Reading the credit-loss section

View as Markdown Open in Claude Open in ChatGPT

The Credit Loss section shows how much of your cohort’s original balance (the cohort being the loans your segment selects) has been lost to charge-offs at each loan age, and where that path is heading by the end of the loans’ term. A charge-off is the write-down of a loan’s balance as uncollectible; Vintage measures it net of any recoveries you report. The section answers one question: how many cents of every dollar this kind of loan was originally written for are lost over its life.

Each mark on the section is explained below. The measurement itself is defined in full in Credit loss measurement.

The cumulative loss curve

The chart’s horizontal axis is Loan age (months): months since each loan was originated. The vertical axis is Cumulative loss (%): the share of the cohort’s original balance lost to net charge-offs by that age.

The curve has two parts:

Solid where your data covers the cohort. This is measured experience.
Dashed past the last age your data covers, labeled (projected). This is a projection to the end of the cohort’s term, described below.

Where the solid part ends is set by how many loans back each age, not by the last row your files happen to contain. The curve is drawn as observed through the last age backed by enough loans to be reliable (at least a tenth of the largest number of loans backing any age), and projected past it. A thin tail of ages beyond that point, backed by a handful of loans, is replaced by the projection rather than drawn as measured. Short interior gaps of up to three months between well-backed ages are bridged on a straight line and drawn dashed.

The curve ends at the cohort’s contractual term, its balance-weighted average term rounded to whole months. A cohort observed past its own term contributes nothing beyond it.

How the curve is built, in brief

At each age, Vintage divides the net charge-offs booked at that age by the original balance of every loan old enough to have reached that age by the as-of date, including loans that have since paid off or charged off, so a young cohort that has not yet had time to take its losses cannot bias the lifetime figure downward. The cumulative curve is the running sum of those per-age rates, so more data can never make it fall. The formula, and the rules for which loans count at which ages, are in Credit loss measurement.

All of this is measured against one as-of date, the latest snapshot month across your whole portfolio, so every segment you model sits on the same clock. See Vintage analysis.

The lifetime loss headline

Above the chart, Expected lifetime loss is the cumulative curve read at the cohort’s contractual term. Its caption reads Cumulative loss at the contractual term. The same figure appears in the screen’s header as Expected Lifetime Loss.

It is a percentage of original balance, not current balance. A 1.50% lifetime loss on a cohort that lent $10 million means about $150,000 of net charge-offs over the cohort’s life.

The shaded range

The shaded area around the curve is labeled for what it is, and it is two different things:

Over the solid part, it is a ~95% interval: a statistical range that widens where the data is thin. It also widens on a concentrated book, because a few large loans carry more uncertainty than many small loans of the same total, and it widens again where the month-to-month loss rates are lumpier than the number of loans behind them would explain. Each widening can only make the range wider, never narrower than the plain count-based interval.
Over the dashed part, it is a Projected range: a widening fan that shows the projection grows less certain with distance. It is not the same statistic as the interval, and it is never labeled as one.

No range ever runs past 100%, because a cumulative loss share cannot exceed everything that was lent. The vertical axis follows the curve itself, with modest headroom. When a range fans out higher than the chart, as a long projection on a young cohort will, it is cut off at the top and the legend says Range continues above the chart. Tap or hover a point for its full range.

The bars behind the curve

The faint bars are Loans with data at this age: how many loans in the cohort were old enough to have reached that age by the as-of date. These are the loans whose original balances form the denominator at that age, so the count includes loans that had already paid off or charged off by then, and a loan that left your book is counted only through the age it left. The bars show how much experience backs each point. Hover or tap any point to see its age, its value, its range, and its loan count.

The chart carries no sentence describing a single point. The per-age counts are on the bars and in each tooltip.

The dashed part is a projection, and says what it assumes

Past the last age the cohort reliably observes, Vintage does not invent a loss rate or fit a trend line. It starts from the cohort’s own measured remaining balance at that age (the outstanding balance of the loans in the denominator, as a share of their original balance) and carries it forward month by month to the contractual term. Each month books the cohort’s measured net loss rate on the balance still outstanding, never more than the principal that defaults that month, and the balance then shrinks by what charged off, what the schedule retired at the cohort’s average note rate, and what prepaid. The default and prepayment rates are the cohort’s own, continued past the edge of the data by easing toward their long-run levels. Losses taper in the tail because the balance runs off, not because a curve was bent toward zero.

That walk has to assume one thing: how loans repay. Vintage assumes level payments (equal monthly payments that retire the balance by the end of the term), because Vintage has no field that says whether a loan is interest-only or balloon. The sentence under the chart says so:

Past the last age your data covers, the dashed line is projected: this cohort’s measured remaining balance carried forward at its own measured loss, default and prepayment rates, month by month, to the contractual term. The projection assumes loans repay on a level-payment schedule. If this cohort holds interest-only or balloon loans, their balances stay outstanding longer than shown.

Two further notes can appear under the chart:

When prepayment could not be measured for the cohort, the projection assumes none, and the note says which of two reasons applies: no loan in the cohort reports scheduled principal, or the cohort does not yet have two consecutive months of data. Because loans that pay off early would have left less balance exposed, the projected loss is likely on the high side. See Reading the prepayment section.
When there is no balance to carry forward, the note says the curve could not be projected past a given month and that the lifetime figure is the measured loss through that month only, not a full-life projection.

A cohort observed all the way to its term has nothing to project, and the projection notes do not appear.

Vintage method or WARM fallback

A badge beside the section heading names the method that produced the curve:

Vintage method: the curve described above, built from per-age measured rates.
WARM fallback: used when the cohort is too young or too thin to support a vintage curve, which means one of two things: the cohort has been observed for less than half of its contractual term, or fewer than five loans back it at its best-covered age. WARM (weighted-average remaining maturity) is the same walk with one constant rate in place of the measured curve: the cohort’s average charge-off rate on its outstanding balance, meaning the net charge-offs it took over all the months you reported divided by the balance those loans were carrying at the start of each of those months. The loss the cohort has already taken stays as the observed part of the curve; only the part past the last observed age is carried at the constant rate.

A WARM curve is drawn with no range around it, on observed ages and projection alike. A band around one rate held constant would describe a spread nothing measured. The note under a WARM chart says in plain words that the cohort’s average charge-off rate so far stood in for a loss curve.

“Thin history, directional only”

When the lifetime figure rests on very little, the headline says so: Thin history, directional only. That happens in two cases:

the WARM fallback was used, or
five or fewer loans back the deepest age the curve draws as observed (the end of the solid line).

The qualifier applies to any figure, not only a worrying one. A 1.20% resting on one loan is as directional as a 0.00% resting on one.

A dash, a measured zero, and the difference

Vintage keeps two very different situations apart.

No charge-off data at all. If none of your organization’s uploads supplies a charge-off amount, credit loss cannot be measured. The headline shows a dash with No charge-off data uploaded, and the section shows No charge-off data in your uploads in place of the chart. It says plainly that this is missing data, not a measured 0% loss rate, and names both remedies: upload charge-off amounts in your snapshots or a transaction file, or, if your transaction file reports charge-offs with a type code, classify that code as a charge-off on the Fields screen. A Net Charge-Off Amount or Charge-Off Amount column and a classified transaction ledger count equally.

A measured zero. If your data does carry charge-off amounts and this cohort genuinely lost nothing, the section’s headline says so in words: No charge-off dollars measured against this cohort. The header figure reads No charge-off dollars measured in this cohort. Vintage withholds that affirmation when some loans in the cohort were marked charged off by a flag or status with no amount, or arrived in your data already closed. There the zero is partly unknowable, and the screen claims neither a clean book nor missing data.

A cohort whose curve is zero at every age is drawn on a full-scale axis, 0 to 100%, so a flat zero reads as zero instead of being stretched to look like a loss.

What the curve leaves out, and why

Several rules decide which loans and months the curve counts. Each has its own home page; in brief:

A loan marked charged off by a flag or status with no loss amount is left out of the loss curve rather than counted as a $0 loss, which would understate losses. See Charge-offs and recoveries.
A loan that was already seasoned when your data begins and has no reported original amount is left out of this per-original-dollar curve only. Its losses still count in the rates measured against outstanding balance, which drive the projected part of the curve and the Pricing section. Supplying an Original Loan Amount column brings it in. See Credit loss measurement.
A loan that left your book through a sale, participation, or transfer counts toward loss experience up to the age it left and no further. The month it left still counts as a month at risk, because the loan was exposed at the start of it and did not default. See Payoffs and exits.
A loan that arrived already closed (its ending is on the first row Vintage ever sees for it) books no loss at that age, because it was never observed while open. See How a loan’s ending is decided.
Months before a loan first appears in your data are never reconstructed. When every loan in the cohort entered your data mid-life, the earliest ages cannot be measured for that cohort and stay dashed. See Missing and unreported data.
Months reported after a loan’s ending are excluded, and a later true-up books at the ending’s age. See How a loan’s ending is decided.