0

I have some mortgage data and I am trying to detect when a borrower refinances. To do so, I use a moving average filter to smooth my data and then I run a Bai and Perron (1998) structural break test on the borrowers monthly payments. I am wondering if it would make more sense to smooth the levels of the data or the monthly payments. I have tried both but I get pretty different results. What makes more sense to best identify mean shifts in the monthly payment?

I have read a little bit about this, and I never see a discussion about what makes more sense to smooth. I imagine it is on a case-by-case basis.

fairlife4life
  • 1
  • 1
  • Welcome to DSP.SE! How noisy is the data (typical variation of monthly payments for a consistent interest rate)? Can that noise be explained well by a model? I would expect taxes to change at regular intervals, if that amount is included in your data. Refi's might be more difficult to distinguish from PMI removal. Other than that, I would (perhaps naively) expect your data to follow the amortization formula. So if you fit a sliding window of that data to a model, you would essentially be detecting a change in model parameters. – Ash Dec 18 '23 at 17:55
  • Thanks @Ash! I would be slightly worried about trying to model the noise since there are a lot of borrowers and variation in structure is quite large. Payments generally follow the monthly payment formula. On average, the variation in monthly payments is not too large. Units are in UF, so on average they vary by around 2 - 3 UF. – fairlife4life Dec 18 '23 at 19:56

0 Answers0