Vintage Analysis Explained: How to Build and Read Vintage Curves

Vintage analysis groups loans by the period they were originated, called a vintage, and tracks how each group performs as it ages. A vintage might be all personal loans booked in January, or all credit cards opened in a given quarter. For each vintage, you measure something like the share of accounts 30 or more days past due, or cumulative net charge-offs, at each month since origination.

Plotting those results produces vintage curves, with months on book along the bottom and the performance measure up the side. Each line is one vintage. Because every line starts at the same point in its life, you can compare a new vintage directly with older ones at the same age.

Loans have a life cycle. Most retail products show very little delinquency in the first few months, a rise as the portfolio seasons, a peak and then a leveling off as weaker borrowers default and stronger ones pay down. Where a loan sits in that cycle drives its delinquency more than almost anything else.

That is why portfolio-level delinquency can mislead. If a lender grows quickly, the portfolio fills with young loans that have not had time to go bad, and overall delinquency falls even if the new loans are weaker. When growth slows, delinquency rises as those loans season. Comparing vintages at equal months on book removes the age effect so you can see the real change in credit quality.

The method is simple, but each step has choices that affect the result, and two analysts who make different choices can reach different conclusions from the same data. Write the choices down so the analysis can be repeated next month and checked by someone else. The steps below work for cards, auto loans and personal loans with small adjustments.

Example only, with made-up round numbers. A lender booked 10,000 auto loans in the first quarter of one year and 12,000 in the first quarter of the next. At 12 months on book, 300 loans from the older vintage had ever been 60 or more days past due, a rate of 3.0%. At the same 12 months on book, 480 loans from the newer vintage had reached that point, a rate of 4.0%.

Meanwhile, the portfolio's overall 60+ delinquency rate barely moved, because rapid growth added many loans too young to show problems. The vintage view shows that the newer loans are performing about a third worse at the same age. That is the signal to look at what changed in underwriting, pricing, dealer mix or the economy during the second origination period.

A vintage chart with many lines can look like noise at first. It helps to focus on a few recent vintages against a longer-standing baseline, rather than every line at once, and to read the chart alongside the counts behind each point. These patterns are the ones most worth looking for.

Vintage analysis is the natural way to measure the effect of a change. If a lender loosens a score cutoff, adds a new marketing channel or changes its credit line assignment, the vintages booked after the change can be compared with those booked before it at the same age. Splitting by the segment affected, such as the new score band, makes the effect clearer.

For credit cards, the same idea applies to line increases. Grouping accounts by when they received an increase and tracking their delinquency and utilization afterward, against a similar group that did not, shows whether the strategy added risk. Testing a strategy on a smaller group first and reading its results by months since the change keeps surprises small.

The most frequent mistake is using a denominator that changes over time, such as current balance, which shrinks as loans pay down and makes rates look worse with age. Fix the base at origination unless you have a specific reason not to. Another is comparing vintages of very different sizes without noting how small the newest ones are; a rate based on a few hundred accounts can move a lot on noise.

Analysts also run into trouble when product terms differ across vintages. A 36-month loan and a 72-month loan season differently, so mixing them hides real changes. Finally, remember that later vintages may have been affected by events the earlier ones never saw, so check calendar effects before blaming underwriting.

Because vintage curves follow a typical shape for a given product, they are often used to project how younger vintages will perform. If older vintages of a personal loan product reach most of their lifetime losses by 24 months on book, a new vintage's results at 6 or 12 months can be compared with the older curves at the same age to estimate where it is heading. Lenders use this kind of projection in budgeting, pricing and loss allowance work, usually alongside economic assumptions.

These projections need care. A new vintage may differ from older ones in borrower mix, loan terms or economic conditions, so a curve that fit past vintages may not fit the new one. Good practice is to show the projection as a range, compare it with actual results each month as the vintage ages, and explain any gap. Keeping numerators and denominators visible lets reviewers check that an early rate is not based on too few accounts to mean much.

Aarvion Risk OS Retail Credit includes a performance explorer for cards, auto and personal loans that shows vintages at equal months on book, with numerators and denominators shown for each rate. The same view covers trends and roll rates and migration, so analysts can move from a vintage signal to the delinquency flow behind it.

Credit line strategy testing lets teams compare a proposed line strategy before rolling it out, and investigations and approved monitoring plans keep follow-up on a weak vintage organized. Every AI step is checked against the bank's own rules, which can allow, hold or block it, or stop all activity, and each step is recorded.