Net flow rates versus roll rates as non-performing consumer loans forecasting methodologies

Abstract

Roll rates and net flow rates can be seen as the evolution of ageing of accounts receivable and Markov chains. They are accepted methodologies to model the behavior of non-performing consumer loans by buckets and to predict losses, but we find that quite often they are wrongly used as interchangeable concepts, although roll rates track individual accounts across buckets in consecutive months and net flow rates just compare consecutive buckets in consecutive months. We determine their matrices of transition probabilities and analyze them in both stationary and steady-state conditions. Net flow rates have many advantages over roll rates, but a quite important finding for financial institutions and supervisors is that historical flow rates are not conservative for forecasting: when the level of new delinquencies soars, contemporary flow rates will tend to be lower than they would be in steady-state conditions, creating a feeling of false confidence and leading to the underestimation of future losses.