RitcheyRch
11-19-2007, 02:25 PM
Something received in an e-mail. Havent had a chance to check snopes or any other sites for verification.
This is a little dense but not overwhelming…interesting model – (Ladera Ranch is in Irvine, Ca)…no factor for the impending ‘mega-disaster’ though…
What I did was try to build a predictive model using the same statistical
techniques used by economists and scientists to glean insights from data.
I did this because most of the time you just see “median price” or “low tier
median price” or whatever, and this tells you very little, and with such a
shallow market (low # of sales), these medians are all over the place… so
I wanted to get the best estimate I could, so I used the tools I know how to
use * statistical regression. The quick summary is that price may be
correcting a lot faster than the medians are letting on.
I took the last 6 months of condo sales in Ladera Ranch off of the MLS
database. I only used 1, 2 or 3 bedrooms condos to try to allow some
breadth of data, but to mostly be comparing apples to apples.
Anyway, before I get into the nitty-gritty of the model, here are my
important findings, the ones that I’m very confident about:
- 1/2/3 bedrooms in Ladera Ranch, on average, are losing $334 in value PER
DAY (over the last 6 months, and there is weak evidence this is
accelerating, and no evidence it is decelerating). This represents a 2.6%
value decrease per month on a 400k home!!! This is a lot higher than a lot
of other estimates, but ties in to all the talk about the low-tier market
falling faster.
- Each additional square foot you add to a property in this band adds $134
to the price, holding all else equal.
- Bedrooms and Bathrooms each add about 25k in value to a house, all other
factors held constant.
- Overpricing your house causes you to lose $0.23 on the final sale price
for every dollar you list it over its eventual sale price (if it sells at
all). Note that this is an anomaly * in good markets, over-pricing often
causes you to get more. Note 2 (sellers): price to sell!
Details:
My approach was to start with what I thought the major factors would be:
- Square Feet
- # of Bedrooms
- # of Bathrooms
- Garage Capacity
- Time at which it was sold (just raw market trend, not “seasonality” –
since I was only looking at 6 months of data its hard to do stuff like say
that Christmas is slow and prove it)
- Square of time at which it was sold (there is some rejiggering in here to
make this work, but this is to crudely capture some
accelerating/decelerating market trends)
- Fixed effects for development (this is an adjustment for whatever tract
the home was in)
- (there were other factors I wanted too look at but which were not in the
data set * HOA dues, taxes, quality of property, the development it is in,
if it was a REO, etc, but I had to work with what I had)
I was somewhat hampered by the lack of data points * I only had 61 to work
with, and more would allow me to make a much better model.
After messing around for a while, I realized I could only get meaningful
results using the following variables, given my shortage of data(if only I
had a full dump of the past year of MLS sales in OC…):
Square Feet
# of Baths
# of Beds
# of days ago the sale closed (I did this analysis on Nov 2)
Results (for Ladera Ranch):
1) As every day goes by, the average price of a house drops by $334. No
joke. On a $400k house that represents price going down 2.6% per month.
This means if you were offering on a house there, and your comparables were
telling you “$400k right now”, you would at the very minimum add 6 weeks of
depreciation in the offer to calculate when the house would actually close.
2) Garage capacity does not seem to predict housing price. This is probably
due to my limited data and because “garages” on MLS are not very descriptive
– you can’t tell if its tandem, or wide, or whatever. It’s also because
garages strongly predict baths and vice versa (bizarre!)
3) For every bed you add to a house, the value goes up about $25,000… But
this value tweaks around depending on how you do the model * it’s not super
accurate, but adding beds, holding all else equal, almost always increases
values.
4) For every bath you add to a house, value goes up about $25,000. This has
a lot of the same issues as the bed count, and has some other issues because
beds and baths “predict” each other.
5) For every additional square foot you add to a property, given that you
have a set # of beds, baths, etc, the value of the property rises by $143.
6) Yes, houses have some intrinsic value in this model at very small sqft
and 1 bed/1 bath, but the lower you go, the less accurate the model becomes.
7) This model still has a fair amount of swing for properties, as you might
expect… obviously, I’m not capturing a lot of other relevant factors and
my model only accounts for about 3/4 of the pricing factors.
Interesting other random results
- For every $1 you list your property over market price, the actual amount
you get for it goes down by about $0.23 in this market. This is
fascinating because other studies I’ve read have shown the opposite, that
every $1 you list over market price, you get $0.50 more. Note that I was
able to prove this result with high confidence, but it wasn’t in my model
because it had a minor interaction with the beds/baths variables and due to
my lack of data, I couldn’t separate it out enough.
- I can definitely show that different complexes sell for more or less, but
didn’t have enough data to make it work out with enough confidence for me to
include it in the model.
Technical notes for mathematicians:
This is an OLS regression. I whitewashed the results, though prior to that
heteroskedasticity wasn’t proven, with my Breusch-Pagan / Cook-Weisberg test
p value at 27%. Boxcox showed a theta of 1.8 reinforcing my choice of a
linear-linear model, which didn’t surprise me much since sqft to price
should be a somewhat linear relationship and sqft was the largest predictor.
Multicolinearity was weak - my VIFs were not exceeding 2, and I don’t see
any reason why my chosen variables should have high multi-colinearity. I
did have to filter out some weak predictors because I Felt they were likely
to have too much multi-colinearity.
Linear regression Number of obs = 61
F( 4, 56) = 79.56
Prob > F =0.0000
R-squared =0.7402
Root MSE =23976
This is a little dense but not overwhelming…interesting model – (Ladera Ranch is in Irvine, Ca)…no factor for the impending ‘mega-disaster’ though…
What I did was try to build a predictive model using the same statistical
techniques used by economists and scientists to glean insights from data.
I did this because most of the time you just see “median price” or “low tier
median price” or whatever, and this tells you very little, and with such a
shallow market (low # of sales), these medians are all over the place… so
I wanted to get the best estimate I could, so I used the tools I know how to
use * statistical regression. The quick summary is that price may be
correcting a lot faster than the medians are letting on.
I took the last 6 months of condo sales in Ladera Ranch off of the MLS
database. I only used 1, 2 or 3 bedrooms condos to try to allow some
breadth of data, but to mostly be comparing apples to apples.
Anyway, before I get into the nitty-gritty of the model, here are my
important findings, the ones that I’m very confident about:
- 1/2/3 bedrooms in Ladera Ranch, on average, are losing $334 in value PER
DAY (over the last 6 months, and there is weak evidence this is
accelerating, and no evidence it is decelerating). This represents a 2.6%
value decrease per month on a 400k home!!! This is a lot higher than a lot
of other estimates, but ties in to all the talk about the low-tier market
falling faster.
- Each additional square foot you add to a property in this band adds $134
to the price, holding all else equal.
- Bedrooms and Bathrooms each add about 25k in value to a house, all other
factors held constant.
- Overpricing your house causes you to lose $0.23 on the final sale price
for every dollar you list it over its eventual sale price (if it sells at
all). Note that this is an anomaly * in good markets, over-pricing often
causes you to get more. Note 2 (sellers): price to sell!
Details:
My approach was to start with what I thought the major factors would be:
- Square Feet
- # of Bedrooms
- # of Bathrooms
- Garage Capacity
- Time at which it was sold (just raw market trend, not “seasonality” –
since I was only looking at 6 months of data its hard to do stuff like say
that Christmas is slow and prove it)
- Square of time at which it was sold (there is some rejiggering in here to
make this work, but this is to crudely capture some
accelerating/decelerating market trends)
- Fixed effects for development (this is an adjustment for whatever tract
the home was in)
- (there were other factors I wanted too look at but which were not in the
data set * HOA dues, taxes, quality of property, the development it is in,
if it was a REO, etc, but I had to work with what I had)
I was somewhat hampered by the lack of data points * I only had 61 to work
with, and more would allow me to make a much better model.
After messing around for a while, I realized I could only get meaningful
results using the following variables, given my shortage of data(if only I
had a full dump of the past year of MLS sales in OC…):
Square Feet
# of Baths
# of Beds
# of days ago the sale closed (I did this analysis on Nov 2)
Results (for Ladera Ranch):
1) As every day goes by, the average price of a house drops by $334. No
joke. On a $400k house that represents price going down 2.6% per month.
This means if you were offering on a house there, and your comparables were
telling you “$400k right now”, you would at the very minimum add 6 weeks of
depreciation in the offer to calculate when the house would actually close.
2) Garage capacity does not seem to predict housing price. This is probably
due to my limited data and because “garages” on MLS are not very descriptive
– you can’t tell if its tandem, or wide, or whatever. It’s also because
garages strongly predict baths and vice versa (bizarre!)
3) For every bed you add to a house, the value goes up about $25,000… But
this value tweaks around depending on how you do the model * it’s not super
accurate, but adding beds, holding all else equal, almost always increases
values.
4) For every bath you add to a house, value goes up about $25,000. This has
a lot of the same issues as the bed count, and has some other issues because
beds and baths “predict” each other.
5) For every additional square foot you add to a property, given that you
have a set # of beds, baths, etc, the value of the property rises by $143.
6) Yes, houses have some intrinsic value in this model at very small sqft
and 1 bed/1 bath, but the lower you go, the less accurate the model becomes.
7) This model still has a fair amount of swing for properties, as you might
expect… obviously, I’m not capturing a lot of other relevant factors and
my model only accounts for about 3/4 of the pricing factors.
Interesting other random results
- For every $1 you list your property over market price, the actual amount
you get for it goes down by about $0.23 in this market. This is
fascinating because other studies I’ve read have shown the opposite, that
every $1 you list over market price, you get $0.50 more. Note that I was
able to prove this result with high confidence, but it wasn’t in my model
because it had a minor interaction with the beds/baths variables and due to
my lack of data, I couldn’t separate it out enough.
- I can definitely show that different complexes sell for more or less, but
didn’t have enough data to make it work out with enough confidence for me to
include it in the model.
Technical notes for mathematicians:
This is an OLS regression. I whitewashed the results, though prior to that
heteroskedasticity wasn’t proven, with my Breusch-Pagan / Cook-Weisberg test
p value at 27%. Boxcox showed a theta of 1.8 reinforcing my choice of a
linear-linear model, which didn’t surprise me much since sqft to price
should be a somewhat linear relationship and sqft was the largest predictor.
Multicolinearity was weak - my VIFs were not exceeding 2, and I don’t see
any reason why my chosen variables should have high multi-colinearity. I
did have to filter out some weak predictors because I Felt they were likely
to have too much multi-colinearity.
Linear regression Number of obs = 61
F( 4, 56) = 79.56
Prob > F =0.0000
R-squared =0.7402
Root MSE =23976