Validation of PARX Models for Default Count Prediction.

Area of Science:

• Financial Mathematics
• Econometrics
• Risk Management

Background:

• The interconnected nature of financial technology (FinTech) platforms necessitates advanced credit risk measurement models.
• Evaluating the predictive accuracy of these models is crucial for investor protection and overall financial stability.

Purpose of the Study:

• To apply Poisson autoregressive stochastic processes to default data for investigating credit contagion.
• To assess the predictive accuracy of these models using standard metrics and a novel criterion extended for discrete data.

Main Methods:

• Application of Poisson autoregressive models with exogenous covariates (PARX) to quarterly default counts.
• Utilizing standard predictive accuracy metrics and a new criterion robust to variable type.

Main Results:

• The study extends a new predictive accuracy criterion to discrete data, yielding results consistent with classical measures.
• Poisson autoregressive models incorporating a contagion component demonstrated a decisive improvement in accuracy compared to autoregressive models alone.

Conclusions:

• The integration of contagion effects into credit risk models is vital for accurately assessing default probabilities in interconnected financial systems.
• The proposed methodology, using PARX models, offers a robust framework for evaluating credit contagion and enhancing model performance in FinTech contexts.