A1521
Title: abslife: Estimating discrete left-truncated lifetime random variables with applications to asset-backed securities
Authors: Lucas da Cunha Godoy - University of California Santa Cruz (United States) [presenting]
Jackson Lautier - Bentley University (United States)
Abstract: The consumer auto lending market in the United States operates on a massive scale: consumer auto asset-backed securities~(ABS) issuance tops $200 billion and total consumer automobile debt exceeds $1,400 billion. Despite this scale, there is a lack of ready-to-use software implementing statistical methods for drawing inference on the lifetime distribution of individual consumer loans sampled from ABS. Such tools would benefit ABS investors because the source of the ABS level cash flow is the sum of its underlying individual assets. Hence, investors require precise time-to-event distribution estimates to accurately model ABS trust level performance. Furthermore, the convenience of statistical software designed for this purpose would be valuable in the high-paced environment of fixed-income trading. The ABS data is nontrivial to analyze, however, and a review is provided of statistical methods for left-truncated, right-censored, discrete time-to-event data, including competing risks. Next, asymptotic distributions for the cumulative and probability mass functions for the time-to-event distribution estimates are introduced. The abslife package for R is then presented, designed specifically to analyze such data. The package's utility is illustrated by analyzing 275,948 consumer auto loans spanning four distinct ABS bonds, demonstrating how abslife enables financial analysts to perform robust inference for discrete time-to-event data.
