Zillow responded in another forum to my question on how these numbers were created:
Hi, it's David from Zillow;
Terry -
Zillow is a database of all homes not just those that are currently on the market and so we calculate Zestimate values for all of the homes in our database. With this database we can report the value trends of all homes not just those that recently sold. The beauty of this approach is that it is not skewed by a change in the mix of homes that are selling - a common problem for the traditional approach of reporting value trends using only an analysis of recent sales
I think I understand that David, but I have two questions:Your national and local MSA numbers are then the addition of all of your Zestimates, which were created from the extrapolations from actual comparable property sales, correct? Essentially additions of estimates based on comparable sales, right? Then if I took ALL the sales over a longer period of time, and compared them to the same time and place last year, our numbers should be similar shouldn't they? Well they aren't, they are way off.
Your President has stated that as much as 38% of your Zestimates are more than 10% off, why isn't your margin of error compounded in a fast moving market where prices could be going up or down relatively quickly. I'm not a statistician, but it seems that margin of error must go up in these conditons.
While our numbers and yours do correspond in some regards- for example our more expensive homes have taken the hardest hit- your data indicated the medium to large home have taken a harder hit, a pretty similar picture, there is still a huge variance in our actual results from the MLS which while not 100% complete is a good representative sample of non-builder sales.
I've recently done research on two of the top three leading zips, with appreciation based on average square foot prices from the MLS, reports showing them for the 3 month period ending July 31. In one zip code, there was price appreciation 50% below yours, and in the other? Double digit depreciation! The zip codes in question, 28277 appreciated 4.5%, and 28278 depreciated 11.97%--over May, June and July- quite a difference to Zillow's +9%, the actual reports are below for your review.
Why is this important? Because I am sure one of our local papers will pick up the Zillow press release and report it, probably uncritically, and I'll have to face a seller next week who thinks his property has gone up 9% when it might have fallen 10% from last year.
David, I'd like to be able to use your numbers--really I would if I had a sense they were atleast roughly accurate. They are impressive by their sheer volume and I know that a lot of big-time crunching has been done, but I just can't not explain the difference between our methods or why your addition of estimates would be a better measure of my market than the actual data, as long as there are sufficient transactions to be meaningful statistically. Here are my zip code Market Reports
Market Report 28277- July 2007
Market Report 28278- July 2007
Don't get me wrong, I know Charlotte has fared much much better than most of the country, but the headlines need to be right. Your comments are appreciated. terry
3 comments:
Hi Terry, it's David again;
Your math is incorrect - we don't add up or average the Zestimates to arrive at the Zindex. That would yield useless information.
This is a common misunderstanding - the important point is that the Zindex is the *median* Zestimate. What that means is if you lined all homes up from highest to lowest Zestimate, the value of the home in the middle of the line is the median - the Zindex - and that's the number we use to report trends. We have seen that Zestimate accuracy has no bias - i.e. when we're wrong, we have an equal chance of being high as we do of being low. That implies that inaccurate Zestimates fall evenly on either side of the Zindex - so, when we report the median we are effectively excluding inaccurate Zestimates.
Note that Zillow's quarterly reports measure annual appreciation - i.e. Q2 2007 values are compared against house values at the end of Q2 2006, not 3 month appreciation as per the MLS analysis you did. Your comparison was not apples to apples.
So, the Zindex is actually a rather robust metric for house value trends. It is not subject to the inaccuracy in individual homes' Zestimates and it is not skewed by a change in the market or the mix of housing inventory like an analysis of recent MLS sales can be.
I hope that helps.
Well good... I didn't think I had your National numbers right-- it did appear meaningless to me as well, but not being a statistician, I wasn't quite prepared to say that.
I have read elsewhere that your error margin is off half up and half down, so the median should remove that error.
The 2nd part of your explanation doesn't quite work, because I did use 3 months of 2007 versus the same 3 month period in 2006, albeit not the 2nd quarter, just one month removed though... and we are still far apart.
I have the numbers for our 2nd quarter in a spread sheet, I'll just have to post and comment on them, then we'll know the answer better.
Comment moderation is off.
Thanks Terry - if you remember, please ping me when you post your Q2 YoY data [davidg at zillow]
Post a Comment