Modeling Prepay Speeds

by AMG

As the regulatory scrutiny of interest rate risk has increased, the criticism of ALM model inputs and assumptions has soared as well.  Don't expect this to change any time soon.

"When was this last updated?" 

This will invariably be one of the first questions asked about any model assumption.  Unfortunately, many of you do not have good answers for this question, especially for some of the more "hard to find" assumptions.  In fact, the answer is very often an honest "I have no idea."  Did you use one of these to put it into the model?


Using stale assumptions will not only get you in hot water during exams, it will also create model errors that may lead to poor decision making.  This is especially true of the critical assumptions, such as non-maturing deposit durations, pricing betas, and prepay speeds.

Prepay speeds is a weakness we see in lot of models that we run across.  Modeling prepay speeds, especially for loan portfolios, is notoriously difficult.  Heck, it's hard to even figure out what your actual prepay speeds are, much less how to model for the future.  However, with all of the long duration assets that have been accumulating on bank balance sheets, getting the direction and basic magnitude of changes in prepay speeds will be vital to model accuracy.  And if you are using static and/or old assumptions, how will you capture the changes in an evolving marketplace? 

We are already seeing changes in prepay speeds due to higher rates:


Total refinancing volume, including refinances through the Home Affordable Refinance Program, dipped in the third quarter of 2013 amid rising interest rates, the Federal Housing Finance Agency’s third-quarter 2013 Refinance Report revealed.
The total refinance volume hit just below 900,000 in the third quarter, with an estimated 200,000 homeowners refinancing through HARP in that period.

Have your models changed along with the market?  In our model, pools of loans are matched with an MBS proxy that has similar maturity and coupon characteristics.  We then combine the MBS forecasted prepay speeds with actual loan speeds to come up with projected speeds that are based in reality for your loans, but have dynamic expectations of direction and magnitude from the MBS market.  This keeps the assumption from getting stale, and generates very reasonable cash flow projections:

There are many other methodologies that work, but I would suggest including a market based component to your forecasted prepay speeds.  I have seen some models that try to use only bank specific data, but this is incredibly difficult, especially in community banks.  How old is the data from a rising rate environment?  And do you have a high enough loan count to generate statistically significant data?  Even if you can't automate it, at least use fresh (definitely less than a year old, preferably updated quarterly) MBS forecasts for a sanity check on your prepay models.  If you have not been asked about it before, you will soon!