Dr. Tom Davis joined FactSet in 2014 as the Director of Fixed Income and Derivatives Research. Before joining FactSet, he led a team of quants at FINCAD and subsequently oversaw the flagship product at Numerix. At FactSet, Tom is focused on ensuring FactSet is providing the highest quality fixed income analytics and growing the coverage across all asset classes. His team also conducts cutting edge research in the models and methods of quantitative finance, recently investigating how machine learning techniques can be applied to produce analytics, or to provide insights into the financial markets. In 2006, Tom earned a Doctor of Philosophy in theoretical physics from the University of British Columbia in Vancouver, Canada.
Machine Learning and Mortgage Analytics: Predicting Defaults in Agency Pools
The mortgage market in the US is the perfect candidate for the application of machine learning techniques, being such a sizeable market with vast amounts of data open to the public. In the US, mortgages are bundled, sold by the Government Sponsored Agencies (GSEs), who guarantee the investors against mortgage defaults. Typical agency prepayment models include many effects, but do not implicitly consider defaults due to this guarantee. We have shown that applying machine learning techniques produces good predictive power on defaults in agency pools. In this talk I will briefly introduce the mortgage markets, define the problem that we are solving, and show the results of the study. Given the successful outcomes, more ambitious projects are planned and will be discussed.