Elliott
Elite Member
- Joined
- Apr 23, 2002
- Professional Status
- Certified General Appraiser
- State
- Oregon
Winter 2020 edition of The Appraisal Journal, written by Hans Isakson, PhD, Mark Ecker, PhD, and Lee Kennedy has a 12 page article on their analysis of AVMs.
Its not a pretty picture. While AVMs are good at mean errors (who isn't?), they have a lot of dispersion when put to the test. Of the metrics that are most meaningful, I find FSD (Forecast Standard Deviation) to be very useful. For GenPress it was 24.7%. Given the task of estimating market value they are within 15% of the value, only 60% of the time. I doubt an appraiser would survive their monthly AVM evaluation being that far off, though they'd get a 5 for beating due date. R^2 of .75, which is good enough for friends of regression. Its a good read. I hope someone at Fannie and FHFA reads it.
Its not a pretty picture. While AVMs are good at mean errors (who isn't?), they have a lot of dispersion when put to the test. Of the metrics that are most meaningful, I find FSD (Forecast Standard Deviation) to be very useful. For GenPress it was 24.7%. Given the task of estimating market value they are within 15% of the value, only 60% of the time. I doubt an appraiser would survive their monthly AVM evaluation being that far off, though they'd get a 5 for beating due date. R^2 of .75, which is good enough for friends of regression. Its a good read. I hope someone at Fannie and FHFA reads it.
