RCA
Elite Member
- Joined
- Jun 27, 2017
- Professional Status
- Certified General Appraiser
- State
- California
So, some very recent events have me thinking:
1. TAF making REVAA a partner, no (public) discussion, 100% "Yea" vote by all members. REVAA being largely a legal and political representaton of AMCs, since its members are nearly all AMCs, despite goals stating something broader for the appraisal community.
2. Peter Christensen, 13.5 years with LIA, 10 years an attorney for real estate appraisal, ..., becomes General Counsel and Chief Compliance Officer for BBG, Inc.
3. Statements by Amorin and other leaders in the appraisal hinting at major changes in residential appraisal. They are tip toeing
4. Hints that the GSEs may be released from conservatorship and that this will cause their shares to skyrocket. They are apparently already pretty profitable, if their shares take off, after all of these years of conservatorship, they will be fat pigs. They will also become interesting targets for lawsuits.
.... and so on.
====== My conclusions:
1. The AMCs via REVAA will probably try to push more advanced AVMs based on AI, to replace residential appraisers. The hope is of course to drastically reduce costs by elimininating appraisers in at least 80% of loans in metro and suburban areas.
2. So, yes, they think they can use AI AVMs to replace residential appraisers. Yes they can, - in that 80%. They could use MARS regresssion, or they may think they can use easier to use Black Box approaches such as Random Forests (in particular this one), Neural Nets, XGBoost and so on to create price estimates, calculate residuals, rank properties and then use AI to rank the subject among the comparables based on photos ( and LLMs would certainly be good at this) by comparing the subject photos to the comparable MLS photos. MARS requires more knowledge and experience to use, but it transparent. Random Forests is easier to use, but is a Black Box approach. These Black Box approaches will need changes to USPAP to use effectively. Hard to say how effective they will be in getting the desired changes.
3. However, they largely don't know what they are dealing with . And the software engineers who know how to work with AI don't have a very good understanding of Real Estate or much of an interest in it.
4. -- So, we can expect major problems and lawsuits. That must scare the AMCs, TAF and the appraisal organizations. It looks like they are preparing for the worst.
5. Many residential appraisers will indeed be pushed out of the field and many 10 years before retirement, - with little to no retirement savings or substantial social security. If they want, they can stand a good chance, I think, of initiating and winning class action lawsuits against the GSEs and/or appraisal organizations for being misled and ill-treated. The GSEs have the money, but the appraisal organizations are more culpable, IMO. The AMCs would be all over the place.
6. The various Appraisal Organizations, and the GSEs, may become subject to very sizeable lawsuits for not doing the right thing.
======= Kaggle Ames Housing Data
Professior Dean De Cock created a good size data set of housing data from Ames, Iowa in 2011. It is on Kaggle.com an is often used in Kaggle Data Mining competitions.
You can join kaggle.com (free) and search for "Ames housing data" to find the data to download. It is 82 attributes on about 2600 homes. If you run earth on it you should get an R2 of about 86% (or 0.86). One model produced is in the link below.
Note that MARS (or as it is called in R - "earth") gives you a model that tells you how it is going to create the estimated sale process for any data that you enter.
www.kaggle.com
Random Forests or Neural Networks, on the other hand, will only give you a function that allows you to enter the property data, and then give you an estimate of the sale price, without any explanation whatsoever.
===== A Good Book on Kindle:
"Hands-On Machine Learning wtih R"
Bradley Boehmke & Brandon Greenwell
1. TAF making REVAA a partner, no (public) discussion, 100% "Yea" vote by all members. REVAA being largely a legal and political representaton of AMCs, since its members are nearly all AMCs, despite goals stating something broader for the appraisal community.
2. Peter Christensen, 13.5 years with LIA, 10 years an attorney for real estate appraisal, ..., becomes General Counsel and Chief Compliance Officer for BBG, Inc.
3. Statements by Amorin and other leaders in the appraisal hinting at major changes in residential appraisal. They are tip toeing
4. Hints that the GSEs may be released from conservatorship and that this will cause their shares to skyrocket. They are apparently already pretty profitable, if their shares take off, after all of these years of conservatorship, they will be fat pigs. They will also become interesting targets for lawsuits.
.... and so on.
====== My conclusions:
1. The AMCs via REVAA will probably try to push more advanced AVMs based on AI, to replace residential appraisers. The hope is of course to drastically reduce costs by elimininating appraisers in at least 80% of loans in metro and suburban areas.
2. So, yes, they think they can use AI AVMs to replace residential appraisers. Yes they can, - in that 80%. They could use MARS regresssion, or they may think they can use easier to use Black Box approaches such as Random Forests (in particular this one), Neural Nets, XGBoost and so on to create price estimates, calculate residuals, rank properties and then use AI to rank the subject among the comparables based on photos ( and LLMs would certainly be good at this) by comparing the subject photos to the comparable MLS photos. MARS requires more knowledge and experience to use, but it transparent. Random Forests is easier to use, but is a Black Box approach. These Black Box approaches will need changes to USPAP to use effectively. Hard to say how effective they will be in getting the desired changes.
3. However, they largely don't know what they are dealing with . And the software engineers who know how to work with AI don't have a very good understanding of Real Estate or much of an interest in it.
4. -- So, we can expect major problems and lawsuits. That must scare the AMCs, TAF and the appraisal organizations. It looks like they are preparing for the worst.
5. Many residential appraisers will indeed be pushed out of the field and many 10 years before retirement, - with little to no retirement savings or substantial social security. If they want, they can stand a good chance, I think, of initiating and winning class action lawsuits against the GSEs and/or appraisal organizations for being misled and ill-treated. The GSEs have the money, but the appraisal organizations are more culpable, IMO. The AMCs would be all over the place.
6. The various Appraisal Organizations, and the GSEs, may become subject to very sizeable lawsuits for not doing the right thing.
======= Kaggle Ames Housing Data
Professior Dean De Cock created a good size data set of housing data from Ames, Iowa in 2011. It is on Kaggle.com an is often used in Kaggle Data Mining competitions.
You can join kaggle.com (free) and search for "Ames housing data" to find the data to download. It is 82 attributes on about 2600 homes. If you run earth on it you should get an R2 of about 86% (or 0.86). One model produced is in the link below.
Note that MARS (or as it is called in R - "earth") gives you a model that tells you how it is going to create the estimated sale process for any data that you enter.
House Price Prediction Using MARS Algorithm
Explore and run machine learning code with Kaggle Notebooks | Using data from No attached data sources
Random Forests or Neural Networks, on the other hand, will only give you a function that allows you to enter the property data, and then give you an estimate of the sale price, without any explanation whatsoever.
===== A Good Book on Kindle:
"Hands-On Machine Learning wtih R"
Bradley Boehmke & Brandon Greenwell
Last edited:
