8.1%Real lending data from an open research dataset
The forecast that knew the policy changed
mean error in twelve month default counts, against 19.2% for last year's rate carried forward
- What was observed
- Twelve month default counts forecast at six past dates and checked against what happened, across a change in the lender's credit policy.
- What a generic approach says
- Carry last year's default rate forward: 19.2% out on average. One curve for the whole book did worse, at 66.9%.
- What the engine read
- Curves by term and grade, with their shape from all history and their level from the book written since the policy change, were 8.1% out. Curves from all history were 19.7% out and curves from the last 24 months 22.3%; the method was chosen after those failed on the same dates.
- What it meant for the decision
- A change of credit policy has to be declared to the forecast. No method's own interval held, so the planning range comes from past errors: about 16% either way, from only six forecasts.