Increase in Housing Starts Contributes to Lumber Run-Up, or is Nonexistent

As per my post last night, and today's news on housing starts, I certainly ought to eat a little crow; according to the Census Bureau, January housing starts were up from December, as well as January 2009. Specifically, according to the Bureau's report:

Privately-owned housing starts in January were at a seasonally adjusted annual rate of 591,000. This is 2.8 percent (±11.5%) above the revised December estimate of 575,000 and is 21.1 percent (±12.3%) above the January 2009 rate of 488,000.

So yes, it appears I was wrong with my initial assessment and the housing market indeed showed signs of picking up in January, potentially contributing to the Lumber run-up of the last month. However, the actual meaning of the housing start increase is slightly more complicated than the above quoted numbers and, I would make the case, much less meaningful than reports are making it out to be. Let me explain.

First, regarding the meaning of the number itself, the 2.8% increase in housing starts is an increase in the seasonally adjusted housing start rate. Because the US Housing market is highly seasonal (i.e. more building projects begin in the spring and summer months) examining trends on a purely month to month basis is not meaningful when analyzing long-term trends. Housing starts will almost always go up in March, and they will almost always go down in October. As such, the US Census Bureau developed a statistical method called X12 (and its predecessor X11) used to remove the expected seasonal effects of this type of data. The best article I could find describing the algorithm is, oddly enough, on an old Federal Reserve Bank of Dallas webpage, but suffice to say the algorithm is designed to remove expected seasonal effects for a given data series, thereby showing the actual overarching trend. Thus, the published number is a point estimate for what the current annual housing start rate is; in this case, 591,000 housing units started per year. As with any statistical analysis, there is a margin of error to that point estimate, and the margin is given right in the reporting sentence. "This is 2.8 percent (±11.5%) above the revised December estimates..." Wait a minute, 2.8% ±11.5%? That should give us a range of –8.7% to +14.3% and, if I'm not mistaken, –8.7 ≤ 0 ≤ 14.3. As any statistician can tell you, a confidence range that includes zero is not statistically significant at all. And the Census Bureau uses a 90% confidence interval in their calculations, so it's not as though they're being overly conservative with their estimates and confidence ranges. In other words, this number is fundamentally meaningless. In fact, if you actually bothered to read the report (as apparently no one in the media did) you'd see that the "±11.5%" figure is asterisked with a footnote that reads as follows:

"90% confidence interval includes zero. The Census Bureau does not have sufficient statistical evidence to conclude that the actual change is different from zero."

So, in other words, there is no evidence that there was a change in the seasonally adjusted rate WHATSOEVER. This number is meaningless.

Moving on, despite the lack of evidence for a month over month change, the increase in rate from January last to now does appear to be truly significant, with a confidence interval of 21.1 percent ±12.3% (between 8.8% and 33.4%) for the change from January 2009's seasonally adjusted rate. (Note that that range does not cross zero.)

The report also lists the raw housing start numbers (those that haven't been seasonally adjusted). Here are the highlights (all numbers pulled from the aforementioned report):

• 2008 Year in full: 905,500
• 2009 Year in full: 554,500
• January 2009: 31,900
• November 2009: 42,300
• December 2009: 37,100
• January 2010: 37,800

The takeaway from all of this is that the housing market has improved significantly since 2009. In fact, using either the December or January estimate for the seasonally adjusted housing start rate yields a significant improvement over the 2009 total figure. However, there is not sufficient evidence to suggest that any further progress has been made over the last two or three months, the numbers (both seasonally adjusted and raw) do not demonstrate sufficient statistical significance to draw that conclusion. Starker still, is that the market today remains at about 65% of where it was in 2008.

How did the Lumber market today respond to all this? Somehow, despite the hype, prices moved mostly sideways; either this information was already priced in, or it doesn't really exist.

Full Disclosure: As of writing, author is short May 2010 Lumber (LBK10)