The European Central Bank published an Occasional Paper analysing whether banks’ internal probability of default (PD) estimates reflect signals of earnings manipulation for publicly traded firms. Using the Beneish M-Score as a proxy for potential manipulation, the paper finds a weak negative relationship between M-Scores and PDs in the full sample, suggesting manipulation signals are not consistently incorporated into banks’ internal credit risk assessments and may be captured only with delay. The authors link AnaCredit credit exposures (2019-2022) with Orbis financial statements, computing M-Scores for 4,659 listed corporates and matching PD information for 1,349 firms, yielding 9,681 firm-bank-year observations across 51 creditors. Only 8.9% of observations exceed the Beneish manipulation threshold (-1.78). For this subset, PDs are higher and increase as the M-Score worsens beyond the threshold, implying a non-linear relationship that becomes visible only when manipulation signals are strong. Country and sector breakdowns show higher exposure-weighted PDs in Portugal, Luxembourg and Greece, and in sectors including real estate, financial and insurance activities and construction. On supervisory implications, the paper suggests internal credit risk estimates may need qualitative overrides and expert judgement where manipulation signals are not adequately factored into models. It also introduces a manipulation-adjusted indicator, PD+, intended to show how manipulation risk, proxied by probability of fraud (PF), can amplify default risk.
European Central Bank 2026-04-07
European Central Bank research proposes a PD+ indicator after finding banks’ PD models do not fully absorb earnings manipulation signals
The European Central Bank published an Occasional Paper finding that banks’ internal probability of default estimates for listed corporates do not consistently incorporate earnings manipulation signals, as proxied by the Beneish M-Score, and may reflect them only with delay. The analysis shows default probabilities rise meaningfully only when manipulation signals are strong, with higher exposure-weighted probabilities of default in certain countries and sectors. The paper highlights potential supervisory needs for qualitative overrides and introduces a manipulation-adjusted indicator, PD+, to show how fraud risk can amplify default risk.