- Joined
- Jun 27, 2017
- Professional Status
- Certified General Appraiser
- State
- California
I should be more mindful of your knowledge in appraisal theory and not dismiss you as an AVM wanna be. I have often thought we could come to an agreement on at least one matter, but then we would both be wrong. I praise you on your many years of experience. Intelligent thoughts have always followed you, but you are faster.
I don't think you know what an AVM is. I am not an AVM. I am just like other appraisers, only:
1. I review the MLS and other data, making corrections that I can efficiently make. I may even spend a day gathering new data that doesn't exist anywhere else.
2. I use MARS Regression instead of Linear or Multi-Linear Regression. That means I will almost always get higher R2 values, such as 70% instead of 40%.
3. I use R to do a lot of the statistical work with Excel spreadsheets - but have to go into modify the R for each appraisal.
4. I use correct mathematical constraints, so all of my adjusted sale price values are as they should be: Exactly identical.
5. I measure accurately, including wall thickness, and use Chief Architect software for floor-plans.
6. If there has been recent construction I check for permits and inspections. I will also check any existing history on building permits.
7. I check the usual public records for title and so on
8. I do make clerical mistakes if pushed to get some time-consuming report by some required date and regret it later.
AVMs process thousands of properties at a time and often use black box statistical methods such as Random Forests, which do not provide models based on the different contributing factors to value. They rely largely on averages for variables that don't have measurement data such as condition, quality and so on. So, AVM results are largely for properties that are in average condition and built with average quality for the neighborhood - unless there is some statistical indication otherwise. They are getting better - but are still pulled to the center of value because of averaging what they do not know. Nowadays, AVMs may be given inspection data for a subject property which improves their estimates. But most likely they will have very little of that inspection data for the hundreds of comparables that go into their calculations.