Where do your rates come from?
Turn your default and loss history into model assumptions. Choose a method below, then replace the example with your own observations.
What happened, by grade
Best grade at the top. The order is the model — each grade's estimate is taken over itself and everything worse than it.
| Grade | Borrowers | Defaults | Observed rate | |
|---|---|---|---|---|
| 0.00% | ||||
| 0.00% | ||||
| 1.11% | ||||
| 6.32% |
These are example figures, not your data — a book where the two best grades have never had a default. Type over them.
Your probabilities of default
An upper confidence bound, not a best estimate. Deliberately conservative, and describe it that way wherever it reaches a disclosure.
| Grade | Observed | Borrowers used | Defaults used | PD to use |
|---|---|---|---|---|
| A | 0.00% | 1,005 | 8 | 1.290% |
| BBB | 0.00% | 585 | 8 | 2.212% |
| BB | 1.11% | 275 | 8 | 4.682% |
| B | 6.32% | 95 | 6 | 10.823% |
Each grade is measured over itself and every worse grade — that is why the borrower counts fall as you go down. A grade with no defaults of its own cannot be measured on its own, but a scale ordered by credit quality says its rate is at most that of everything below it, and that is enough. It also means the result cannot come out with a better grade riskier than a worse one, which an ungrouped calculation does whenever the top grade is small.
Put these rates into the model
Written into one sector's through-the-cycle curve, with this calibration recorded as its basis. It becomes a draft change, reviewable and reversible like any other.
These are still the shipped example figures. Type your own default history into the table above and this becomes available — demonstration numbers must not enter a model under a basis that calls them your experience.