The Impact of Statistics on Sales Prices.

[RCA]

I'm in the process of running regressions on every area (using the MLS defined areas) of every city in the major counties around the SF Bay Area, as covered by my MLS, pro.mlslistings.com. It's a lot of work - that I have been putting off. I should finish San Mateo County tomorrow, then Santa Clara is next, then Monterey, then Alameda and if I get around to it Santa Cruz. You might be surprised by how many cities and neighborhoods there are. Powershell scripts make it possible to set up all the workfiles without too much effort, in any case, once the scripts are written and debugged.

To my surprise, I am getting really high R2 values in the range of 0.79 to 0.90+ just on the first stage of my regression (not taking into consideration subjective variables.) for all but a handful of towns. -- And this is on 16 years of data, typically several thousand sales per city; of course more for the large cities. This would not have been possible in say 2005.

I believe this is the impact of real estate agents, appraisers and buyers relying on AVM services which are based on statistical analysis - and perhaps also appraisers using regression. But the change is very evident to me. It applies to all neighborhoods from the low end to the high end.

I may put some of my Stage I regression models, which are rather straightforward runs of Minitab/Salford Systems MARS on GitHub. Haven't decided yet.

 

[Randolph Kinney]

I believe this is the impact of real estate agents, appraisers and buyers relying on AVM services which are based on statistical analysis

So ... are you going to be right up there with Zillow? Maybe launch a competing AVM service?

 

[Ariba]

So ... are you going to be right up there with Zillow? Maybe launch a competing AVM service?

Amen Why even bother doing the observation.

I'm in the process of running regressions on every area (using the MLS defined areas) of every city in the major counties around the SF Bay Area,

How does that relate to individual neighborhoods in a county/city? Just more BS

 

[RCA]

So ... are you going to be right up there with Zillow? Maybe launch a competing AVM service?

Zillow does not provide statistical models - only values.

In fact, many appraisers provide "fly by the seat of your pants" for subjective adjustments such as view - they don't know how to do otherwise.

I provide models, that can aid in generating values. However, "Stage I" models do not get into subjective valuation; they only provide objective processing by regression software on input data, that will account for hopefully 80% of price differences. "Stage II" gets into dealing with the really difficult touchy subjective valuation, - typically the other 10-20% of price variation.

So, you are comparing completely disparate things.

 

[Renee Healion]

snipped
I believe this is the impact of real estate agents, appraisers and buyers relying on AVM services which are based on statistical analysis - and perhaps also appraisers using regression. But the change is very evident to me. It applies to all neighborhoods from the low end to the high end.

I wondered something similar in a small town where one agency was quite active on the list side and sales were at full list price. The sale prices were 161% of the assessment, several of them. I think they were listing based on the relationship to assessment (or assessor's full value). This is a rural town not a place with very consistent development.

No Prop XX effect on taxes here, they are ad valorem, done town by town.

 

[Randolph Kinney]

Zillow does not provide statistical models - only values.

I provide models, that can aid in generating values.

Although your value estimate uses different methods than Zillow, the public will compare your value estimate with Zillow value estimates. Will the public distinguish your methods from Zillow when it comes to value? If you assert that your value conclusion is an appraisal, then you make that distinction whereas Zillow asserts its value is not an appraisal.

According to Zillow:

Zestimate methods
Zillow publishes Zestimate home valuations for 97.5 million homes across the country, and uses millions of statistical and machine learning models that can examine hundreds of data points for each individual home.

To calculate a Zestimate, Zillow uses a sophisticated and proprietary algorithm that incorporates data from county and tax assessor records and direct feeds from hundreds of multiple listing services and brokerages. The Zestimate also incorporates a home's facts and features, which homeowners have the ability to update.

The Zestimate accounts for variables like:

• Home characteristics including square footage, location or the number of bathrooms
• Unique features like hardwood floors, granite countertops or a landscaped backyard
• On-market data such as listing price, description, comparable homes in the area and days on the market
• Off-market data — tax assessments, prior sales and other publicly available records

Currently, we have data for over 110 million U.S. homes and we calculate Zestimates for more than 97.5 million of them.

 

[RCA]

Although your value estimate uses different methods than Zillow, the public will compare your value estimate with Zillow value estimates. Will the public distinguish your methods from Zillow when it comes to value? If you assert that your value conclusion is an appraisal, then you make that distinction whereas Zillow asserts its value is not an appraisal.

According to Zillow:

Zestimate methods
Zillow publishes Zestimate home valuations for 97.5 million homes across the country, and uses millions of statistical and machine learning models that can examine hundreds of data points for each individual home.

To calculate a Zestimate, Zillow uses a sophisticated and proprietary algorithm that incorporates data from county and tax assessor records and direct feeds from hundreds of multiple listing services and brokerages. The Zestimate also incorporates a home's facts and features, which homeowners have the ability to update.

The Zestimate accounts for variables like:

• Home characteristics including square footage, location or the number of bathrooms
• Unique features like hardwood floors, granite countertops or a landscaped backyard
• On-market data such as listing price, description, comparable homes in the area and days on the market
• Off-market data — tax assessments, prior sales and other publicly available records

Currently, we have data for over 110 million U.S. homes and we calculate Zestimates for more than 97.5 million of them.

It's interesting, after creating a Stage I model for Pacifica, I plugged in my house values into the model and it popped out $1,528,000, - not far from Zillows $1,529,583. Stage I assumes that home has average characteristics compared to recent sales. Now, is Zillow using MiniTab/Salford MARS? They say they "use" a sophisticated and proprietary algorithm - but that could be MARS, or perhaps another product of theirs called TreeNet that would give similar values but is a black-box approach. Maybe Zillow has stabilized its valuation generation methods - they can always change.

However, appraisers need to go beyond Stage I. Not all houses are average quality, condition, have average views for the neighborhood, and so on. Stage II/III valuation is, I am very sure, missing from Zillow. The day will come when they will perhaps attempt to analyze MLS and streetside photos to develop estimates of subjective features, yet that is where fraud could surface with a little imagination.

 

[Randolph Kinney]

It's interesting, after creating a Stage I model for Pacifica, I plugged in my house values into the model and it popped out $1,528,000, - not far from Zillows $1,529,583.

So ... using your method versus Zillow's method gives you $1,583 difference? A difference of 0.10%?

That really says your method compared to Zillow does not have a discernible difference on value.

Tough sell Bert, saying your method is better than Zillow.

 

[Ariba]

In fact, many appraisers provide "fly by the seat of your pants" for subjective adjustments such as view - they don't know how to do otherwise.

Since the view in the MLS by agents is very subjective, to say the least, I assume your model incorporates the Topographical map of every neighborhood to determine view? Come to think of it, how much of the data reported in the MLS is subjective?

Your talent is wasted being an appraiser you should be in the financial sector and make a fortune.

 

[Ariba]

Double Post