Social, governance data refine small firm default risk

The Banca d'Italia paper investigates the predictive power of social (S) and corporate governance (G) indicators in assessing default risk for Italian non-financial firms within its In-house Credit Assessment System (ICAS).

S indicators leverage data from the Italian National Social Security Institute (INPS), while G indicators are derived from a proprietary knowledge graph (KG19) detailing corporate relationships and governance traits.

The study grouped firms by size and employed feature selection, including random forests, to identify statistically significant indicators.

Key workforce-related metrics, such as layoff rates, employee turnover, and average wages, consistently demonstrate significant predictive power across all firm sizes.

Governance indicators, however, show varied contributions by firm size; board age diversity is a significant predictor for all classes, but board size is relevant only for micro and small firms.

The augmented S-ICAS model, integrating these S and G scores, improves the discriminatory power of the baseline model, with AuROC increases of 0.5-0.9 percentage points for micro and small firms.

This enhancement is notable given the high baseline AuROC.

The research adopts a purely predictive perspective, aiming to improve default risk assessment without evaluating firms' social or governance quality.

The default definition aligns with EU Regulation No 575/2013, identifying default when an obligor is unlikely to pay or is past due more than 90 days, making it generally reversible.

While governance information is typically incorporated qualitatively in ICAS expert assessments, this paper represents a novel attempt to include a wide range of S and G indicators statistically.

The overall contribution of these indicators to the model's performance is moderate.

For medium-sized firms, only S indicators are statistically significant, with the governance score showing no significance.

For large firms, neither the governance score nor the AuROC increment proves significant, a finding potentially attributable to the low number of large companies and infrequent default events.