EconPapers    
Economics at your fingertips  
 

HERA-XAI: explaining one-year-ahead corporate rating forecasts through hierarchical artificial intelligence

Greta Falavigna ()
Additional contact information
Greta Falavigna: CNR-IRCrES, Consiglio Nazionale delle Ricerche, Istituto di Ricerca sulla Crescita Economica Sostenibile, Economica Sostenibile, Strada delle Cacce 73, 00135 Torino, Italia, https://www.ircres.cnr.it

IRCrES Working Paper from CNR-IRCrES Research Institute on Sustainable Economic Growth - Torino (TO) ITALY - former Institute for Economic Research on Firms and Growth - Moncalieri (TO) ITALY

Abstract: Corporate rating forecasting models are often characterized by a trade-off between predictive performance and interpretability. While advanced machine learning algorithms can generate accurate forecasts, understanding the financial mechanisms underlying their predictions remains a significant challenge. This paper presents HERA-XAI, an explainable extension of the Hierarchical Artificial Intelligence framework for one-year-ahead corporate rating forecasting. Building upon the sector-specific hierarchical architecture introduced in HERA, the proposed framework combines Random Forest classifiers with a comprehensive explainability layer based on variable importance measures, Partial Dependence Plots (PDPs), Shapley Additive Explanations (SHAP), and simulation-based analyses. The analysis is conducted on a large dataset of Italian firms and considers three macro-sectors: Production, Trade, and Services. First, alternative probability threshold calibration criteria are investigated within the hierarchical classification process. Second, predictor importance rankings are extracted across the hierarchy to identify the financial indicators most relevant to rating assignment. Third, PDPs are employed to characterize the direction, magnitude, and non-linear effects of key financial variables on predicted outcomes. Fourth, local and global SHAP analyses are used to quantify the contribution of individual predictors to both firm-level classifications and overall model behaviour. Finally, scenario simulations are performed to evaluate how changes in firms’ financial conditions propagate through the hierarchical architecture and affect future rating assignments. The results show that the integration of hierarchical classification and explainable artificial intelligence provides transparent and economically interpretable rating forecasts while preserving predictive effectiveness. By combining forecasting, explainability, and simulation capabilities within a unified framework, HERA-XAI allows researchers, financial institutions, and policymakers to identify the financial drivers of rating transitions, understand the mechanisms underlying machine learning predictions, and explore the potential consequences of alternative financial scenarios.

Keywords: Corporate Ratings; Explainable Artificial Intelligence; SHAP; Partial Dependence Plots; Hierarchical Classification. (search for similar items in EconPapers)
Pages: 68 pages
Date: 2026-10
References: Add references at CitEc
Citations:

Downloads: (external link)
https://www.ircres.cnr.it/images/wp/wp_06_2026.pdf (application/pdf)

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:csc:ircrwp:202610

Access Statistics for this paper

More papers in IRCrES Working Paper from CNR-IRCrES Research Institute on Sustainable Economic Growth - Torino (TO) ITALY - former Institute for Economic Research on Firms and Growth - Moncalieri (TO) ITALY Contact information at EDIRC.
Bibliographic data for series maintained by Anna Perin () and Giuseppe Lanzafame ().

 
Page updated 2026-11-15
Handle: RePEc:csc:ircrwp:202610