On the Improvement of Default Forecast Through Textual Analysis

Area of Science:

• Financial Risk Management
• Data Science
• Computational Linguistics

Background:

• Traditional models for predicting account default rely on limited financial variables.
• Textual data from account transactions offers a rich, underutilized source of information.
• Integrating qualitative insights from text can enhance predictive accuracy.

Purpose of the Study:

• To augment conventional default drivers with novel text-based variables.
• To classify bank account transactions into qualitative macro-categories.
• To assess the predictive power of client profiles derived from textual analysis for default risk.

Main Methods:

• Application of textual analysis techniques, including ad hoc dictionaries and distance measures.
• Classification of individual account transactions into predefined macro-categories.
• Utilization of supervised classification models to evaluate predictor effectiveness.

Main Results:

• Text-based variables derived from transaction data can be effectively generated.
• Client profiles based on transaction text show potential as default predictors.
• The proposed methodology offers a new approach to enhancing financial default prediction.

Conclusions:

• Textual analysis provides valuable, complementary information for default risk assessment.
• Classifying user behavior through transaction text can lead to more accurate client profiling.
• This approach offers a promising avenue for improving the robustness of financial risk models.