Description:
We employ a sample of 10,136 Italian micro-, small-, and mid-sized enterprises (MSMEs) that borrow from 113 cooperative banks to examine whether market pricing of publicly listed firms adds additional information to accounting measures in predicting default of private unlisted firms. Through a novel public-private firms’ mapping approach, first we match the asset prices of listed firms following a data-driven clustering by means of Neural Networks Autoencoder so as to evaluate the firm-wise probability of default (PD) of our sample of unlisted MSMEs. Then, we adopt three statistical techniques, namely linear models, multivariate adaptive regression spline and random forest to predict the firms’ default, comparing the models’ performance when using accounting measures predictors only and when including the firm-wise PD. Finally, we make use of Shapley values to explain the relevance of each predictor. We find a significant improvement in model performance when including the estimated PD in the predictive specifications, suggesting banks to expand the spectrum of information used in private firms’ credit risk assessment. Our results carry important policy implications for financial institutions and policy makers as they provide a tool to mitigate issues related to the informational opacity of MSMEs due to their poor information disclosure, while reaching accurate credit risk assessment. Our findings assume even greater relevance due to the incorrect accounting practices adopted by SMEs following the COVID-19 pandemic