Greta Falavigna
\nAffiliazione: CNR-IRCrES, Consiglio Nazionale delle Ricerche, Istituto di Ricerca sulla Crescita Economica Sostenibile, Strada delle Cacce 73, 00135 Torino, Italia
\ncorresponding author: greta.falavigna@cnr.it<\/p>\n
Abstract<\/strong> Keywords:<\/strong> Random Forest, Corporate Credit Risk, Sectoral Analysis.<\/p>\n DOI:<\/strong> 10.23760\/2421-7158.2026.004<\/p>\n ISSN (online):<\/strong> 2421-7158<\/p>\n How to Cite<\/strong><\/p>\n Falavigna, G. (2026). HERA: Hierarchical Engine for Rating Assessment. A Sector-Specific Hierarchical Artificial Intelligence Framework for One-Year-Ahead Corporate Rating Forecasting (CNR-IRCrES Working Paper 4\/2026). Istituto di Ricerca sulla Crescita Economica Sostenibile. http:\/\/dx.doi.org\/10.23760\/2421-7158.2026.004<\/a><\/p>\n <\/p>\n","protected":false},"parent":0,"menu_order":0,"template":"","categoria-pubblicazioni":[40],"class_list":["post-11347","pubblicazione","type-pubblicazione","status-publish","hentry","categoria-pubblicazioni-working-paper"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/pubblicazione\/11347","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/pubblicazione"}],"about":[{"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/types\/pubblicazione"}],"version-history":[{"count":3,"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/pubblicazione\/11347\/revisions"}],"predecessor-version":[{"id":11350,"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/pubblicazione\/11347\/revisions\/11350"}],"wp:attachment":[{"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/media?parent=11347"}],"wp:term":[{"taxonomy":"categoria-pubblicazioni","embeddable":true,"href":"https:\/\/www.ircres.cnr.it\/en\/wp-json\/wp\/v2\/categoria-pubblicazioni?post=11347"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
\nThis study develops HERA (Hierarchical Engine for Rating Assessment), a sector-specific artificial intelligence framework for one-year-ahead corporate rating forecasting. The proposed methodology decomposes the original multi-class rating problem into a hierarchical sequence of binary classification tasks estimated through Random Forest models. Separate HERA architectures are developed for the Production, Trade, and Services sectors, allowing the framework to capture sector-specific financial characteristics and heterogeneous risk profiles.
\nThe predictive performance of the classifiers is evaluated through Out-of-Bag validation, while optimal classification thresholds are determined through F1-score maximization. Beyond forecasting accuracy, the framework provides information on the financial indicators associated with specific rating transitions through variable importance analysis.
\nThe empirical results show that the determinants of corporate ratings differ across both economic sectors and rating categories. Debt-related indicators emerge as key predictors for rating transitions involving the highest and lowest rating classes, whereas measures of financial sustainability and capital structure play a more prominent role in intermediate rating transitions.
\nOverall, HERA provides not only one-year-ahead rating forecasts but also insights into the sector-specific financial drivers associated with future rating outcomes. These findings may be of interest to firms, financial institutions, and policymakers seeking to better understand the determinants of future corporate creditworthiness.<\/p>\n