Skip to content

How Vintage Loss Curves Work

Vintage loss curves answer a cohort question: how much cumulative loss has a group of loans produced as it ages, and how much loss is still likely to emerge? They are useful when cohort-level history is available before a full transition or cashflow model is justified.

Why Curves Often Have An S-Shape

Newly originated loans usually produce little observed loss immediately. Defaults emerge after borrowers season, then cumulative loss grows more slowly as the cohort pays down and the remaining exposure becomes smaller. The result is often an S-shaped cumulative curve: a slow start, a steeper middle, and a flattening tail.

Why Fit A Parametric Curve

Raw interpolation can describe the observations you already have, but it does not provide a stable tail for an incomplete cohort. Parametric families such as Weibull and Gompertz encode a smooth shape with an asymptote. That makes the forecast easier to compare across vintages and less sensitive to one noisy month.

The package also supports smoothing comparisons. Start with vintage.run() when you want ranked methods, validation, and optional tail projection. Use CurveFitter directly when you specifically need one parametric family.

What Ultimate Loss Means

Ultimate loss is the fitted asymptote of the cumulative-loss curve. It estimates the total cohort loss rate after enough time has passed for the curve to flatten. For an immature vintage, the difference between observed cumulative loss and ultimate loss is the remaining tail estimate.

Warning

Ultimate loss is a cohort-level approximation. It is not a substitute for loan-level amortization, prepayment, or scenario-dependent cashflow modeling.

Choose The Right Workflow

Starting point Use this workflow
Cohort loss history by month on book Vintage Analysis
Portfolio snapshot plus transition assumptions or loan history Lifetime Loss Forecasting
Loan-level balances and roll-rate cashflow assumptions Portfolio Simulation
Monthly aggregated payments, charge-offs, and balances Rollforward Workflow

Vintage analysis is the simplest useful choice when the business question is about cumulative cohort loss. Move to Lifetime Loss Forecasting, simulation, or Rollforward when the question depends on state migration, cashflows, or operating diagnostics.