Inventory changes vs. Price Changes. verrry interesting
Rhodes asked in her headline What gives? Local home-sales market is softening, but prices keep rising
Local home sales cool off; why are prices still hot?
I think I may have a good answer.
My theory was that while supply drives price, price changes probably lag inventory changes - so I took the 2001-2007 data, and started playing around with different comparisons of monthly inventory changes vs. monthly median price changes. I compared no lag, 3, 6, 12, and 15 month lags to see which had the best correlation.
What I found was that a 12 month lag was the sweet spot, giving me the best explanatory value with an r-squared of 0.624. So as inventory goes up, the change in price doesn't really show up for a FULL YEAR.
Where this gets interesting is, I now have a relationship that I can use to forecast the next 12 months of prices, based on the last 12 months of inventory growth.
That model gets ugly quickly...check it out
Month InvCh Predicted Price
Jun-06 17.17% 6.27%
Jul-06 19.29% 5.82%
Aug-06 23.91% 4.81%
Sep-06 28.79% 3.75%
Oct-06 30.78% 3.32%
Nov-06 27.72% 3.98%
Dec-06 22.94% 5.02%
Jan-07 21.93% 5.24%
Feb-07 22.50% 5.12%
Mar-07 32.59% 2.93%
Apr-07 38.42% 1.66%
May-07 44.17% 0.42%
Next I am going to run this against Case Shiller data - which I expect will give me an even better model
Local home sales cool off; why are prices still hot?
I think I may have a good answer.
My theory was that while supply drives price, price changes probably lag inventory changes - so I took the 2001-2007 data, and started playing around with different comparisons of monthly inventory changes vs. monthly median price changes. I compared no lag, 3, 6, 12, and 15 month lags to see which had the best correlation.
What I found was that a 12 month lag was the sweet spot, giving me the best explanatory value with an r-squared of 0.624. So as inventory goes up, the change in price doesn't really show up for a FULL YEAR.
Where this gets interesting is, I now have a relationship that I can use to forecast the next 12 months of prices, based on the last 12 months of inventory growth.
Predicted MedianPrice = -0.217(YoY inventory change 12 months earlier) + 10%
That model gets ugly quickly...check it out
Month InvCh Predicted Price
Jun-06 17.17% 6.27%
Jul-06 19.29% 5.82%
Aug-06 23.91% 4.81%
Sep-06 28.79% 3.75%
Oct-06 30.78% 3.32%
Nov-06 27.72% 3.98%
Dec-06 22.94% 5.02%
Jan-07 21.93% 5.24%
Feb-07 22.50% 5.12%
Mar-07 32.59% 2.93%
Apr-07 38.42% 1.66%
May-07 44.17% 0.42%
Next I am going to run this against Case Shiller data - which I expect will give me an even better model
Comments
A 12 month lag again turned out the best model (which gives me comfort about the approach) and using this data, I get an r-squared of 0.829!!
CS is 2 months in arrears in their reporting - meaning we get 14 months of forecast.
Here's the fugliness. I backed in to the M2M numbers as well.
Month Y2Y M2M Actual
Apr-07 9.13% 0.8% 9.6%/1.3%
May-07 6.94% -0.5%
Jun-07 4.99% -0.2%
Jul-07 4.40% 0.6%
Aug-07 3.11% -0.2%
Sep-07 1.74% -0.7%
Oct-07 1.18% -0.2%
Nov-07 2.04% 0.9%
Dec-07 3.38% 1.4%
Jan-08 3.66% 0.2%
Feb-08 3.50% 0.4%
Mar-08 0.68% -1.9%
Apr-08 -0.96% -0.8%
May-08 -2.57% -2.2%
Check out the scatterplot...
Good question. Because inventory is derived from last month's inventory + houses added - houses sold, I think adding sales as another variable is to some degree auto correlated with inventory.
So if I tried a multivariate regression, I would probably use net adds and sales as the two independent variables - not inventory and sales.
that said, my CS model above is pretty freaking good. I know that correlation <> causation - but it fits pretty well with a generally accepted economic theory....
I think the 12 month lag works, but the slope of the line will be different for SD vs. Seattle - so I need the inventory data for that market.
If you know where to find it, give me a pointer and I'll run the model. The general concept shoudl be applicable.
I'd bet that the down trend will be smoother than what I show here - w/o that little run-up, but the slope and depth are probably in the ball park
Can you get the info needed from this link?
http://sandicor.com/statistics/stats200 ... stics.html