From housing economist Tom Lawler: Yellen on Household Formations “Not Keeping Up;” What Is Her Data Source?
In Federal Reserve Chair Yellen’s surprising “dovish” speech at the Economic Club of New York last week, she noted that a relatively slow pace of household formation was one of the “headwinds” that up to now had continued to restrain the U.S. economy. Here is a brief excerpt from her speech.
“Looking beyond the near term, I anticipate that growth will also be supported by a lessening of some of the headwinds that continue to restrain the U.S. economy, which include weak foreign activity, dollar appreciation, a pace of household formation that has not kept up with population and income growth and so has depressed homebuilding, and productivity growth that has been running at a slow pace by historical standards since the end of the recession.”
emphasis added
I found the statement on the “lagging” pace of household formations somewhat fascinating, in that it is pretty widely known by competent housing economists and demographers that there currently are no good, reliable, or timely data on US household formations. Rather, there are multiple and often conflicting estimates of US household formations calculated based on different surveys conducted by the Census Bureau. In fact, there are even conflicting estimates of US household formations based on the SAME survey conducted by Census, but “controlled” to different benchmarks (in one case housing stock estimates, and in another case population estimates).
Since Chair Yellen was focusing on recent developments, I am assuming she was referring to household estimates from the Housing Vacancy Survey, which is a supplement to the Current Population Survey that (among other things) (1) produces estimate of the share of the housing stock that is occupied vs. vacant; and (2) which produces estimates of the number of occupied homes based on these occupancy share estimates and independent estimates of the housing stock from Census’ Population Division. Census released its report for the fourth quarter of 2015 in late January, and the estimates of the occupied housing stock (i.e. households) in that report did at face value suggest anemic growth in US households last year, after explosive growth in the latter part of 2014.
Here is a chart showing (1) quarterly changes in HVS-based household estimates (annualized, both not seasonally adjusted and seasonally adjusted (by me)), and (2) year-over-year changes in HVS-based household estimates.
As the chart suggests, HVS-based US household estimates are incredibly volatile, with most of the “noise” reflecting sizable (and unrealistic) quarterly fluctuations in estimates of the gross vacancy rate (as opposed to the housing stock). The chart also shows that HVS-based US household growth for the fourth quarter of 2014 was inconceivably high, reaching annualized growth of 5.6 million on an unadjusted basis and 4.7 million on a seasonally-adjusted basis. HVS-based household growth from Q4/2013 to Q4.2014 was 1.9 million, the highest YOY gain since the middle of 2005, while HVS-based household growth slowed to a paltry 462,000 from Q4/2014 to Q4/2015.
With respect to the acceleration in HVS-based household growth beginning in the second quarter of 2014, there were reasons unrelated to actual trends to expect an increase. Beginning in April 2014, a new sample based on the Master Address File compiled during Census 2010 was “phased in” to the HVS estimates, with the phase in period being 16 months. To the extent that the previous (and quite dated) sample was at least partly related to the HVS’ overestimate of the gross vacancy rate (as well as the homeownership rate), one would have expected the HVS estimates beginning in the second quarter to (1) show accelerated (and above “actual” household growth; and (2) shown “surprisingly large” declines in the homeownership rate. While the phase-in of the new sample can’t explain the volatility in the HVS estimates, it probably was at least partly behind the acceleration in household estimates Note that a similar but smaller acceleration in HVS based household estimates followed the phase-in of a new sample beginning from April 2004 to July 2005.
Less easy to understand, however, is why the HVS-based household estimates showed such anemic growth last year.
As noted before, the HVS-based household estimates are “controlled” to independent housing stock estimates, and in essence assume that (1) the HVS estimates for occupancy and vacancy shares are “correct,” and (2) the housing stock estimates are correct. Other household estimates from essentially the same survey (the HVS is a supplement to the CPS) but “controlled” to independent population estimates often show decidedly different household growth estimates. These estimates in essence assume that (1) household characteristics from the CPS/ASEC are “correct,” and (2) population estimates (total and by demographic characteristics) are correct. Unfortunately, such estimates are available only with a considerable lag (and only available for March of each year), and the last estimate available is for March 2015. In addition, the “time series” for CPS/ASEC household estimates are for the most part not adjusted to reflect updated estimates for population counts (total and demographic category, such as age). As such, these time series are of limited usefulness.
Even more disconcerting is that CPS-based household estimates by various characteristics (e.g., age of householder, tenure, etc.) are not at all consistent with decennial Census results, and have not been since 2000. It appears as if the implementation of the “Computer Assisted Information Collection” in 1994 is partly behind some of the “issues” of CPS-based household and housing tenure biases since then. (I’ll have more on this issue in a later report).
ACS estimates are a somewhat better “match” to decennial Census results, but such estimates are only available annually, and with a long lag. (The latest available estimates are for 2014).
Given both the huge volatility in, as well as the at times inconsistency of HVS-based household estimates relative to other household estimates, it is not clear how one should interpret last year’s slowdown in HVS-based household estimates.
Stated another way, the “household estimates conundrum” is alive and well, and quite frankly policymakers don’t currently have any reliable timely data on US households.