4. Voluntary rent reporting should show selection, with participants outperforming non-participants beyond what the payment data explains.
5. Coverage should correlate with provider size, since furnishing is a fixed cost.
6. Recently arrived households should perform better than their domestic files predict, since the file records network participation rather than reliability.

The third is the easiest and would establish the asymmetry as a measured fact rather than a characterization. Counting positive versus negative tradelines by furnishing sector across a large sample of files would show directly how one-sided the coverage is — and it requires no new data collection, only an audit of what's already there.

The conclusion we'd hold: the credit system's coverage boundary is a historical accident that has become an eligibility rule. It records one category of obligation in full, records the rest only when they fail, and the people it cannot see are not the people who are unreliable — they're the people whose reliability was never carried by anything.

Frequently asked questions

---

This report presents an analytical framework and the authors' interpretation; it is not legal or policy advice. What may be furnished to consumer reporting agencies, by whom, and subject to what obligations is governed by law that varies and continues to develop, and the general characterizations here are not a description of any particular requirement. No estimate is offered of how much non-credit payment history would improve prediction; the implications identified as testable are hypotheses.