As inflation numbers are rising to levels not seen since the seventies in the Netherlands, Point-in-Time models trained on the most recent economic cycle will yield unexpected outcomes. This problem was first encountered exactly two years ago during the Covid period when unemployment and GDP forecasts went to extreme levels.
IFRS9 PD models, that must exhibit Point-in-Time behavior, are extremely sensitive to those changes especially when trained to only small shocks as was the case in 2014-2020. Besides large impact on IFRS9 Expected Credit Loss estimates, we foresee much attention to this question from the accountant for the half year figures. We therefore recommend to monitor the behavior of the PD estimates with respect to the macro-economic input variables constantly. Consider if the relationship between PD and for example CPI modelled previously still holds. One can argue if the relationship should be linear or if it should be capped at a certain level. A management overwrite can be considered for the time being to eliminate extreme shocks.
At RiskQuest we have experience in developing Point-in-Time models, but also understand the unknown future part. Sometimes historical data is not representative.
Feel free to contact [email protected] for more information on this topic.