Can there be anything funnier than watching CNBC correspondents struggle to make sense of economic data?
I’m sure there is but still, it’s pretty humorous.
Wednesday's New Residential Construction Report really threw them for a loop as it reported a “surprise” jump in national housing starts while also showing a somewhat expected decline in national housing permits.
Various CNBC “Realty Check” segments were dedicated to this supposed anomaly finally culminating with a Diana Olick blog post titled “Starting to Defy Logic” with some of the following text:
“The permits number makes more sense, down nearly 9%, which seems to say builders get that if they build a house right now, it’s going to be hard to sell. But I have to go back to the starts number. What’s up with that? What are these builders thinking??? Every expert I talk to, and trust me I talk to a lot, from Wall Street analysts, to DC industry wonks, tells me that until the builders get their inventory under control, any recovery in the housing industry is going to stall. There’s demand out there, but not that much!”
Well, looking at the data the way Diana views it (look at the following chart and click for a larger version), there’s no wonder she’s having difficulty.
Also, looking at the materials released by the Census Department, not only are there large margins of error, but there are various other distortions that are introduced into the data when it’s compiled.
In order to shed a little better light on what’s going on with permits through completions, Ill go through the steps I took to clean up and chart the data resulting in a much more sensible view.
First, there is one distortion present in the headline permits versus starts and completions that needs to be factored out, namely the fact that not all starts (and subsequent completions) require a permit.
On average roughly 2.3% of projects are started without a permit resulting in there always being a slight disconnect between permits and overall starts and completions.
Fortunately, the Census Department releases two additional series of starts and completions that are compiled ONLY from permit issuing places that can be used to make an apples-to-apples comparison between permits, starts and completions.
Next, as you can see in the prior chart, these series are very erratic and seasonal and also subject to many revisions so instead of attempting to compare the actual series, I smoothed things out a bit by calculating the 12 month moving average for each series.
Keep in mind that we are working with the “raw” unadjusted data series so as not to incorporate any additional smoothing provided by the Census Departments seasonality adjustments.
Now, it’s important to remember that these three series are not simply independent time-data series but are, in fact, three logically related and dependent series.
In the process of a building project, first you get the “permit”, next you “start” building, and finally you “complete” the project.
For this reason, one must adjust expectations prior to reading a newly released Census Department report to account for the true nature of the data published simultaneously each month.
In general, permits “lead” starts by roughly a month so this month’s permits are for next months starts.
By the same token, starts lead completions by roughly six to eight months (it takes that amount of time to build a house) so this month’s starts will complete at least a half a year from now.
Because of this, it would be helpful, for comparative purposes, to shift the starts back one month, and the completions back roughly six months.
This way, you can see that permits are “indicating” next months starts result and starts are indicating the result for completions six months from now.
It’s important to keep in mind that this is not a perfect science as there are many factors that can limit the effectiveness of this kind of manipulation.
For example, its likely that the time between getting a permit and starting a project could be less than a month so in many cases, permits and their subsequent start might occur in the same month.
Additionally, the average amount of time between starts and completions may change over time so comparing a lengthy data series with one time adjustment (either 6 or 8 months but you can’t use both) may exhibit times where the completions correlate well with starts and other times when it does not.
That being said, for the series I’m working with, shifting seems to work fairly well.
Except for being an “order of magnitude” separate, the three series are following a very predictable trend, each one leading and indicating the future of the next.
Regarding the order of magnitude, Ill provide more analysis on this phenomena in a later post but I think for now it’s safe to say that cancellations are playing a role.
For now though, Ill factor out the order difference between each series by normalizing the data to a base of 100 as well as restricting the dates a bit.
The trend is not at all enigmatic as CNBC made it out to be but, in fact, very logical and orderly.