Avm Risk? Avm Accuracy Above That Of Appraisals

[George Hatch]

The lack of sophistication is very obvious when you look at zillow in highrise condo buildings where the units in the same tier are identical. You have 10-20 identical units stacked together. The model value for these units in the same tier should all be very close. But I looked through a building recently on zillow looking at all of the units in a tier, any unit that traded recently had a estimated value close to the sale price but any unit that did not trade recently were estimated to be 10-15% higher than the ones that traded recently. It also does not seem to put any value on the parking space which is about a 10% adjustment. If it had any kind of sophistication then the estimated value of a unit in a tier would be based on the units in the tier which are identical.

We already build small models that are tuned to certain situation-driven specifics. If we were doing enough assignments in those market segments to justify maintaining those models we could reuse them over and over and save ourselves a lot of "reinvent the wheel" time/effort.

Building a single model that is capable of handling the myriad of combinations out there is a much more daunting task.

 

[Denita Neuenhaus]

I've never been an agent but have met many over the years....
It's my opinion good/successful agents are proactive and "teach" sellers and buyers....
You sound like a great agent!!!

Thanks. I consider it important to stay on top of the lending criteria so we can spot a strong buyer vs a weak buyer in accordance with today's guidelines. Not taking the place of a great LO but learning what works in the lending world stops a lot of 'contracts' that have no possibility of actually closing.

@Dublin ohio: As to that agent that "claimed geographical competence because my office was not located in same zip code. But I was less than 5 miles away." I apologize on behalf of those of us that are in it for a career. That is the most unprofessional reason I have heard in a long time. If the appraiser is coming 200+ miles one way in a huge metro area I can see this as potentially a valid reason, but 5 miles? Not even a blip on the radar. Something was wrong there.

 

[Thomas Holding]

By the method I use, comparables are most often just window dressing. ... they will simply be regarded as demo cases for the model the appraiser has come up with , rather than as the source of adjustments. An investigation or court case may open up the set of sales transactions used for creating the model, - but nothing short of that.

The system that you developed is very similar to several mass appraisal systems that I have seen in use across the country. The idea is sound and it works very well depending on your models. Typically, when a taxpayer goes to the county tax website, they will see a representation of several comparable properties (determined by a similar ranking methodology btw.), that gives the impression that a particular property was directly compared to the selected comps. In actuality the assessed value is derived from a hybrid cost/sales/regression model. The comparables are provided in this way not to mislead but, a) mainly because the public is familiar with this format, and b) to illustrate that their valuation was equitable.

The main drawback for individual appraisers is that developing a working system like this is very labor intensive and expensive. It's the main reason why counties in my state typically perform revaluations only every 4-8 years. If your system works as well as you say it does with easily obtainable data, then that is quite impressive and I hope that you continue to refine it so that ultimately it will be a tool that other appraisers can use in their businesses.

Concerning the whole AVM accuracy / risk equation, I know that most fee appraisers don't give much credence to residential tax values, but I can tell you from personal experience that (at least in my state) mass appraisal systems are getting much more sophisticated and a considerable amount of effort and resources are being allocated to increase the accuracy of tax data and of valuations. The danger that I see for residential appraisers in the future is that most of the data that tax departments collect is in the public domain and thus available to anyone who asks for it. When combined with other data sources (like MLS and advanced GIS), this enormous dataset will ultimately enable AVMs to increase their overall accuracy to a level that balances any risks associated with their use. It's why I personally have been advocating that appraisers work together and combine their own (typically more accurate data) into a database that they can control and monetize.

AVMs don't have to be perfect before they become more widely accepted, they just have to be good enough. When that happens, our profession will face even more severe disruptions.

 

[RCA]

The system that you developed is very similar to several mass appraisal systems that I have seen in use across the country. The idea is sound and it works very well depending on your models. Typically, when a taxpayer goes to the county tax website, they will see a representation of several comparable properties (determined by a similar ranking methodology btw.), that gives the impression that a particular property was directly compared to the selected comps. In actuality the assessed value is derived from a hybrid cost/sales/regression model. The comparables are provided in this way not to mislead but, a) mainly because the public is familiar with this format, and b) to illustrate that their valuation was equitable.

The main drawback for individual appraisers is that developing a working system like this is very labor intensive and expensive. It's the main reason why counties in my state typically perform revaluations only every 4-8 years. If your system works as well as you say it does with easily obtainable data, then that is quite impressive and I hope that you continue to refine it so that ultimately it will be a tool that other appraisers can use in their businesses.

Concerning the whole AVM accuracy / risk equation, I know that most fee appraisers don't give much credence to residential tax values, but I can tell you from personal experience that (at least in my state) mass appraisal systems are getting much more sophisticated and a considerable amount of effort and resources are being allocated to increase the accuracy of tax data and of valuations. The danger that I see for residential appraisers in the future is that most of the data that tax departments collect is in the public domain and thus available to anyone who asks for it. When combined with other data sources (like MLS and advanced GIS), this enormous dataset will ultimately enable AVMs to increase their overall accuracy to a level that balances any risks associated with their use. It's why I personally have been advocating that appraisers work together and combine their own (typically more accurate data) into a database that they can control and monetize.

AVMs don't have to be perfect before they become more widely accepted, they just have to be good enough. When that happens, our profession will face even more severe disruptions.

 

[RCA]

Yes, it is labor intensive. It takes me sometimes forever to get appraisal out, because - I have to do so many things to get things done the way I want, plus do all the things my good mentors taught me I "should" do - like extract tax assessor and property details, which is sometimes easy and sometimes a lot more time consuming than it should be.

Creating models takes more than loading data and pushing a button. -- Well if you want to do simple linear regression on a number of properties and just live with the result, it can be a fast process. But if you are striving to get an R2 (adjusted R2) of close as far over 0.70 as possible with something like MARS, that can take a lot of searching and probing against different subsets of variables. Salford-Systems MARS can quickly create dozens of possible models, which are ranked and you have to study them to see if they make sense. You have to extract out "confounding" and "collinear" variables. Do you want to use "AreaID", Census Tract/Block, Longitude/Latitude or some other variable for location. How can you separate out houses with ocean views - lots and lots of problems.

Then you wrestle with quirks and bugs in the software, and find that you could use some plug-in that costs another $100/month of some such thing. I'm currently wrestling with InDesign. It gives you a lot of control and I really like that way you can split up your report into independent chapters that are isolated from each other - which makes a long doc more stable. While the long doc is broken up into semi-independent pieces, you can still do global searches and changes across all "chapters". But it seems infinitely complex at times, with lots of hidden and silent features to hide the complexity. For example, if you want to "unanchor" a small diagram you imported into your doc, which for some reason got imported as an anchored object, so you can't center it, you have to know to Ctrl-X cut it and the paste it back in WITHOUT AN ACTIVE CURSOR!! Well, I am learning it and hopefully will master it. Germans use it a lot. It is for publishers and typesetters who strive for excellent publications. So, I like it - and I do have other uses for this skill .....

My work in appraisal is more like interesting research, not just doing appraisals. Maybe I learn something that can be useful to others, but certainly what I am doing is probing into unknown areas ..... Why? Because I see Valuation as a more general skill beyond Real Estate, as something that is the backbone of decision making.

 

[Mike Garrett RAA]

Does anyone else find it ironic that Core Logic now owns Alamode...the largest provider of appraisal software in the country? They also own a great number of multiple listing systems in the country.

 

[Carolina1]

Yes, it is labor intensive. It takes me sometimes forever to get appraisal out, because - I have to do so many things to get things done the way I want, plus do all the things my good mentors taught me I "should" do - like extract tax assessor and property details, which is sometimes easy and sometimes a lot more time consuming than it should be.

Creating models takes more than loading data and pushing a button. -- Well if you want to do simple linear regression on a number of properties and just live with the result, it can be a fast process. But if you are striving to get an R2 (adjusted R2) of close as far over 0.70 as possible with something like MARS, that can take a lot of searching and probing against different subsets of variables. Salford-Systems MARS can quickly create dozens of possible models, which are ranked and you have to study them to see if they make sense. You have to extract out "confounding" and "collinear" variables. Do you want to use "AreaID", Census Tract/Block, Longitude/Latitude or some other variable for location. How can you separate out houses with ocean views - lots and lots of problems.

Then you wrestle with quirks and bugs in the software, and find that you could use some plug-in that costs another $100/month of some such thing. I'm currently wrestling with InDesign. It gives you a lot of control and I really like that way you can split up your report into independent chapters that are isolated from each other - which makes a long doc more stable. While the long doc is broken up into semi-independent pieces, you can still do global searches and changes across all "chapters". But it seems infinitely complex at times, with lots of hidden and silent features to hide the complexity. For example, if you want to "unanchor" a small diagram you imported into your doc, which for some reason got imported as an anchored object, so you can't center it, you have to know to Ctrl-X cut it and the paste it back in WITHOUT AN ACTIVE CURSOR!! Well, I am learning it and hopefully will master it. Germans use it a lot. It is for publishers and typesetters who strive for excellent publications. So, I like it - and I do have other uses for this skill .....

My work in appraisal is more like interesting research, not just doing appraisals. Maybe I learn something that can be useful to others, but certainly what I am doing is probing into unknown areas ..... Why? Because I see Valuation as a more general skill beyond Real Estate, as something that is the backbone of decision making.

Excellent posts Bert. Adaptation and evolution are necessary.

 

[Joe Flacco]

Zillow acknowledges that AVM's do not work in high density urban areas such as the northeast by avoiding high density areas with ibuyer. They are not willing to put their money where their mouth is with the ibuyer busines because it is "unpredictable" and "singular". I don't know why banks or GSE's would rely on AVM's. I am sure it is no secret in the industry that it does not work well in high density urban areas regardless of the "accuracy" they tout.

 

[Carolina1]

View attachment 37989


Zillow acknowledges that AVM's do not work in high density urban areas such as the northeast by avoiding high density areas with ibuyer. They are not willing to put their money where their mouth is with the ibuyer busines because it is "unpredictable" and "singular". I don't know why banks or GSE's would rely on AVM's. I am sure it is no secret in the industry that it does not work well in high density urban areas regardless of the "accuracy" they tout.

Joe,

If you were setting up a large scale test of value accuracy for several valuation options, what would you use as the "true" value from which you would judge the accuracy of the options?

 

[Joe Flacco]

Joe,

If you were setting up a large scale test of value accuracy for several valuation options, what would you use as the "true" value from which you would judge the accuracy of the options?

I think sale price is the only judge of accuracy or true value. Certainly not a automated model.