Vintage Analysis: Measuring the Quality of Credit You Granted

Vintage analysis tracks how groups of customers perform based on when the credit was granted. It is the only reading that precisely answers the most important question in the discipline: are the decisions we are making now better or worse than the ones we made six months ago?

Why total delinquency cannot answer that

Portfolio delinquency mixes old and new cohorts, different volumes and different terms. A fast-growing book shows artificially low delinquency — the denominator grows faster than the problem surfaces. When growth slows, delinquency appears to "explode" even though nothing changed in underwriting.

Vintage isolates that effect because it follows the same group through time.

Building the matrix

1. Group customers by the month credit was granted (the vintage)
2. For each vintage, measure cumulative delinquency at the end of each month of life
3. Build the matrix: vintages as rows, months on book as columns

Read it vertically: compare the same column across vintages. In the example, March and April are clearly worse at 3 and 6 months — something changed in underwriting, and the effect has not hit total delinquency yet.

What the curve tells you

• Early slope — the faster the curve rises in the first months, the more likely an underwriting or fraud problem
• Plateau — the level where the curve flattens is that vintage's expected loss
• Cross-vintage comparison — consistent deterioration points to looser criteria, a changed mix or a weakening market

Cuts worth running

The overall matrix says something changed; the cuts say what. Build separate vintages by:

• Origination channel (rep, e-commerce, distributor)
• Score band
• Decision path: automated, analyst, exception
• Segment and region
• Granted limit band

The decision-path cut is usually the most revealing: when the exception cohort runs far above the rest, you have a number to support the conversation about what exceptions cost — instead of an opinion.

What to do with the result

Frequency and horizon

Refresh monthly and read with a horizon of at least 12 months — a two-month-old vintage says little. In short-term, fast-turning operations, six months already gives a reasonable read.

The connection to the rest of the tracking is in credit portfolio monitoring and credit and collections KPIs.

Construction mistakes

• Grouping by due date instead of origination date — destroys the logic
• Mixing products with very different terms in one matrix
• Using absolute dollars instead of a percentage of the vintage
• Comparing vintages at different maturities — only column against column

What to take from this

Vintage analysis is the thermometer for underwriting quality. Build the matrix by origination month, always compare the same month on book, and use the channel and decision-path cuts — that is where the cause shows up.