What Vintage Modeling does
Vintage Modeling helps community banks and credit unions answer three questions from their own loan history: how much their loans lose to charge-offs, and when in a loan’s life; how much faster than the contract borrowers pay them back; and, given both, what rate the next loan needs to carry to be worth making. You upload the files your core system already exports. Vintage reconciles them into one portfolio, measures how slices of it have behaved, and prices a loan against that history.
What you put in
Section titled “What you put in”You upload loan-level files exactly as your core system produces them. Each file belongs to one of three file groups, and you say which when you add it. Vintage never guesses a file’s type.
| File group | What each row is | Typical contents |
|---|---|---|
| Loan Snapshots | One loan at one point in time (monthly, quarterly, or yearly) | Snapshot month, current balance, status, payment and principal figures |
| Origination Files | One loan as it was when it was made | Origination date, original loan amount, term |
| Transaction Files | One dated loan event | Charge-offs, recoveries, payoffs, partial prepayments |
You can upload any combination: snapshots alone, an origination file alone, a transaction file alone, or all three together. Your files do not need to match a template. Column names are whatever your core calls them, dates and amounts can be written in the usual ways, and columns Vintage does not recognize are kept rather than dropped. Vintage accepts useful partial data and tells you plainly what it can and cannot do with it. Preparing your data covers what to export.
The first upload is usually a backfill, meaning as much history as you have. After that, the usual rhythm is one new snapshot a month.
What you get back
Section titled “What you get back”A reconciled portfolio. Vintage joins every file you have uploaded into one record per loan, tied together by the Loan ID. Each loan carries its history of snapshot values, its transactions, and every column you uploaded, including the ones Vintage keeps as a custom attribute (a column it does not interpret but keeps so you can slice by it). The Portfolio screen shows every loan, and the Segment Builder narrows the book to the loans you care about: auto loans originated 2021 to 2023 with a FICO score under 680, for example.
Measured behavior for a segment. The Modeling screen takes the segment you defined and shows, by loan age, how those loans have actually behaved:
- Credit loss. A cumulative loss curve: the share of the cohort’s original balance lost to net charge-offs as the loans aged, with a headline expected lifetime loss.
- Prepayment. A speed curve, read as a single monthly mortality (SMM) or a conditional prepayment rate (CPR): how quickly principal came back ahead of schedule, through full payoffs and partial prepayments together.
Both curves are drawn solid where your data backs them and dashed where they are projected to the cohort’s term.
A price for a loan. In the same screen’s Pricing section you enter a loan amount, a term, your funding curve (the funds-transfer pricing, or FTP, curve: your institution’s cost of funds at each term), your costs and fees, and a return target. Vintage solves for the required rate, meaning the note rate at which this loan’s projected return meets your target. It shows that rate as a build-up of components (funding, expected credit loss, operating cost, fees, required return) and projects the loan month by month to its profitability. You can also enter the rate you plan to offer and see how far it sits from the required one.
When you upload, the Review step reports readiness for three modeled outputs: credit loss, payoff (a loan paid off in full), and prepayment (a partial prepayment). The Modeling screen then folds payoffs and partial prepayments into one prepayment measure, so it shows two behaviors: credit loss and prepayment.
The screens you will use
Section titled “The screens you will use”| Screen | What it is for |
|---|---|
| Upload | Start an upload and resume one. The page lists every past upload in Upload History. |
| Portfolio | Every loan in your reconciled portfolio, its status, and the Segment Builder. |
| Fields | Your fields and what they mean: types, value vocabularies, rate formats, and how amounts are reported. |
| Modeling | Credit loss, prepayment, and pricing for the segment you defined. |
| Settings | Your organization, your profile, your team, and (for admins) network security. |
Viewers do not see Upload or Fields. See Accounts and organizations for the three roles.
How the measurements work, in brief
Section titled “How the measurements work, in brief”Vintage measures by vintage analysis. A cohort (a group of loans you expect to behave alike) is lined up by loan age (months since origination), a rate is measured at each age, and the loss rates are accumulated into a curve as the loans season. This is the life-table idea. Each age’s rate is weighted by the loans actually exposed at that age, so a young cohort that has not had time to take its losses does not drag the long-run picture down. Rates are weighted by balance rather than by loan count, the convention used in current expected credit loss (CECL) work and in mortgage-market (SIFMA) analytics.
There is no machine learning, no fitted distribution, and no hidden parameter. Every number is meant to be re-derivable by a banker or an examiner from the cohort’s own figures. Vintage analysis explains the method in full.
Modeling is segment-first. The screen models a slice you define, not the whole book, because a curve that averages auto loans with mortgages describes neither. A segment is modeled up to a maximum size set for your organization, and the screen names that number if you go past it.
Measuring and projecting are kept apart. Everything Vintage measures from your history uses figures you reported. Anything it carries forward past the edge of your data (the projected part of the loss curve, the pricing projection) assumes a level-payment schedule, and the screen states that assumption beside every figure it affects.
The boundary Vintage holds
Section titled “The boundary Vintage holds”Vintage normalizes and reconciles the facts you supply, asks you only questions you can actually answer, and never manufactures an economically material fact. Where data is missing, it says plainly what it cannot measure and exactly what would unlock it, rather than inventing a value that would read as real.
In practice:
- A blank is unknown, never zero. A blank cell never overwrites a value Vintage already knows, and never counts as a zero balance or a zero loss.
- It never guesses what your codes mean. Status, flag, and transaction-type values mean an event only because you said so when you mapped them. See Classifying values.
- It never reconstructs an amortization schedule to measure anything. A prepayment speed needs a schedule figure you reported. Without one, the speed reads as a dash, not a confident 0%. See Prepayment speeds.
- It keeps its approximations apart from your facts. An origination date estimated from a loan’s first appearance, or a payoff inferred because a loan dropped out of your file before maturity, is never presented as something you reported. Review tells you before you finalize when origination dates will be estimated.
- An unmeasurable figure is left out, never priced at zero. If no charge-off amounts exist anywhere in your data, the credit-loss line is omitted from the price and you are told what that costs you.
Some things Vintage deliberately does not do. It applies no macroeconomic or qualitative overlay, has no refinance (rate-incentive) model of prepayment, and models two behaviors (credit loss and prepayment), not more. What Vintage does not do lists each boundary with its reason.
Your data stays protected
Section titled “Your data stays protected”Personal information never leaves your browser. Before anything is uploaded, your browser removes the columns you mark as personal and replaces the Loan ID and Tax ID with protected IDs: one-way scrambled values that cannot be reversed, which still let Vintage recognize the same loan in next month’s file. Vintage never receives a raw loan number or tax ID. See Removing personal information and Security and data protection.