The views and opinions expressed in this article are those of the thought leader as an individual, and are not attributed to CeFPro or any particular organization.
By Aruna Joshi, VP, Model Risk Management, Visa
What challenges arise with building models on historic data?
There’s a saying – “garbage in garbage out”. This couldn’t be more applicable to model risk. Using historical data for building models also contradicts the warning given by financial advisors – “past performance is no guarantee of future results”. Take an example of a credit model where you want to determine the probability of default. If the only historical data used for modelling is from benign economic times with extremely low default rates, your models may be too lenient and may not be able to predict defaults if the economic situation suddenly deteriorates. Similarly, if you are trying to model prepayments of mortgages and the only time period used for modelling is during low interest rates, the model may not pick up the relationship between prepayments and falling interest rates. Unfortunately, the best data available for model building is still historical data. Hence, the biggest challenge for using historical data is to ensure that the time period covers a full economic cycle that includes a recessionary period. Ensuring that the data is representative of the situations one is trying to model is paramount. This will mitigate the risk of inadequate model performance however will not eliminate it.