Credit Bureaux: What They Hold and How a Score Works

A credit information company keeps a record lenders build. Each lender reports the credit it extended and how it came back, the company holds those reports against an identity, and it returns them to lenders entitled to ask. A credit score is a number produced from that record by a method, and it ranks accounts inside a population by observed repayment. The score describes a record of borrowing rather than a person.

Credit information companies are called credit bureaux, and the shorter name has done some quiet damage. Bureau makes the arrangement sound like an office that investigates people. A credit bureau investigates nobody. A credit bureau receives, keeps and hands back. Every property it has, including the ones that matter most to the person the entries are held against, follows from the single problem it was built to solve, and that problem belongs to a lender rather than to a borrower. The honest account of an arrangement is usually the useful one, and the honest account of a credit bureau starts with the lender.

Why does a shared record exist at all?

Start on the wrong side of the balance sheet on purpose. Rukmini Finance Limited is an invented finance company that lends and takes no deposits. Rukmini Finance carries assets under managementThe total a lender has out on loan and manages on its own book at a stated date. It measures the size of the lending side and is not a profit figure. of Rs 18,000 crore, and it funded that book with borrowingsMoney the lender has itself borrowed, from banks or from the market, and must repay on terms it agreed. This is the funding side of a lender that takes no deposits. of Rs 14,400 crore standing against net worthWhat is left of a lender's own assets once everything it owes has been taken off. It is the buffer belonging to the lender itself rather than to anybody it borrowed from. of Rs 3,600 crore. Dividing the borrowings by the net worth gives 4.0 times, on Rs 3,600 crore of net worth, so four rupees in five that went out of the door as a loan came in through the door as somebody else's money first.

Look now at what Rukmini Finance can and cannot see. A single asymmetry sits there, and it is the whole reason a shared record exists. Rukmini Finance can see every loan it has itself made, to the rupee and to the day. The same lender can see nothing at all about the loans anybody else has made to the same person. So without something outside itself, it is pricing an obligation while blind to every other obligation sitting alongside it, and it is doing that with money it has to give back.

Here is the everyday shape of it. Three shopkeepers on one street each extend a month of credit to the same customer, on the quiet understanding that a month of sales will cover it. Not one of the three knows about the other two. Each of them has made a perfectly sensible judgement on the information in front of them, and the three judgements together add up to something none of them agreed to. Nobody misbehaved. The information simply was not in one place, and no amount of care inside any one shop could put it there.

So the arrangement was built to solve a lender's information problem, and everything it later came to give the person described was designed in afterwards rather than being the reason it exists. Saying that plainly, instead of softening it, explains the arrangement's shape better than any friendlier account does. The lender's problem explains why the record holds what lenders report and nothing else. The same problem explains why a correction has to travel back to a lender rather than being made where the record sits. And it explains why anybody a lender never lent to is invisible: they were never part of the problem it was built for.

What is a credit information company, and what does it not do?

Define it by its verbs and the confusion clears in a sentence. A credit information company receives reports from lenders about credit those lenders extended and how it was repaid. The company keeps those reports against an identity. The company returns them to lenders entitled to ask. Three verbs, and there is no fourth one hiding.

The company originates nothing. Readers get that one sentence wrong more often than any other, and getting it wrong sends every follow-on question to the wrong address. Every entry in a credit record arrived there because a lender wrote it and sent it. Nothing in the record was observed, inferred, judged or decided by the company holding it. Which means a question about an entry is never really a question about the company. The question belongs to the lender that reported the entry, and the whole correction route set out below falls straight out of that fact.

One more distinction separates two parties that often sit in the same application, and it is worth fixing before going further. The lender that holds the loan on its own book is the party that reports. A lending service providerA party that finds, sources or services a loan on a lender's behalf without holding the loan itself. What is required of one is set by the Reserve Bank of India at rbi.org.in. that found the borrower, took the application and chases the instalment is not the party whose name the entry belongs to. Nor does the money movement create anything: a repayment instruction travelling through an invented payment system such as Setu Payments Limited moves value and creates no entry whatsoever. The entry appears because a lender reported it.

The Reserve Bank of India at rbi.org.in sets what such a company may hold and who may see it, and the requirement moves.

What does a credit record actually hold?

Describe the contents by kind rather than by field, and two useful things happen at once. The description stops going out of date the moment a field is added somewhere. The absences also become visible, and a field list never manages that.

Four kinds, and there is no fifth. The first is the identity the entries are held against. An identity is what turns a collection of reports about one person into a record rather than a heap. The second is the obligation: what was borrowed, from whom, and on what terms. The third is the repayment record against that obligation as it ran, and it is the only part with time in it. The fourth is the record of who asked to see the whole thing. Most readers do not know that record exists, so it gets its own section further down.

Now the part that teaches more than the list does: the record holds credit, so it does not hold an income, an employment, or anything at all about the circumstances a repayment was made in. Neither absence is an accident, and neither is coming later. An income and an employment were never inside the problem the arrangement was built to solve, so nobody ever built a route for them to arrive by. A reader who assumes the record knows what somebody earns has misread what the arrangement collects, and a lender that assumes it has skipped a question it still has to ask somewhere else.

What is a credit score, and what is it a ranking of?

The definition rewards a slow reading, and it comes apart three ways. A credit score is a number produced from the record by a method, and it ranks accounts within a population by observed repayment. Three separate claims are packed into that one sentence, and each of them changes what can honestly be said about the number.

First, the score is a ranking. A ranking is a position relative to a population rather than an absolute measurement of anything. A thermometer measures. A finishing position does not. Move the same runner into a different race and the position changes while the runner does not, and nobody thinks the second race proves the runner got worse. A score behaves like the position, not like the thermometer, and almost every misuse of one starts by forgetting which of the two it is.

Second, it is produced by a method, and the method is not published, so nobody outside can reproduce the number. An unpublished method is an unusual property for a figure that gets treated as decisive, and the property is worth naming rather than working around. An interest calculation can be checked. A ratio can be rebuilt from two figures on a balance sheet and tested against what somebody said it came to. Neither of those is possible with a score, and the impossibility of checking is the single strongest reason to hold a score at the distance described below.

Third, it is computed from a record of credit, so it can only rank what the record contains, and the record contains what lenders reported. Everything the previous section listed as absent is absent here too. The method cannot weigh what was never sent to it.

Taken together, the three yield the central line: a score describes a record of borrowing, and it is not a measure of a person. The line is not a softening or a kindness. The line states what the arithmetic can reach, and a reader who remembers one sentence about scores should remember that one.

What can a score not know?

A limit somebody can name is a limit somebody can work around, and a limit left vague just becomes a vague feeling of unease. So list the limits rather than implying them.

No income was ever reported into the record a score is computed from, so a score cannot know an income. Nothing about circumstances travels down a reporting chain, so a score cannot know what a particular month cost somebody. And an entry only begins to exist when a lender has already said yes to something, so a score can know nothing at all about credit that was never extended.

Keep the consequence on the lender, where it belongs: a score is an input to the four things an underwritingThe work a lender does before agreeing to a loan: settling whether to lend, how much, at what price and on what terms. Covered separately and used here rather than rebuilt. decision settles, and it is not one of those four things. Whether to lend, how much, at what price and on what terms are four questions, and the number that came back from the record is one of the materials each of them is answered with. The number is not an answer to any of them, and it is certainly not an answer to all four at once. The failure section further down is what happens when somebody treats it as one.

How does an entry become wrong when nobody did anything?

Write this one mechanically rather than sympathetically. Sympathy is not actionable and the mechanics are, so here are four routes named as the processes they actually are.

The first is timing. A report was prepared and sent on one date and the record shows the position as at that date, so a payment made afterwards has not arrived in the record yet. The record is not lying. The record is simply older than the fact. The second is a closure never carried through: an obligation was settled and the settlement was never reported as a closure, so an entry that should have stopped is still standing. The third is identity. Where identifiers are similar, a report can be filed against the wrong one, and the entry is now sitting in a record it was never about. The fourth is terms: what was recorded differs from what was agreed, and every later entry against that obligation inherits the difference.

The four routes have one thing in common, and it is the lesson of the whole section: not one of them involves anything the person described did or failed to do, and every one of them is a processing outcome inside a reporting chain. A chain with four handovers has four places to drop something, and it will drop something eventually, at a rate that has nothing to do with anybody's conduct. The correction route below exists for exactly that reason, and nothing else about the arrangement matters more to the person the entries are held against.

How is an entry put right, and who actually fixes it?

The shape of the route is teachable without stating a single rule, so here is the shape. A person may obtain their own record. An entry in it may be disputed. The dispute goes back to the lender that reported the entry. The entry came from there, and the lender is the only place the underlying fact lives. And the record is amended where the lender's report turns out to have been wrong.

Notice that the correction travels back up the identical line the entry came down, one handover at a time. The return path is not an administrative quirk. The path falls directly out of the fact established earlier: the company holding the record originated nothing, so it has nothing to check the entry against. Only the lender does.

How all of this works, in what time, and by whom, is set by the Reserve Bank of India at rbi.org.in, and it moves. A correction timeline written out in full would not merely go stale on the day the timeline changed. A stale timeline would be wrong, in the confident voice of a finished text, to a reader who had come looking for exactly that.

And here is the sentence worth carrying away from this section: the correction route is the single most valuable property this arrangement has for the person the entries are held against, and it is the property that an alternative signal usually does not have at all. A record that can be obtained, questioned and amended is a fundamentally different kind of thing from a signal inferred about a person somewhere that person cannot see, cannot ask about, and cannot send back.

What does the record of who asked contain?

Most readers do not know this part exists. Every time a lender looks at a record, that look is itself recorded. The reading leaves a trace in the thing being read.

Two things about the record of enquiries are settled without inventing anything. The record exists, as one of the four kinds set out earlier. And the record distinguishes between a person looking at their own record and a lender looking at it in connection with an application. A person's own look and a lender's look are two different events, and there would be no point recording them identically. On what basis a lender may look at a person's record at all is set by the Reserve Bank of India at rbi.org.in, and it moves.

The effect of an enquiry on a score cannot be stated, and the reason for that silence is itself worth following. Three reasons stack up, and they are cumulative rather than alternative. No scoring method is published, so nobody outside can derive the effect. The scales and distributions an effect would have to be measured against are not published either, so no outsider holds the ingredients a derivation needs. And a wrong statement on this particular subject would not land on a reader as an interesting general claim. A wrong statement would land as a fact about the reader's own record, and the reader might then act on it. A plausible wrong number is worse than a stated silence.

Where does the whole record actually come from?

No borrower is worked through: a treatment of credit records is the last place an invented person belongs. The working goes instead to where the record comes from. A record comes from the lenders' own books, and that is exactly where it should be examined.

Take the two invented lenders together. Rukmini Finance Limited carries assets under management of Rs 18,000 crore. Suvarna Commercial Bank Limited, an invented commercial bank, carries gross advancesThe lending book counted before anything is set aside against advances that may not come back. Settled earlier in this material and used here as a stated base. of Rs 1,44,000 crore. Adding them gives Rs 1,62,000 crore of lending across two lenders, and every rupee of it becomes entries in somebody's record, all of it reported by the lenders themselves. Suvarna Commercial Bank's book is 8.0 times Rukmini Finance's on those two figures, so between just these two, roughly eight entries in nine by value trace back to the larger one.

Now the central division, worked through rather than asserted. At Suvarna Commercial Bank Limited, gross non-performing advances are Rs 6,480 crore. Dividing that by gross advances of Rs 1,44,000 crore gives 4.50 per cent, struck on gross advances. Taking that off leaves Rs 1,37,520 crore. Against the same base of Rs 1,44,000 crore, that is 95.50 per cent performingSaid of an advance on which repayment is arriving as agreed. Where an advance stops being treated as performing is set by the Reserve Bank of India at rbi.org.in. by value.

So a credit record is overwhelmingly a record of repayment that happened, and a reader who pictures it as a list of trouble has the proportions of the thing backwards by a factor of about twenty. The proportion is not a reassuring aside. The proportion changes what a credit record is for. A file that is mostly evidence of obligations met is a fundamentally different instrument from a register of problems, and the two would be read in opposite directions by anybody using them.

Whom can this arrangement not describe at all?

The answer was sitting in the first section the whole time, so close where that section started. An entry begins with a lending decision. No decision, no entry. So the arrangement can describe only people who have already been lent to, and everybody else is not badly described by it. The arrangement is simply silent about them.

And the population the arrangement cannot describe is precisely the population a lender most often wants to reach. Reading other kinds of record is therefore a subject in its own right, covered separately and earlier. The circularity is the sharpest thing in the whole arrangement. A lender wants a record before it lends. A record only exists because somebody already lent. The gap that produces is structural, built into the design, and no amount of care by anybody inside the arrangement closes it. The gap is not a coverage problem waiting to be solved by more effort. The gap is the arrangement.

Why can a score not be put under a slider?

A control that moves a score until an approval flips or a price moves is the usual way this subject gets illustrated, and it leaves the reader feeling that something has been understood. No such control appears below, for two reasons.