Here we go again:
Yesterday’s New Home Sales data of plus 13% month over month was . . . how shall I politely phrase this . . . somewhat questionable.
It seemed like such an outlier, that I had to dig into the details, especially given all the other non-confirming data we have seen: The same time we learned about that huge New Home Sales, we learned elsewhere that Unsold house inventory is at its highest since April 1986. (Existing Home Sales number about 6 times the number of New Homes).
I’ll have more on this tomorrow, but here’s the key takeaway:
a) The data appears to be "statistically insignificant," according to the Census Bureau;
b) Strong historical numbers (like plus 13%) tend to be subject to revision, but mostly stay net postive, albeit somewhat moderated;
c) Over the past 10 years, double digit months have been followed by flat to negative data the very next month (Mean Reversion).
The actual data can be found here.
The first item is the margin of error: Its actually higher than the increase for October as well as the revision for September:
“Sales of new one-family houses in October of 2005 were at a seasonally
adjusted annual rate of 1,424,000 . . . This is 13% (+/- 17.7%)* above the
revised September rate of 1,260,000 and 9.0% (+/-18.2%)* above the
the key: anytime your margin of error is greater than the
estimated increase in New Home sales, confidence levels inthat data are low to non-existent. (A Census Bureau Economist I spoke to agreed with this interpretation).
As significant as that is, let’s ignore it for the moment: Looking back over the past 15 years of data, we see that a mean regression has followed nearly all double digit monthly gains. The subsequent month’s data was significantly lowered — flat to negative in nearly every case:
New Homes Sales
|Month, Year||Double Digit Gain||Subsequent Month||Increase / Decrease|
|June 2003||10.7%||July 2003||(-2.1%)|
|December 2000||11.7%||January 2001||(-4.8%)|
|July 2000||11.9%||August 2000||(-4.4%)|
|November 1998||11.4%||December 1998||(-4.6%)|
|January 1998||10%||February 1998||(-0.7%)|
|March 1995||10.2%||April 1995||0.8%.|
|*February 1994;||10.82%||March 1994||8.89%|
|April 1993||16.45%||May 1993||(-10.70%)|
|September 1993;||12.56%||October 1993||(-3.03%)|
|January 1992||21.15%||February 1992||(-5.47%)|
In nearly all of these months, the subsequent month’s data was significantly lowered. The one exception was *February 1994, which was followed by a strong March and April — but they came on top of January 1994, which has the honor of being the very worst month ever in the history of the Census Construction data: Down -23.77%.
One final factoid: According to the Census Bureau, it takes 6 months to establish a trend for new houses sold. They note this in the fine print:
"These statistics are estimated from sample surveys. They are subject to sampling variability as well as nonsampling error including bias and variance from response, nonreporting, and undercoverage…Changes in seasonally adjusted statistics often show irregular movement. It takes 6 months to establish a trend for new houses sold. Preliminary new home sales figures are subject to revision due to the survey methodology and definitions used. The survey is primarily based on a sample of houses selected from building permits…Explanations of confidence intervals and sampling variability can be found on our web site listed above.
a) A high margin of statistical error means the October data is unreliable;
b) we should expect to see a revision downwards, but not by a whole lot;
c) The November data should be flat to negative.
Hey, I’m an optimistic guy. I was an X-Files fan — I want to believe. <sigh> But sometimes, the dope is so bad that’s its misleading to call it anything but. I only want to get an
accurate read of this planet’s economy in order to know how to position
capital into various asset classes.
Damn! Now I have to go look at GDP . . .
Note: Since this nonsensical data issue has become a recurring theme, I have added Data Analysis as a new subtopic . . .
New-Home Sales Surged in October; Cooling Still Seen
JOI PRECIPHS, JAMES R. HAGERTY and KEMBA DUNHAM
Wall Street Journal, November 30, 2005; Page A2
NEW RESIDENTIAL SALES IN OCTOBER 2005 (PDF)
Census Bureau, Manufacturing and Construction Division
NOVEMBER 29, 2005 AT 10:00 A.M. EST
Comparing New Home Sales and Existing Home Sales http://www.census.gov/const/www/existingvsnewsales.html