A1397
Title: Discrete time-to-event regression analysis under left-truncation with applications to consumer finance
Authors: Jackson Lautier - Bentley University (United States) [presenting]
Jun Yan - University of Connecticut (United States)
Vladimir Pozdnyakov - University of Connecticut (United States)
Abstract: Asset-backed securities (ABS) play a vital role in financing American consumer automobile debt. Recently, economic analysis into ABS has benefited from the public release of data, which provides a new, rich source of loan level consumer auto loan information. Because this loan lifetime data is discrete-time and subject to random left-truncation, however, it is nontrivial to analyze. This has attracted recent study, but there is still no suitable approach to model this ABS loan lifetime data that incorporates regression coefficients for the lifetime of interest. The conditional, bivariate distribution is thus generalized to link to covariates, while keeping the left-truncation distribution unspecified. The high-dimensional, constrained likelihood-based parameter estimation problem is solved numerically, using a block coordinate descent design. Under suitable regularity conditions, the complete large sample, multivariate normal distribution of the estimators is provided. This allows for large sample inference into variable and model selection. All results are proven and verified through simulation studies. The methods are then applied in an economic study of borrower prepayment behavior for 1,553 consumer auto loans from the 2017-3 Ally Auto Receivables Trust ABS bond. It is found that borrowers with pick-up trucks prepay slower, all else equal, among other consumer finance insights.
