I was all excited to write something about the latest housing starts data from the Census Bureau based on having read some news accounts, but then I read the release itself (PDF) and it turns out the data comes with a ridiculous margin of error:
Privately-owned housing starts in April were at a seasonally adjusted annual rate of 717,000. This is 2.6 percent (±14.8 percent)* above the revised March estimate of 699,000 and is 29.9 percent (±15.2 percent) above the revised April 2011 rate of 552,000.
And note that those large error bands are for a meager 90 percent confidence interval. What's striking is that the very same release includes a much sounder data set about permits:
Privately-owned housing units authorized by building permits in April were at a seasonally adjusted annual rate of 715,000. This is 7.0 percent (±1.0 percent) below the revised March rate of 769,000, but is 23.7 percent (±1.9 percent) above the revised April 2011 estimate of 578,000.
I think my new rule is going to be that we should just watch the permits data and forget about the starts. The downsides to this are that some housing is built in nonpermitting jurisdictions and also that construction projects employ people and building permits don't. But in the aggregate, housing starts are going to be a function of building permits and the government data on permits is way more reliable.