Predicting Student Loan Debt: A Hierarchical Time Series Analysis
In recent years, and especially in response to the Covid-19 pandemic, much attention has been brought to the issue of rapidly increasing student debt. Yet in the field of time series analysis, there is a dearth of studies examining trends in student loan debt. This is likely due to the impression of simple, yet steep, linear increase in student loan debt over the last decade. However, trends in this type of debt are much more complicated when a complete picture of the hierarchical nature of this data is considered. One objective of this project is to generate accurate forecasts with the impact of Covid-19 in mind not only for outstanding student loan debt, but also for sub-categories of this value based on loan status: such as loans in default or loans in repayment. To facilitate this, traditional hierarchical forecasting methods were compared to newer methods, namely MinT and its recent adaptations. Our findings indicate that MinT forecast reconciliation with the use of structural scaling results in the mostaccurate forecasts across the aggregation structure. Although not the main focus of this study, the second-level forecasts indicate a forecasted 3 1% increase in default rates between the first quarter of 2020 the second quarter of 2022 and a 12% decrease in dollars outstanding for enrolled students during the same time period.