Modelling Net Loan Loss with Bayesian and Frequentist Regression Analysis

TL;DR

This paper develops and compares Bayesian and frequentist nonlinear regression models to analyze net loan loss, incorporating financial, sociological, and temporal data to enhance prediction accuracy.

Contribution

It introduces two distinct nonlinear regression models for net loan loss, integrating temporal variables and comparing Bayesian and frequentist approaches for better understanding.

Findings

Both models improve prediction accuracy over simpler models.

Bayesian and frequentist methods provide complementary insights.

Temporal variables significantly influence net loan loss predictions.

Abstract

We create two distinct nonlinear regression models relating net loan loss (as an outcome) to several other financial and sociological quantities. We consider these data for the time interval between April 1st 2011 and April 1st 2020. We also include temporal quantities (month and year) in our model to improve accuracy. One model follows the frequentist paradigm for nonlinear regression, while the other follows the Bayesian paradigm. By using the two methods, we obtain a rounded understanding of the relationship between net loan losses and our given financial, sociological, and temporal variables, improving our ability to make financial predictions regarding the profitability of loan allocation.