Publication Date

Winter 12-9-2025

Document Type

Student Project

First Advisor

Dr. Michelle Xia

Degree Name

B.S. (Bachelor of Science)

Department

Department of Statistics and Actuarial Science

Abstract

This project models private household debt among U.S. consumers using data from the Consumer Expenditure Survey (CES) between 2013 and 2023. The analysis focuses on identifying how demographic and economic characteristics, such as income, housing expenditures, education, and occupation, relate to non-mortgage “other” loan balances. After initial model development produced poor residual behavior due to zero-inflation from imputed debt values, the analysis was refined to include only households reporting verifiable debt. Multiple modeling techniques, including AIC-based variable selection and Lasso regularization, were compared under a five-fold cross-validation framework. The Lasso model achieved superior predictive accuracy (RMSE = 1.55, MAE = 1.18), while the AIC model provided greater explanatory depth and slightly better residual diagnostics. In this restricted sample of debt-reporting households, housing outlays, education, occupation, housing tenure, and several interactions with age and year emerged as consistent predictors of other-loan balances. Both models demonstrate statistically sound behavior and support cautious inference about the demographic and financial factors associated with private consumer debt.

Download

Share

COinS