Quote:
Originally Posted by Normal
I also use algorithms. The matrix may drop differently, but data is only so accurate. One person puts in a pool, or pavers, or skylights, or a fireplace wall. Another changes to all tile flooring and does closets. Maybe the outside landscape is totally redone, none of which is recorded in county clerk records.
There has to be a standard deviation formula for human interaction verses years living in a place; please let me know if you ever come up with that.
|
So the flaw in my algorithm was that most of the comparables were updated / modernized with new kitchens, or flooring, appliances, etc. My parents house was structurally near perfect for 60 years old, but never updated for 60 years, with 20 year old appliances, 50 year old linoleum flooring, 50% of the windows with broken seals and cloudy, etc. Out of 14 comparables, only 2 were not upgraded, which means that the comparables were really not very comparable. I had a variable for modernized/updated but there weren't enough data points to get an accurate reduction in FMV for lack of upgrades.
So the algorithm overstated the FMV by probably 50-80 k to account for the cost of modernization, which I understood but would have to create some randomized data to account for the differences, which i didn't attempt. . . so the buyer's agent had the output of the model, and used it to buy the house slightly below the estimate, with someone who was dying to get into the town. . and they even waived the inspection after their real estate agent went through the house herself before the sale.
It only takes one buyer to get it done for reasons which is known only to them. .