Lecture series
Credit scoring, from first principles
What a credit score is, why lenders use one, how a scorecard is built and tested, and how to put one to work in a lending business. Ten lessons, with two interactive examples.
10 of 10 lessons published. Lesson 1 is free to read.
Listen to this course
Now playing
Press play to start
- 01What a credit score is, and why lenders use oneA credit score is a number that ranks borrowers by how likely they are to fail to repay. It exists because lending is a decision made thousands of times, and those decisions need to be consistent, fast and measurable.
- 02The data behind a scoreA score can only be as good as the information behind it. Credit bureau data, what the applicant tells you, how existing customers behave, and alternative data each answer a different question.
- 03How a scorecard worksA scorecard turns information about a borrower into points, adds them up, and maps the total to the odds of repaying. Each step is simple, which is why scorecards remain widely used in consumer lending.Interactive
- 04Does the score work? Measuring itTwo questions decide whether a score is fit to use: does it put risky borrowers below safe ones, and do its probabilities match what actually happens? They are measured differently, and a score can pass one and fail the other.
- 05Setting the cut-offA score decides nothing until someone draws a line. Where to draw it is a business decision that trades approvals against losses, and it should be made with the numbers in front of you.Interactive
- 06Beyond approval: limits, pricing, monitoring and collectionsApproval is only the first decision in the life of a loan. The same probabilities of default also help decide how much to lend, at what price, how to watch the account and whom to contact first when payments stop.
- 07When data is thin: new customers, alternative data and rejected applicantsA score learns from past borrowers. People with little or no credit history, and applicants the lender turned away, are exactly the people it knows least about. Each gap has its own remedies, and each remedy has limits.
- 08Machine learning and scorecardsMachine learning can find patterns that a scorecard misses, and in a large benchmark study it ranked borrowers better than the usual scorecard method. The gain comes with costs: a flexible model can learn noise, become unstable and be hard to explain. A lender has to weigh both, and still be able to tell each declined applicant why.
- 09Fairness, explanations and the rulesA score decides who gets credit and on what terms, so it has to be fair, explainable and lawful. Regulators approach this in different ways: principles and assessment methods in Singapore, notice requirements in the United States, and duties for high-risk AI in the European Union. Leaving sensitive data out of a model does not settle the question.
- 10Running a score in a lending systemA score earns its keep only once it runs inside the lending system: placed correctly in the decision flow, recorded with its version, tested against the score it replaces, watched for drift and recalibrated or rebuilt when it slips. Bank supervisors set expectations for much of this under model risk management.
Quiz and certificate 10 questions; those on lesson 1 are free to try. Read every published lesson and score 80% or more for a certificate of completion.
Take the quizCredit bureaus: developments and credit decisions →Lending rules by market →All courses →