- Find a Real Estate Professional
- Realtors®
- Mortgage Lenders
- Home Improvement Pros
- Other Real Estate Services
- Review an Agent, Lender or Pro
- Marketing on Zillow
- Real Estate Agent Advertising
- Join the Professional Directory
- Popular
- Real Estate Market Reports
- More
Best Answer
- Dunes....
- Contributions:3894
I admire your patience pj20..
I hope Zillow responds to your question but the truth is they seem to be enthusiastically responding to anything but the Zestimate Questions
I hope Zillow responds to your question but the truth is they seem to be enthusiastically responding to anything but the Zestimate Questions
- Rachel Rosen, "RachelRosen"
- Contributions:1493
@PJ20:
First item to note is that the zestimate is always changing. Frankly, we could have the opposite issue next month, where your neighbor contacts us saying his went down, and yours went up. So, these frequent changes are not meant to be the "be all end all" when it comes to market value. That brings me to the next point. Each zestimate is calculated individually, so the zestimate is not comparing you to your neighbor. This home could easily be a mile away from you as well (as far as the zestimate is concerned). One key item I see is that we do not have past sale history for you, which does affect your zestimate. The amount of data we have affects the Zestimate accuracy.
Regarding Zillow answering Zestimate Questions:
The forums are a place for the community to ask each other questions. If forum members know the answer to a question and answer it, there is no need for a zillow employee to give the same answer. We are not staffed to answer every question that is asked in the forums. It is not needed for us to be, since many members already know the answer and reply before one of us sees it. Often, I see the answer in the thread so I do not see a need to respond just to show my zillow badge and say- Correct!. We do our best to chime in when needed. We are adding employees to better handle the growing user base, but we will never answer every question in advice since it is not always needed to specifically have a zillow employee answer.
We do have FAQ's on the site, and the zestimate value and accuracy link at the bottom of every page.
Here it is: http://www.zillow.com/how ... uracy.htm
First item to note is that the zestimate is always changing. Frankly, we could have the opposite issue next month, where your neighbor contacts us saying his went down, and yours went up. So, these frequent changes are not meant to be the "be all end all" when it comes to market value. That brings me to the next point. Each zestimate is calculated individually, so the zestimate is not comparing you to your neighbor. This home could easily be a mile away from you as well (as far as the zestimate is concerned). One key item I see is that we do not have past sale history for you, which does affect your zestimate. The amount of data we have affects the Zestimate accuracy.
Regarding Zillow answering Zestimate Questions:
The forums are a place for the community to ask each other questions. If forum members know the answer to a question and answer it, there is no need for a zillow employee to give the same answer. We are not staffed to answer every question that is asked in the forums. It is not needed for us to be, since many members already know the answer and reply before one of us sees it. Often, I see the answer in the thread so I do not see a need to respond just to show my zillow badge and say- Correct!. We do our best to chime in when needed. We are adding employees to better handle the growing user base, but we will never answer every question in advice since it is not always needed to specifically have a zillow employee answer.
We do have FAQ's on the site, and the zestimate value and accuracy link at the bottom of every page.
Here it is: http://www.zillow.com/how ... uracy.htm
- Rachel Rosen, "RachelRosen"
- Contributions:1493
PJ20,
Please email your address so I can investigate and reply. I'll reply to the of the thread tomorrow.
My email is rachelr at zillow dot com
Please email your address so I can investigate and reply. I'll reply to the of the thread tomorrow.
My email is rachelr at zillow dot com
- Pasadenan
- Contributions:21466
"Please, take a deep breath and find your better self" -
I was tempted to take a deep breath and find the specific house in question, as there is plenty of data posted to find it.
But it would be a complete waste of time, as no one on this thread wants any answers, they only want to state their opinion is that the calculations for the previous month for that neighborhood were "WRONG" by 1.3% because there was one fewer sale on record for the modeling.
Estimates moving in opposite directions month to month on the same street for the same house size, same lot size, same #bedrooms, same #bathrooms, same number of stories, same year built, and same #covered parking spaces is a "feature" of the estimating model, not a "flaw".
As for Zillow responding? Why would they when it is clear that there never was any question?
And if someone really believed it was a "bug", they would have posted in the "bugs" category to begin with.
I was tempted to take a deep breath and find the specific house in question, as there is plenty of data posted to find it.
But it would be a complete waste of time, as no one on this thread wants any answers, they only want to state their opinion is that the calculations for the previous month for that neighborhood were "WRONG" by 1.3% because there was one fewer sale on record for the modeling.
Estimates moving in opposite directions month to month on the same street for the same house size, same lot size, same #bedrooms, same #bathrooms, same number of stories, same year built, and same #covered parking spaces is a "feature" of the estimating model, not a "flaw".
As for Zillow responding? Why would they when it is clear that there never was any question?
And if someone really believed it was a "bug", they would have posted in the "bugs" category to begin with.
- hpvanc
- Contributions:2579
Not Zillow either, but the current low transactions volumes are a major factor in the inaccuracy and wild swings.
You stated that your property is one of the more valuable on the street, which could be an indication for the algorithm to use different factors causing it to skip some of the comps or modeling that is driving the other values in your neighborhood and pick different ones that have either more distressed properties in the equation or forces it to search in a larger geographical area that encompasses entirely different neighborhoods to use those factors that differ from your neighbors.
Some things that I have noticed that seem to drive these anomalies are:
Lot sizes that are not consistent with the neighborhood.
Number of bathrooms (even if it is a higher number).
You stated that your property is one of the more valuable on the street, which could be an indication for the algorithm to use different factors causing it to skip some of the comps or modeling that is driving the other values in your neighborhood and pick different ones that have either more distressed properties in the equation or forces it to search in a larger geographical area that encompasses entirely different neighborhoods to use those factors that differ from your neighbors.
Some things that I have noticed that seem to drive these anomalies are:
Lot sizes that are not consistent with the neighborhood.
Number of bathrooms (even if it is a higher number).
- Dunes....
- Contributions:3894
Unexpected but appreciated pj20..Thank-you
Pretty much agree with your entire last Post...
Unfortunately IMO this refusal to respond to Zestimate Questions while having numerous Staff respond to Fluff Questions has entirely changed how I view this Companies Direction or Intentions and Value as it relates to the Consumer.....
Pretty much agree with your entire last Post...
Unfortunately IMO this refusal to respond to Zestimate Questions while having numerous Staff respond to Fluff Questions has entirely changed how I view this Companies Direction or Intentions and Value as it relates to the Consumer.....
- pj20
- Contributions:18
Pas dean, A moment ago you were belittlig my credentials, suddenly you are celebrating them grandiosely. Of course, neither are correct and the obvious question is why are you taking the time to stoop to personal attacks? By the way, you stated that I 'know what needs to be fixed and how to fix it'. That is false I stated no such thing. I said there is a glaring error and that zillow shoul find out what is causing it an fix it and any related problems. When we began this conversation, you were not making personal attacks or accusing me of saying things I hadn't. Please, take a deep breath and find your better self.
- Pasadenan
- Contributions:21466
pj20,
It is obvious that you have more knowledge of the U.S. housing market, and more experience in modeling and statics than
Stan Humphries
Svenja Maarit Gudell
Dong Xiang
Yeng Bun
Steve Brownell
Thomas G. Thibodeau
and
Stephen Malpezzi
And more experience and understanding than
First America CoreLogic
Fiserv, Inc. ("Fiserv")
IntelliReal, LLC ("IntelliReal")
Interthinx, Inc. ("Interthinx")
Lender Processing Services, Inc. ("LPS")
Precision Appraisal Services, Inc. ("Precision Appraisal")
Real Data, Inc. ("RDI")
RealEC Technologies, Inc. ("RealEC")
Zillow, Inc. ("Zillow")
American Flood Research, Inc. ("AFR")
Electronic Appraiser, Inc. ("Electronic Appraiser")
Espiel, Inc. ("Espiel")
Homesand.net
Eppraisal
cyberhomes.com
trulia
chase
homegain
So, I'm baffled why you are not building your own home estimating website...
You state you know what needs to be fixed and how to fix it, so why aren't you fixing it?
Surely with all your graduate studies you can have it on line within the next 6 months?
It is obvious that you have more knowledge of the U.S. housing market, and more experience in modeling and statics than
Stan Humphries
Svenja Maarit Gudell
Dong Xiang
Yeng Bun
Steve Brownell
Thomas G. Thibodeau
and
Stephen Malpezzi
And more experience and understanding than
First America CoreLogic
Fiserv, Inc. ("Fiserv")
IntelliReal, LLC ("IntelliReal")
Interthinx, Inc. ("Interthinx")
Lender Processing Services, Inc. ("LPS")
Precision Appraisal Services, Inc. ("Precision Appraisal")
Real Data, Inc. ("RDI")
RealEC Technologies, Inc. ("RealEC")
Zillow, Inc. ("Zillow")
American Flood Research, Inc. ("AFR")
Electronic Appraiser, Inc. ("Electronic Appraiser")
Espiel, Inc. ("Espiel")
Homesand.net
Eppraisal
cyberhomes.com
trulia
chase
homegain
So, I'm baffled why you are not building your own home estimating website...
You state you know what needs to be fixed and how to fix it, so why aren't you fixing it?
Surely with all your graduate studies you can have it on line within the next 6 months?
- pj20
- Contributions:18
Dunes,
You can encounter a lot on these boards - intelligent conversation; answers to pressing questions; infantile behavior; all of the learned tricks of a debater not wanting to lose - obfuscation, red-herrings, character assassination, attacking straw men,etc.
I'm just glad when I get the rare intelligent comment or conversation.
As far as Zillow, it seems to me that they have great potential, but they will never get there with such a seriously flawed model. I have only been focusing on one glaring error. Others have cited many serious ones. Pasadenan even intentionally or not acknowledged the error of anchoring to a very dated sales price. If Zillow actually invested the time and built a better zestimate, 2.31 perhaps, then they have a chance at being a great company.
You can encounter a lot on these boards - intelligent conversation; answers to pressing questions; infantile behavior; all of the learned tricks of a debater not wanting to lose - obfuscation, red-herrings, character assassination, attacking straw men,etc.
I'm just glad when I get the rare intelligent comment or conversation.
As far as Zillow, it seems to me that they have great potential, but they will never get there with such a seriously flawed model. I have only been focusing on one glaring error. Others have cited many serious ones. Pasadenan even intentionally or not acknowledged the error of anchoring to a very dated sales price. If Zillow actually invested the time and built a better zestimate, 2.31 perhaps, then they have a chance at being a great company.
- pj20
- Contributions:18
Pasadean,
My graduate academic background has more than adequate statistical
courses, as well as modeling courses.
It is unfortunate that rather having an intelligent conversation, you have chosen to resort to slightly veiled personal attacks. I won't go there.
The zestimate has an obvious flaw and I do hope the folks at zillow go about correcting the flaw.
My graduate academic background has more than adequate statistical
courses, as well as modeling courses.
It is unfortunate that rather having an intelligent conversation, you have chosen to resort to slightly veiled personal attacks. I won't go there.
The zestimate has an obvious flaw and I do hope the folks at zillow go about correcting the flaw.
- yahui168
- Contributions:18
It seems a logic flaw that requires "neighboring properties which are alike in all key variables" as fact and given this fact then the model must be wrong. It's possible that the model do not treat this as identical but the info provided does not allow other people to show what is not identical between the properties.
Perhaps one house is built in 1978 and one in 1981 and houses built in the 70's are selling better because 70's retro is trendy and 80's elegance not so much.
Perhaps one house is built in 1978 and one in 1981 and houses built in the 70's are selling better because 70's retro is trendy and 80's elegance not so much.
- Pasadenan
- Contributions:21466
If you want validity information, that is posted on
Zestimate Tolerance range by region, Excel File
Your misunderstanding would likely be resolved if you took a statistics course or an introductory physics course at a local college.
I also suggest reading some of the Zillow Research Briefs, such as:
upgrading the Zestimate
or
Case-Shiller not just a home price index
Zestimate Tolerance range by region, Excel File
Your misunderstanding would likely be resolved if you took a statistics course or an introductory physics course at a local college.
I also suggest reading some of the Zillow Research Briefs, such as:
upgrading the Zestimate
or
Case-Shiller not just a home price index
- pj20
- Contributions:18
Pasadenan,
You state - 'There is no "flaw" to correct'. With that statement you have invalidated yourself as one who can judge any type of model.
As, the flaw is obvious, it gives divergent answers for neighboring properties which are alike in all key variables. That sir is a glaring error and it is potentially worse, because you don't know how the error has shown up, what mistaken calculation, linkage, assumption caused it are unknown. So, those mistaken calculations, links, assumptions could be causing other errors.
You say that the model will NEVER do what I am asking. Well, I am only asking for validity.
You state - 'There is no "flaw" to correct'. With that statement you have invalidated yourself as one who can judge any type of model.
As, the flaw is obvious, it gives divergent answers for neighboring properties which are alike in all key variables. That sir is a glaring error and it is potentially worse, because you don't know how the error has shown up, what mistaken calculation, linkage, assumption caused it are unknown. So, those mistaken calculations, links, assumptions could be causing other errors.
You say that the model will NEVER do what I am asking. Well, I am only asking for validity.
- Pasadenan
- Contributions:21466
There is no "flaw" to correct. If you gave the street name, I would have already shown that the tax assessed value and the year purchased are substantially different and the cause of the movement in opposite directions.
The model is being modified everyday. It will be better in a few months. But it will NEVER do what you are asking, and never can do what you are asking.
If you want an estimate that ignores tax assessed value and last purchase price/date, use homesand.net, or eppraisal.com, or trulia.com.
It appears that you went to the wrong website.
The model is being modified everyday. It will be better in a few months. But it will NEVER do what you are asking, and never can do what you are asking.
If you want an estimate that ignores tax assessed value and last purchase price/date, use homesand.net, or eppraisal.com, or trulia.com.
It appears that you went to the wrong website.
- pj20
- Contributions:18
I have been trying to be polite, but it is getting harder as you consistently ignore my question and expound at length on tangential matters.
If any serious model gives two divergent conclusions for two subjects alike in all key characteristics , then there is a serious flaw in the model. How serious depends on the nature of the flaw itself. I would hope that Zillow would determine the cause of the flaw and correct it.
If any serious model gives two divergent conclusions for two subjects alike in all key characteristics , then there is a serious flaw in the model. How serious depends on the nature of the flaw itself. I would hope that Zillow would determine the cause of the flaw and correct it.
- Pasadenan
- Contributions:21466
I thought you were interested in an answer and explanation.
Perhaps I was "wrong" and you are only interested saying why you want an different model, or why you don't like the numbers. In which case, I may have wasted my time since this thread only has a views to post ratio of 7.4, meaning very few people are reading it. But if I can't even explain a simple model to others, how are better models going to be developed? Without dialog on the issues, algorithms don't improve.
I known nothing about cardiology, and wouldn't even attempt a computer solution for that; even rough diagnosis.
We are not talking about Cardiology; and the "characteristics" of a house are NOT defined by:
1) house sqft
2) lot sqft
3) # bedrooms
4) # bathrooms
5) # stories
6) year built
7) Neighborhood
For example take a cheep lot next to lots of traffic; but flat, no trees; hard compacted soil.
pour a slab on grade; build a square house, 8 foot ceilings (minimal drywall cutting), scrap pine framing, flat roof with 1:12 slope directly on the ceiling joists (no attic); no roof overhang, minimal insulation, 10 year asphalt rolled roofing; minimal underlayment, off the shelf residential grade windows and doors, stucko exterior on chicken wire and felt, no exterior paint; apartment grade carpet glued to slab throughout, walls and ceilings all spray painted the same color with 1 year interior latex paint, watered down. minimal kitchen cabinets, all particle board, prefab kitchen counter with vinyl finish. No garbage disposal, dishwasher, or any "extras". Short asphalt driveway with minimal carport on asphalt.
Use that to define all the 7 parameters listed earlier.
And you really think that the home is "equivalent" to a neighboring home custom built the same year with the same 7 parameters, but great view and no traffic noise, and large mature trees and designer landscaping?
If you think they are "equivalent", then you are missing a key element in purchasing decisions.
The "details" are not in the public records. Even if they were, modeling to 100 variables or more would cause even more issues with sparse data. So, instead, to "capture" the difference of amenities and quality, and views, and desirability..., the last sold price adjusted for last sold date is used (in addition to the other parameters). And to "capture" possible improvements since the sold date, the tax assessed value adjusted for local tax rules is used. But as already mentioned, the tax assessed value does not fully capture improvements since purchase, especially if the purchase was made on a property that had substantial deferred maintenance.
If you need more than a machine model with no site visit can provide, you need an appraisal, and there is no getting around that.
Perhaps I was "wrong" and you are only interested saying why you want an different model, or why you don't like the numbers. In which case, I may have wasted my time since this thread only has a views to post ratio of 7.4, meaning very few people are reading it. But if I can't even explain a simple model to others, how are better models going to be developed? Without dialog on the issues, algorithms don't improve.
I known nothing about cardiology, and wouldn't even attempt a computer solution for that; even rough diagnosis.
We are not talking about Cardiology; and the "characteristics" of a house are NOT defined by:
1) house sqft
2) lot sqft
3) # bedrooms
4) # bathrooms
5) # stories
6) year built
7) Neighborhood
For example take a cheep lot next to lots of traffic; but flat, no trees; hard compacted soil.
pour a slab on grade; build a square house, 8 foot ceilings (minimal drywall cutting), scrap pine framing, flat roof with 1:12 slope directly on the ceiling joists (no attic); no roof overhang, minimal insulation, 10 year asphalt rolled roofing; minimal underlayment, off the shelf residential grade windows and doors, stucko exterior on chicken wire and felt, no exterior paint; apartment grade carpet glued to slab throughout, walls and ceilings all spray painted the same color with 1 year interior latex paint, watered down. minimal kitchen cabinets, all particle board, prefab kitchen counter with vinyl finish. No garbage disposal, dishwasher, or any "extras". Short asphalt driveway with minimal carport on asphalt.
Use that to define all the 7 parameters listed earlier.
And you really think that the home is "equivalent" to a neighboring home custom built the same year with the same 7 parameters, but great view and no traffic noise, and large mature trees and designer landscaping?
If you think they are "equivalent", then you are missing a key element in purchasing decisions.
The "details" are not in the public records. Even if they were, modeling to 100 variables or more would cause even more issues with sparse data. So, instead, to "capture" the difference of amenities and quality, and views, and desirability..., the last sold price adjusted for last sold date is used (in addition to the other parameters). And to "capture" possible improvements since the sold date, the tax assessed value adjusted for local tax rules is used. But as already mentioned, the tax assessed value does not fully capture improvements since purchase, especially if the purchase was made on a property that had substantial deferred maintenance.
If you need more than a machine model with no site visit can provide, you need an appraisal, and there is no getting around that.
- pj20
- Contributions:18
A model developer starts testing a new prototype, a cardiologist robot. He sends two patients to him, each has identical characteristics where they are relevant for diagnosis, but they are quite dissimilar in irrelevant characteristics. The cardio/robot refers the first patients to a cv surgeon for a bypass, for the second patient he recommends stress management, dietary changes and regular exercise.
The model developer seeing this would realize he has a serious flaw, as identical patients, in relevant characteristics, yielded divergent diagnoses and treatment. A conscientious developer would not allow his model to be used and he would search for the flaw, to determine was it specific to this problem or more general and would lead to other spurious diagnoses The developer would do this no matter how many times before the cardio robot had been successful. We have a similar situation here, houses that are identical in key attributes are yielding divergent results. Zillow should be as conscientious as this developer, for these are not random noise, they are an instance of substantially identical data yielding divergent results.
By the way, I wonder why you have invested so much time in this, even posting in the middle of the night.
.
The model developer seeing this would realize he has a serious flaw, as identical patients, in relevant characteristics, yielded divergent diagnoses and treatment. A conscientious developer would not allow his model to be used and he would search for the flaw, to determine was it specific to this problem or more general and would lead to other spurious diagnoses The developer would do this no matter how many times before the cardio robot had been successful. We have a similar situation here, houses that are identical in key attributes are yielding divergent results. Zillow should be as conscientious as this developer, for these are not random noise, they are an instance of substantially identical data yielding divergent results.
By the way, I wonder why you have invested so much time in this, even posting in the middle of the night.
.
- Pasadenan
- Contributions:21466
They use less distance for urban and larger distance for rural, but distances are not published anywhere. They also indicate they use sub-county divisions for separate models; over 1 million of them; so that would be some kind of defined borders, with an average of 100 homes each.
They don't use a comparables method, except for the "my estimate", which is temporarily off line as the programmers are doing some revisions.
Here is a thread where a Zillow employee (no longer with Zillow; likely Drew), answered the "comparables" question:
Comps on Zillow
As for random noise causing movement in the same direction? You are incorrect; if that was the case, the noise wouldn't cancel out with large sample sizes and and by taking the median of a large group of units.
You are also making the assumption that the modeling to the parameters is linear, which they have already stated it is not, and similarly, you are assuming there are no negative coefficients.
Again, even with linear regression and only 1st order terms, you would see the movement in opposite directions. (It is believed that there are 2nd order, 3rd order and log terms involved, but that is part of the proprietary learning method, and there is no simple means of verification).
But don't take my word for it; plug the numbers from the recently sold into a multiple linear regression curve fit with a minimum of 6 coefficients and appropriate public records parameters, calculate the coefficients, then use those to calculate estimates from a month ago and present for each home on your block. It is very simple.
Just for illustration... (not applicable to this case as I already explained it is the year purchased and the tax assessed value creating the specific observed phenomenon) say you have 2 bathrooms and the neighbor has 4 bathrooms, and a month ago the sales indicated a bathroom was worth $5000 to a buyer for each over 2, but worth -$8000 for each under two. And this month, the curve fit modeling changes based on what sold, and a bathroom is worth $15000 for each over 3 and worth -$16000 for each less than 3. Then yours drops $16k and your neighbor's increases $5k. Of course this is oversimplified, and doesn't mean the model actually knows what anything specific is worth; only that is what the modeling implies.
And really, if you want an estimate that doesn't fluctuate as much and is easier to understand, use Homesand.net's website:
Homesand.net
It uses Zillow's free API's to get comparable data and then uses the comparable data to do a quick comparable method. The comparables that Zillow provides cross city boundaries, zip code boundaries, and neighborhood boundaries, but not county boundaries (at least it has not been observed yet).
The fluctuations in opposite directions exist on all estimating websites; but Zillow makes it easier to see by giving the delta from the previous month, and by providing 1 year, 5 year and 10 year trend charts for each estimate. The other websites don't want to do that both because it exposes how much the estimates change, and because their interest is different than tracking market changes over time.
They don't use a comparables method, except for the "my estimate", which is temporarily off line as the programmers are doing some revisions.
Here is a thread where a Zillow employee (no longer with Zillow; likely Drew), answered the "comparables" question:
Comps on Zillow
As for random noise causing movement in the same direction? You are incorrect; if that was the case, the noise wouldn't cancel out with large sample sizes and and by taking the median of a large group of units.
You are also making the assumption that the modeling to the parameters is linear, which they have already stated it is not, and similarly, you are assuming there are no negative coefficients.
Again, even with linear regression and only 1st order terms, you would see the movement in opposite directions. (It is believed that there are 2nd order, 3rd order and log terms involved, but that is part of the proprietary learning method, and there is no simple means of verification).
But don't take my word for it; plug the numbers from the recently sold into a multiple linear regression curve fit with a minimum of 6 coefficients and appropriate public records parameters, calculate the coefficients, then use those to calculate estimates from a month ago and present for each home on your block. It is very simple.
Just for illustration... (not applicable to this case as I already explained it is the year purchased and the tax assessed value creating the specific observed phenomenon) say you have 2 bathrooms and the neighbor has 4 bathrooms, and a month ago the sales indicated a bathroom was worth $5000 to a buyer for each over 2, but worth -$8000 for each under two. And this month, the curve fit modeling changes based on what sold, and a bathroom is worth $15000 for each over 3 and worth -$16000 for each less than 3. Then yours drops $16k and your neighbor's increases $5k. Of course this is oversimplified, and doesn't mean the model actually knows what anything specific is worth; only that is what the modeling implies.
And really, if you want an estimate that doesn't fluctuate as much and is easier to understand, use Homesand.net's website:
Homesand.net
It uses Zillow's free API's to get comparable data and then uses the comparable data to do a quick comparable method. The comparables that Zillow provides cross city boundaries, zip code boundaries, and neighborhood boundaries, but not county boundaries (at least it has not been observed yet).
The fluctuations in opposite directions exist on all estimating websites; but Zillow makes it easier to see by giving the delta from the previous month, and by providing 1 year, 5 year and 10 year trend charts for each estimate. The other websites don't want to do that both because it exposes how much the estimates change, and because their interest is different than tracking market changes over time.
- pj20
- Contributions:18
Is that true? They pick all recent sales within a mile radius? If so, that is a problem of a different order. In many urban areas, you can have very different neighborhoods within a mile of each other. Nothing wrong with that, leads to diversity. But grouping all houses within a mile to reach a home value is ludicrous.
I'm guessing though you are pulling my leg.
I'm guessing though you are pulling my leg.
- HomeSand.net, "White Picture"
- Contributions:4396
Pasa is no wrong when he writes ' they use all "none distressed" sales'.
For example: your home is 100 feet from your neighbor home, Zillow picks the comps in ~1 mile radius limited for each property, your home and your neighbor house have same 10 comps in the common 5180 feet radius area, however your home got another new lower purchased price comp on the 100 feet far north where is out of the common 5180 feet radius area and your neighbor's home has a new higher purchased price comp on the far south where is also out of the common 5180 feet radius area. The results are your home got lower Zestimate and your neighbor home got higher Zestimate and you got the differently results if the new comps' purchased prices go otherwise.
- pj20
- Contributions:18
HomeSand,
Perhaps you know Pasadean. He wrote the following to me - I keep telling you, it is not a "comparable" method. They don't use "comparables", they use all "non distressed" sales.
Could you reconcile his comment and your post.
Perhaps you know Pasadean. He wrote the following to me - I keep telling you, it is not a "comparable" method. They don't use "comparables", they use all "non distressed" sales.
Could you reconcile his comment and your post.
- HomeSand.net, "White Picture"
- Contributions:4396
The answer is that Zillow picks the differently comps for the differently houses, if the new comps for your home have the lower purchased prices then your home has a lower estimate and if your neighbor home has the newer comps with the higher purchased prices then your neighbor's home's value will be increased in the Zestimate.
- pj20
- Contributions:18
If both my house and my neighbors house value went up or down at the same time for no explainable reason, that would likely be random noise and nothing to worry about. But when they diverge in different directions for no logical reason, then that is a fundamental problem for the model, as it shows that the internal logic and assumptions are wrong.
.
- Pasadenan
- Contributions:21466
Here is a simple experiment for you to try...
You take your car for a drive to Eureka and back, and you have a friend take their car for a drive from the same starting point to the same destinations.
Then you both calculate your gas mileage.
Then you do the same thing, both driving to Bakersfield. Again both calculating the gas mileage.
Then you compare the difference of the gas mileage for the same vehicle, with the difference for the other.
Does the gas mileage stay within 2% of the previous readings? Does the gas mileage change move the same direction?
There are multiple reasons you will get different answers, and there is a 50% probability that the change in reading between the two vehicles will be the opposite direction. Round off error will easily account for a 5% or more difference. The slope of the car location when filling the tank can easily account for more than a 7% difference. The air pressure for the tires can easily account for a 10% difference. How one accelerates will make a difference. Speed on uphill slopes will make a difference. How one brakes will make a difference. Wind speed and direction will make a difference, especially for headwinds and tailwinds. Weight in the vehicle will make a difference.
If you are not at least 2% off from previous readings for both vehicles, I would be extremely surprised. And if they are off in the opposite direction, there is a 50% probability for that to occur.
Any 1st semester junior college physics student would know that. So would anyone that has ever taken any statistics courses.
You are making a big deal out of a "non-event". And your specifics are indicating the modeling may be much better than the statistics should be able to provide.
Again, if you are familiar with any mathematical or computer modeling methods for any application, you would already understand that those fluctuations are random noise and not the data.
If you want more specifics, I would need the street name and two major cross streets. And even there, I wouldn't be running the same modeling, but only modeling to similar data.
You take your car for a drive to Eureka and back, and you have a friend take their car for a drive from the same starting point to the same destinations.
Then you both calculate your gas mileage.
Then you do the same thing, both driving to Bakersfield. Again both calculating the gas mileage.
Then you compare the difference of the gas mileage for the same vehicle, with the difference for the other.
Does the gas mileage stay within 2% of the previous readings? Does the gas mileage change move the same direction?
There are multiple reasons you will get different answers, and there is a 50% probability that the change in reading between the two vehicles will be the opposite direction. Round off error will easily account for a 5% or more difference. The slope of the car location when filling the tank can easily account for more than a 7% difference. The air pressure for the tires can easily account for a 10% difference. How one accelerates will make a difference. Speed on uphill slopes will make a difference. How one brakes will make a difference. Wind speed and direction will make a difference, especially for headwinds and tailwinds. Weight in the vehicle will make a difference.
If you are not at least 2% off from previous readings for both vehicles, I would be extremely surprised. And if they are off in the opposite direction, there is a 50% probability for that to occur.
Any 1st semester junior college physics student would know that. So would anyone that has ever taken any statistics courses.
You are making a big deal out of a "non-event". And your specifics are indicating the modeling may be much better than the statistics should be able to provide.
Again, if you are familiar with any mathematical or computer modeling methods for any application, you would already understand that those fluctuations are random noise and not the data.
If you want more specifics, I would need the street name and two major cross streets. And even there, I wouldn't be running the same modeling, but only modeling to similar data.
- pj20
- Contributions:18
Pasadenan,
There is no logical or rational reason why the current values of neighboring houses should diverge because one was bought in 1984, the other in 1988.
You previously mentioned a sale of a home on May 3 that was half the size in a different neighborhood as the reason for the divergence, but that makes no sense as well.
I have still have not received a valid reason why this divergence occurs in the model and of course, I won't get a valid reason, because there isn't one.
There is no logical or rational reason why the current values of neighboring houses should diverge because one was bought in 1984, the other in 1988.
You previously mentioned a sale of a home on May 3 that was half the size in a different neighborhood as the reason for the divergence, but that makes no sense as well.
I have still have not received a valid reason why this divergence occurs in the model and of course, I won't get a valid reason, because there isn't one.
- Pasadenan
- Contributions:21466
I already explained it, but you refuse to accept it. You stated you understood computer modeling, but not in the Real Estate industry.
What industry? Weather? Stocks? Bonds? Jewelry? Antiques? Paintings? Horse races? Poker? Traffic? Flood control? Sign wind-loading? Retail Jean sales? Harmonic oscillation failure of bridges? Music?
Perhaps if I know what industry you understood computer modeling I could explain what I'm failing to explain adequately now?
I already told you the difference between your house and the neighbor's house that is causing the different direction of the movement of the estimate is the last sold date and the tax assessed value. And it will be different again in a few months. That is just the nature of the model and the sparse sales data.
I'd like to see you get 98.7% accuracy and repeatability on your math exams.
What industry? Weather? Stocks? Bonds? Jewelry? Antiques? Paintings? Horse races? Poker? Traffic? Flood control? Sign wind-loading? Retail Jean sales? Harmonic oscillation failure of bridges? Music?
Perhaps if I know what industry you understood computer modeling I could explain what I'm failing to explain adequately now?
I already told you the difference between your house and the neighbor's house that is causing the different direction of the movement of the estimate is the last sold date and the tax assessed value. And it will be different again in a few months. That is just the nature of the model and the sparse sales data.
I'd like to see you get 98.7% accuracy and repeatability on your math exams.
- pj20
- Contributions:18
Pasadenan,
That people can be illogical when they buy houses is completely irrelevant to my question. You are right that $14k is not a big deal, but that was not my reason for raising the issue.
My question again was - What is the logical basis for one house to go down $14k in the most recent revision and for the neighbor to go up $14k? Why would this be when there are no differences between the houses that would explain the change in value?.
If a model gives out completely spurious results, as the above truly are, then the model cannot be considered reliable or valid. Again, the issue is not $14k, but two neighbors going in different directions for no logical or rational reason.
That people can be illogical when they buy houses is completely irrelevant to my question. You are right that $14k is not a big deal, but that was not my reason for raising the issue.
My question again was - What is the logical basis for one house to go down $14k in the most recent revision and for the neighbor to go up $14k? Why would this be when there are no differences between the houses that would explain the change in value?.
If a model gives out completely spurious results, as the above truly are, then the model cannot be considered reliable or valid. Again, the issue is not $14k, but two neighbors going in different directions for no logical or rational reason.
- Pasadenan
- Contributions:21466
"What is the logical basis for one house to go down $14k in the most recent revision and for the neighbor goes up $14k?" -
There is NOTHING logical about prices people pay or what they buy, nor when they buy! People by based on emotions, desires, and rumors, not logic.
It is not an issue of "logic", it is a simple math problem. You take six sales, and 6 unknown coefficients, and you solve for the coefficients. Then you plug the numbers back into the resulting model, and you have your results. Six sales is not enough for a good estimate. It is "right" because the math is "right" and it is a "valid method", but that doesn't make it "useful". You need more than 6 sales to get useful results.
So, each new sale changes the coefficients, and you get a different answer, and none of them are moving the same direction because the coefficients changed.
I already explained they don't use linear regression as they have found that many of the relationships are non-linear, and that there is interaction between some of the parameters. But you can try it with multiple linear regression anyway. Look up the LINEST function in your Excel program.
Again, the purpose is not to get consistent suggested list prices! Nor to obtain suggested offer amounts! The purpose is only to track market dynamics on a very small geographical level, and create historic market indexes from these results. The indexes track the market much better than Case Shiller, and includes much more regions and much smaller regions than Case Shiller's method can do.
The "fluctuations" are not the data; that is random noise. I already explained how you can remove the fluctuations.
Besides, you already stated yours was the most valuable on the entire street, thus it must be worth about $1.1 Million? That would mean that $14k is only 1.27% change! Try to get any appraiser to be within less than 1.5% difference from sale price. Even the sold prices tend to be 5% less than list price, and list prices are based on a Realtor®'s CMA.
For a $1.1 million home, Zillow is stating a range of $830k to $2.14 million.
Why would anyone be concerned about a meaningless $14k fluctuation? It is much less than the tolerance range. Round off error would completely account for it.
There is NOTHING logical about prices people pay or what they buy, nor when they buy! People by based on emotions, desires, and rumors, not logic.
It is not an issue of "logic", it is a simple math problem. You take six sales, and 6 unknown coefficients, and you solve for the coefficients. Then you plug the numbers back into the resulting model, and you have your results. Six sales is not enough for a good estimate. It is "right" because the math is "right" and it is a "valid method", but that doesn't make it "useful". You need more than 6 sales to get useful results.
So, each new sale changes the coefficients, and you get a different answer, and none of them are moving the same direction because the coefficients changed.
I already explained they don't use linear regression as they have found that many of the relationships are non-linear, and that there is interaction between some of the parameters. But you can try it with multiple linear regression anyway. Look up the LINEST function in your Excel program.
Again, the purpose is not to get consistent suggested list prices! Nor to obtain suggested offer amounts! The purpose is only to track market dynamics on a very small geographical level, and create historic market indexes from these results. The indexes track the market much better than Case Shiller, and includes much more regions and much smaller regions than Case Shiller's method can do.
The "fluctuations" are not the data; that is random noise. I already explained how you can remove the fluctuations.
Besides, you already stated yours was the most valuable on the entire street, thus it must be worth about $1.1 Million? That would mean that $14k is only 1.27% change! Try to get any appraiser to be within less than 1.5% difference from sale price. Even the sold prices tend to be 5% less than list price, and list prices are based on a Realtor®'s CMA.
For a $1.1 million home, Zillow is stating a range of $830k to $2.14 million.
Why would anyone be concerned about a meaningless $14k fluctuation? It is much less than the tolerance range. Round off error would completely account for it.
- pj20
- Contributions:18
You said that my house was going down because of the May 3 house, while my neighbors went up because of it. But the May 3 house is not similar to any of the houses on this street. It is far smaller on a tiny lot, while the houses here are bigger and on bigger lots.
Here is my question again - What is the logical basis for one house to go down $14k in the most recent revision and for the neighbor goes up $14k? Why would this be when there are no differences between the houses that would explain the change in value?.
You seem to be knowledgeable and intelligent, but please answer this question or leave it alone.
Here is my question again - What is the logical basis for one house to go down $14k in the most recent revision and for the neighbor goes up $14k? Why would this be when there are no differences between the houses that would explain the change in value?.
You seem to be knowledgeable and intelligent, but please answer this question or leave it alone.
- Pasadenan
- Contributions:21466
I keep telling you, it is not a "comparable" method. They don't use "comparables", they use all "non distressed" sales.
It is a curve fit learning modeling method, and since yours is not similar, it has to go down while the others have to go up.
Besides, Zillow won't tell anyone how they define the "vicinity" for each model. We know it is on a sub-county level, over 1 million models; but other than that, we don't have the specifics.
So, here are the 264 recently solds for Zip Code 94610
94610 Recently Sold
And if you don't like Zillow's numbers, I already gave you links to several other sites to get other "free" numbers that will be both much higher and much lower.
And you still didn't give the street name for what you are asking about, so there is no way that I can run the numbers.
I suggest you take it to a local math professor that can explain how the modeling and small sample size of sales affects the numbers.
It is a curve fit learning modeling method, and since yours is not similar, it has to go down while the others have to go up.
Besides, Zillow won't tell anyone how they define the "vicinity" for each model. We know it is on a sub-county level, over 1 million models; but other than that, we don't have the specifics.
So, here are the 264 recently solds for Zip Code 94610
94610 Recently Sold
And if you don't like Zillow's numbers, I already gave you links to several other sites to get other "free" numbers that will be both much higher and much lower.
And you still didn't give the street name for what you are asking about, so there is no way that I can run the numbers.
I suggest you take it to a local math professor that can explain how the modeling and small sample size of sales affects the numbers.
Thus month my zestimate went down while others on the block went up. Why?
Stating a discriminatory preference in an advertisement for housing is illegal. If you think this content is discriminatory or otherwise inappropriate and feel it should be removed from Zillow, please let us know by completing the information above.
We will review this content. Thanks for helping make the site more useful to everyone. To learn more, read Zillow's Good Neighbor Policy.