Probability of default (PD)

Probability of default (PD) is the estimated likelihood that a borrower — an individual, company, or counterparty — fails to meet a debt obligation within a specified time horizon, commonly one year for regulatory purposes. It is expressed as a percentage and is one of the three core inputs, alongside exposure at default and loss given default, used to calculate expected credit loss.

PD models are built with statistical or machine-learning techniques, often logistic regression or gradient-boosted trees, trained on historical borrower characteristics and repayment outcomes, and are closely related to but distinct from credit scoring: a credit score is typically a relative ranking tool, while a PD model outputs a calibrated probability meant to be used directly in loss and capital calculations. Under frameworks such as Basel and IFRS 9/CECL, banks must estimate PD to set regulatory capital and loan-loss provisions.

PD matters because it directly drives loan pricing, credit limits, and reserve requirements, and because errors compound: a systematically underestimated PD understates risk across an entire loan book. Model quality is commonly assessed with discrimination metrics like ROC AUC and is subject to model risk management review, since regulators require PD models to be validated, monitored for drift, and periodically recalibrated against actual default outcomes rather than trusted indefinitely.