Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Explainable Artificial Intelligence for identifying profitability predictors in Financial Statements

Marco Piazza, Mauro Passacantando, Francesca Magli, Federica Doni, Andrea Amaduzzi, Enza Messina

TL;DR

This study investigates predicting one-year profitability changes for Italian listed firms by merging ML on raw financial statements with SHAP-based explainability to satisfy EU AI regulations. It contrasts raw financial data against traditional financial ratios, finding high-dimensional raw features and tree ensembles yield superior predictive performance. XAI analyses reveal that Financial Profile features are consistently influential, and pruning noisy features with SHAP can maintain or improve accuracy, offering interpretable insights for auditors and investors. The work lays the groundwork for multi-modal extensions leveraging textual notes and larger datasets, highlighting practical implications for profitability forecasting and responsible AI deployment in finance.

Abstract

The interconnected nature of the economic variables influencing a firm's performance makes the prediction of a company's earning trend a challenging task. Existing methodologies often rely on simplistic models and financial ratios failing to capture the complexity of interacting influences. In this paper, we apply Machine Learning techniques to raw financial statements data taken from AIDA, a Database comprising Italian listed companies' data from 2013 to 2022. We present a comparative study of different models and following the European AI regulations, we complement our analysis by applying explainability techniques to the proposed models. In particular, we propose adopting an eXplainable Artificial Intelligence method based on Game Theory to identify the most sensitive features and make the result more interpretable.

Explainable Artificial Intelligence for identifying profitability predictors in Financial Statements

TL;DR

This study investigates predicting one-year profitability changes for Italian listed firms by merging ML on raw financial statements with SHAP-based explainability to satisfy EU AI regulations. It contrasts raw financial data against traditional financial ratios, finding high-dimensional raw features and tree ensembles yield superior predictive performance. XAI analyses reveal that Financial Profile features are consistently influential, and pruning noisy features with SHAP can maintain or improve accuracy, offering interpretable insights for auditors and investors. The work lays the groundwork for multi-modal extensions leveraging textual notes and larger datasets, highlighting practical implications for profitability forecasting and responsible AI deployment in finance.

Abstract

The interconnected nature of the economic variables influencing a firm's performance makes the prediction of a company's earning trend a challenging task. Existing methodologies often rely on simplistic models and financial ratios failing to capture the complexity of interacting influences. In this paper, we apply Machine Learning techniques to raw financial statements data taken from AIDA, a Database comprising Italian listed companies' data from 2013 to 2022. We present a comparative study of different models and following the European AI regulations, we complement our analysis by applying explainability techniques to the proposed models. In particular, we propose adopting an eXplainable Artificial Intelligence method based on Game Theory to identify the most sensitive features and make the result more interpretable.

Paper Structure

This paper contains 6 sections, 2 figures, 2 tables.

Table of Contents

  1. Methods: dataset, ML models and XAI approach
  2. Experiments and results
  3. Classification results
  4. XAI results
  5. XAI Validation
  6. Conclusions and Future works

Figures (2)

  • Figure 1: Frequency in Top 10 ranked features, normalized by the number of features composing a specific financial statement' part.
  • Figure 2: Comparative analysis of the distribution along the financial statements of the 50 most important and 50 less important features. The importance metric is computed with KernelShap.