The release of 2011Q2 data marks the second full update of our CreditForecast 4.0 forecast. The combination of these improvements has produced more robust forecasts of performance and volumes that are better aligned with expectations. We will continue to vet the forecast database and refine our models as we explore the nuances of the dataset.
Among the improvements are:
Introduction of non-linear terms to better capture certain relationships between economic drivers and outcomes.
Improved specification of linked (a.k.a. "daisy-chained") models. For example, "90 day delinquency" is a function of "lagged 60 day delinquency."
Improved specification of economic drivers in performance models: Lags, non-linear transformations, etc.
Improved origination volume forecast models by credit score band and geography with stronger correlations to economic conditions.
Refined assumptions to account for the delay between loan origination and performance reporting to the credit bureaus.
A refined error-correction component. Core models define long-term trends and correlations with economic factors. An error-correction component considers model forecast performance during the most recent periods and adjusts the forecasts going forward. For example, if debt modification programs distort the long-run relationship between unemployment and delinquency, the error-correction component will account for this in the near term, gradually transitioning to the longer-term forecast.