What a credit score is, and why lenders use one

What it is

The World Bank's guidelines on credit scoring describe it as a statistical method used to predict the probability that a loan applicant, an existing borrower or a counterparty will default or become delinquent. In practice, a score adds up points for each piece of information about a borrower, and a higher total usually means a lower chance of default.

Two things follow. First, a score is about a defined event (for example, falling 90 days behind within a year) over a defined period; it says nothing about anything else. Second, the number itself has no natural unit: the points can be scaled to any range. What matters is how well it ranks borrowers and, once calibrated, what probability of default each score stands for.

Why lenders use one

Before scoring, consumer lending decisions rested on a loan officer's judgement. Judgement can be good, but it is slow, varies from officer to officer and from day to day, and is hard to check afterwards. Hand and Henley's review of consumer credit scoring describes it as formal statistical methods that classify applicants as good or bad risks, which became more important as consumer credit grew. Scoring brings four things:

• Consistency. The same applicant gets the same answer, whoever handles the application.
• Speed. A decision takes seconds, which is what makes lending inside an app, at a checkout or through payroll possible.
• Measurability. Because every decision follows the same rule, you can test afterwards whether the rule worked and change it deliberately.
• Control. Credit policy is set once, by people accountable for it, instead of being re-decided case by case.

Kinds of score

The World Bank guidelines list several uses along the life of a loan: application scores for new applicants, behavioural scores for existing customers, collection scores for accounts in arrears, early-warning scores, and fraud-detection scores. A lender may also use a score supplied by a credit bureau, built on data pooled across lenders, alongside its own.

What a score is not

A score is not a judgement of a person's character, and it is not a promise: some high-scoring borrowers will default and many low-scoring ones will not. It is only as good as the data and the population it was built on; a score built on salaried city borrowers may rank farmers poorly. And it rarely decides alone. Eligibility rules, affordability checks, fraud checks and the law sit around it, and the cut-off that turns a score into a yes or no is a business decision (lesson 5).

Where this series goes

The next lessons follow the life of a score: the data it is built on, how a scorecard turns that data into points, how to tell whether it works, where to draw the line, how scores are used after approval, what to do when data is thin, where machine learning fits, the rules on fairness and explanation, and how to run a score inside a lending system.