The Pipeline

With primary mortgage rates having risen from the 3-3.5% range to the neighborhood of 6% in six months, the mortgage landscape has changed dramatically, with turnover becoming a greater focus than it has been in a long time. An important consideration into how mortgage rates and spreads evolve is likely to be the newly ramped up QT limits.

Over the course of the last two-plus years, we have seen unprecedented behavior in the mortgage market. Interest rates on mortgages fell to historic lows for an extended period of time, and then rose extremely rapidly to high levels not seen in many years. In this article, we will look at how the interest rate movement affected the composition of pools issued by Freddie Mac by specifically focusing on loan purpose at origination.

On January 26, 2022, the Federal Reserve Board (FRB) released “principles” for reducing its balance sheet size, followed by the release of “plans” for the same on May 4, 2022.

We revisit the PSS model designed in 2012. We attribute much of historical PSS dependence on the level of rates to (a) the dynamics of guarantee fees (G-fees) and (b) the compounding difference between primary and secondary rates. Net of those factors, the new model is much less directional. In addition, we modified the concept of “trend factor” by making it asymmetrical thereby reflecting originators’ profit-protecting behavior. The new model has been tested for different historical intervals, including the recent COVID-19 pandemic.

This article examines neural networks, which provide a general approach to evaluating a large data set with many potential input variables and complex non-linear interactions. Mortgage loan prepayments exhibit complex relationships to a wide range of variables and extensive data for tens of millions of mortgage loans. Therefore, mortgage modeling is a promising area for applying a variety of machine learning and artificial intelligence techniques.