Automated Valuation Models and Non-Agency RMBS Property Evaluation and Analysis

Abstract

This article explores various applications of automated valuation models (AVMs) in the evaluation and analysis of both loans and associated properties underlying non-agency residential mortgage-backed transactions and securities. In relation to available information only a few years ago, structured finance investors are now equipped to take a deeper and increasingly granular look into non-agency RMBS pools. In particular, the article discusses AVM property valuations in conjunction with ZIP indices as they relate to current and forward-looking surveillance activities on active, securitized loans. And it introduces the idea of a dealspecific home price time series, where such an index would track and aggregate property prices of active loans in a given transaction and could be used as a quantitative gauge of dealproperty over-, under-, or fair-valuation. Another analysis compares the results of current property valuations using ABSNet Loan HomeVal AVMs with those based on the FHFA and S&P/Case-Shiller indexes to show how substantial differences can be between the approaches. Finally, the article examines the relative statistical efficiency of the relationship between loan-level delinquency performance by ZIP code aggregation and state- and CBSA-level home price index time-adjusted valuations versus AVM-generated valuations.

TOPICS: MBS and residential mortgage loans, big data/machine learning, statistical methods