I have to challenge somebody else's number
An independent recomputation of a client's allowance, and the two tests that actually find a bad model: whether it has ever been right, and whether the movement in the allowance is explained by the book or by somebody quietly changing the model.
ECL is the estimate most likely to be materially wrong and the hardest to challenge with a sample. Rebuilding the client's model in a spreadsheet to test it costs more hours than the area is budgeted, so it usually gets tested by enquiry.
It is not an audit methodology and it will not tell you whether an estimate is reasonable. It recomputes, it back-tests, and it shows you what moved. The judgement stays yours.
In the order you would meet them. None of them needs another to have been run first, so you can take one and leave the rest.
- 01Has this model ever been right?Back-test against a period that has already finished.
- 02Why did the allowance move?The book separated from the model, each period rebuilt on the model in force then.
- 03Recompute it yourselfSame tape, your assumptions, and the difference itemised.
- 04Does it tie to the accounts?Differences kept as named exceptions, never netted away.
Produce the number. Keep the proof.
Free to use, and nothing to install. Your file is read in this browser and never uploaded. An account is needed only to put a model change under review.