Technology-driven analytics in mortgage-backed securities for single-family mortgage financing

Technology-driven analytics constitute an important means in the analysis of residential mortgage-backed securities for the single-family mortgage finance market. The analytical focus of this paper is to provide a succinct discussion of technology-driven analytics deployed to enhance mortgage-backed security cash flow projection accuracy and performance, traditional and Bayesian model creation and the deployment of supervised machine learning support vector machines to forecast mortgage termination events. Following these discussions, we provide a summary of a representative CASE study: The algorithmic specification of econometric models and implementation of supervised machine learning support vector machines to forecast mortgage termination event probabilities for use in prepayment modeling. With their large dimensionality, responsive specification tuning capability, and proven track record with respect to prediction accuracy, time series-based hazard probit models estimated for the various calendar months of the various event-leading external economic process signals, and probit hazard function/shock conversion specification optimized using actual historical census population estimates, continue to serve as key reference indicators. Successfully employed in past prepayment prediction studies during similar economic periods, these time series models thereby support the prepayment modeling process. Investment in technology-driven analytics is made only when the management of the agency responsible for the creation/sale of the guarantee on the securities are convinced that the promised cash flow characteristics from the mortgage-backed security cash flows represent an extremely high-duty cycle. As technology-driven analytics become more widely used by issuers, traders, institutional and retail investors, and analysts, this increasing awareness raises a prescient question: What impact does the growing use of different modeling techniques have on the accuracy and performance of cash flows for bond valuation and portfolio management purposes? Who has the correct answer to this question?