Low Default Portfolios – From the Usefulness of Pluriannual Data to the Inconsistency of Multi period Estimation

Estimating conservative default probabilities is crucial when banks opt to utilize an internal ratings‑based strategy to calculate their capital requirements. At least five years of historical data should be embraced when adopting internal models, according to the Basel Committee on Banking Supervision. This paper calls for a conceptual shift between pluriannual data and multi‑period estimation. It is shown from a variety of theoretical and practical perspectives that default probabilities computed using the multi‑year process do not reflect the real‑world banking business and are unrealistic or imprudent when the classical or Bayesian approaches are implemented. Different time periods and data aggregation methods are applied in such approaches to illustrate the inconsistency of multi‑period estimation. As a result, any financial risk management tool should refrain from employing the multi‑period methodology recommended by various authors for determining low default probabilities because the outcomes are not prudentially sound. Estimating annual default probabilities via time series (instead of estimating multi‑period default probabilities) is the most accepted practice for both the classical and Bayesian approaches, as detailed here. The conclusions remain the same whether the occurrence of default events follows a binomial distribution or a Poisson distribution. JEL classification numbers: C11, C53, C81, D81.