Machine Learning for the Unlisted: Enhancing MSME Default Prediction with Public Market Signals

Alessandro Bitetto, Stefano Filomeni and Michele Modina

Journal of Corporate Finance, 2025, vol. 94, issue C

Abstract: This paper contributes to the growing body of research on private firms, particularly private firm accounting. We explore the economic factors that drive improvements in the default prediction of unlisted private firms using peers’ market-based information. Specifically, we examine how the market-based default probability of a peer firm can provide valuable insights into the often noisy accounting data of private firms. Our analysis delves deeply into these economic issues to uncover essential insights. To address our research question, we utilize a granular proprietary dataset of 10,136 Italian micro-, small-, and mid-sized enterprises (MSMEs) that are required to disclose their financial statements publicly. We propose a novel public–private firm mapping approach to investigate whether incorporating peers’ market-based information improves the accuracy of default predictions for private unlisted firms. Our mapping approach matches the market information of listed firms with private firms through a data-driven clustering technique using Neural Network Autoencoder. This method enables us to link the Merton Probability of Default (PD) of public peers to the corresponding private firms within the same cluster. We then apply five statistical techniques – linear models, multivariate adaptive regression splines, support vector machines, k-nearest neighbours and random forests – to predict corporate default among private firms, comparing model performance with and without the inclusion of Merton’s PD estimated using peers’ market-based information. To assess the contribution of each predictor, we employ Shapley values. Our results demonstrate a significant improvement in default prediction for unlisted private firms when incorporating peers’ market-based information, confirming that the noisy accounting data of private firms alone hinders accurate default prediction. Furthermore, our findings highlight the importance for banks to broaden the scope of information used in credit risk assessments of private firms. These results have important policy implications for financial institutions and policymakers, providing a tool to mitigate the challenges posed by the noisy information disclosure of MSMEs while ensuring more accurate credit risk assessments.

Keywords: Credit risk; Distance to Default; Machine learning; Market information; Probability of Default; Shapley; XAI (search for similar items in EconPapers)
JEL-codes: C52 C53 D82 D83 G21 G22 (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S0929119925000987
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:corfin:v:94:y:2025:i:c:s0929119925000987

DOI: 10.1016/j.jcorpfin.2025.102830

Access Statistics for this article

Journal of Corporate Finance is currently edited by A. Poulsen and J. Netter

More articles in Journal of Corporate Finance from Elsevier
Bibliographic data for series maintained by Catherine Liu ().