Earnings manipulation signals weakly reflected in bank credit risk models

This ECB paper investigates whether earnings manipulation signals are reflected in banks' internal credit risk estimates, specifically the probability of default (PD).

Using AnaCredit data from 2019-2022 and financial statements from Orbis, the authors analyzed 4,649 publicly traded corporations.

Beneish M-Scores were computed to detect potential earnings manipulation.

The study found a weak and negative correlation between M-Scores and PDs for the full sample, suggesting that manipulation signals are not fully absorbed by internal models.

However, for the 8.9 percent of firms showing actual manipulation, PDs were higher and increased further as M-Scores worsened.

These findings indicate that the impact of earnings manipulation on credit risk is often deferred, rather than immediately captured by internal models, highlighting a context-dependent and non-linear relationship.

The paper highlights that earnings manipulation poses a substantial challenge for credit risk models, as distortions in financial data can lead to misclassification of a firm's risk profile.

Given that internal PD models often rely on historical financial data, the study suggests a potential need for these estimates to incorporate qualitative overrides and expert judgment when earnings manipulation signals are present but not adequately captured.

To enhance risk assessment, the authors introduce a manipulation-adjusted credit risk indicator, denoted as PD+, which quantifies how the probability of fraud amplifies default risk.

Cross-sectional analyses revealed consistent patterns linking default risk to M-Scores in specific countries and sectors, including Portugal, Luxembourg, Greece, and industries such as construction, real estate, and financial services.