Saturday, September 18, 2010

Lawler: US Households: Why Researchers / Analysts are “Confused”

by Bill McBride on 9/18/2010 08:00:00 AM

CR Note: This from economist Tom Lawler.

[On Thursday, the Census Department released a report] entitled “Income, Poverty, and Health Insurance Coverage in the United State: 2009,” which was widely covered in the press, included a table showing an “estimate” of the number of US households in the US as of March 2010 --- 117,538,000, up just 357,000 from March 2009. The report also shows historical data on this estimated number of households, which is derived from a special Current Population Survey. The table with that data has tons of footnotes, which note that there have been multiple revisions in this so-called “time series.”

This ‘household estimate,” which is not subject to rigorous population or housing unit “controls,” is one of at least five household series one can “pick up” from various Census sources. And, of course, they are all unbelievably inconsistent, both in terms of levels and changes.

Below are various household (“occupied housing units") estimates from different reports/sources. Note that the Housing Vacancy Survey has quarterly average “estimates,” but I am just showing its annual data. I am also only showing data back to 2000.

US Households: What's the "Right" Number? (thousands of units)
 AHS (avg)ACS (avg)HVS, 2008 vintage (avg)HVS, unadj (avg)CPS (Mar)Decennial Census (April 1)
2000 104,819102,555105,720106,434105,480
2001105,435106,429103,772107,010108,239 
2002 107,367104,994104,965109,297 
2003105,842108,420105,636105,560111,278 
2004 109,902106,971106,588112,000 
2005108,871111,091108,667108,231113,343 
2006 111,617109,736109,575114,384 
2007110,692112,378110,173110,306116,011 
2008 113,101110,475111,409116,783 
2009111,861 111,344111,344117,181 
2010    117,538 
Annual Change
 AHS (avg)*ACS (avg)HVS, 2008 vintage (avg)HVS, unadj (avg)CPS (Mar)Decennial Census (April 1)
2001 1,6101,2171,2901,805 
20022049381,222-2,0451,058 
20032041,0536425951,981 
20041,5151,4831,3351,028722 
20051,5151,1891,6961,6431,343 
20069115271,0691,3441,041 
20079117614377311,627 
20085857233021,103772 
2009585 869-65398 
     357 
*AHS: annual average for 2-year period

The “HVS” is the Housing Vacancy Survey, which is the quarterly Census report that includes the homeownership rates and vacancy rates. This report is not actually designed to measure the size of the housing stock (or the number of households), but rather vacancy rates. The “2008 vintage” data are attempts to create a household estimate consistent with historical housing stock estimates from other Census reports. The “unadjusted” HVS data use periodic updates of the housing stock estimates (and updated forecasts), but do not correct for past over- or under-estimates of the housing stock – thus creating multiple discrete shifts in this time series.

I include this one because recently someone sent me a report asking me to comment on a piece by a firm which including showing a decline in the number of households in 2009, citing a Census report. This confused me, but I figured out that the hapless “analyst” had used this “unadjusted” series, which is absolutely useless as a time series. The 2009 and 2010 HVS household (and housing stock) data, are going to be revised downward materially in the upcoming Q3/10 report, reflecting the updated July 1, 2009 housing stock estimates released this June, and then revised in September, which show a MUCH lower housing stock than that assumed by the HVS.

All of these data are available on various Census sources or in vendor economic databases, but the caveats/concerns/issues associated with using them as a time series are often either barely mentioned in footnotes, or not mentioned at all.

Even a casual glance at the [above table] indicates that [these] various measures – some of which cover slightly different time spans, and some (such as the ACS) is a different “concept,” at times show vastly different trends, for reasons that are not at first glance clear. As many housing analysts have noted, how fast the current “excess” supply of housing (which in and of itself is extremely difficult to gauge, given the apparent unreliability of the data!) can be absorbed is heavily dependent both on the level of new construction and the growth in households. Sadly, there not only does not exist a reliable time series of household growth that enables one to look at the behavior during business downturns/recoveries, but there is no reliable time series to gauge how fast/slow RECENT growth has been – though the combined data suggest extremely slow growth over the last few years.

On the housing stock, Census – which released updated housing stock estimates through July 1, 2009 in June (based on a pretty simplistic methodology) -- updated those estimates this month to reflect the fact that the June estimates did not incorporate state/local inputs. [The next table] are the revised housing stock estimates back to July 1, 2000. Census does not have a reliable annual time series for earlier periods.

Census Housing Stock Estimates, July 1
 Revised 2009 Vintage
7/1/2000116,300,799
7/1/2001117,905,005
7/1/2002119,456,206
7/1/2003121,076,837
7/1/2004122,824,501
7/1/2005124,711,041
7/1/2006126,500,212
7/1/2007128,132,164
7/1/2008129,313,137
7/1/2009129,969,653
YOY Change
7/1/20011,604,206
7/1/20021,551,201
7/1/20031,620,631
7/1/20041,747,664
7/1/20051,886,540
7/1/20061,789,171
7/1/20071,631,952
7/1/20081,180,973
7/1/2009656,516

In the “Vintage 2008” HVS data, it was assumed that the housing stock from mid 2008 to mid 2009 increased by about 1,140,000, and that from the spring of 2009 to the spring of 2010 the housing stock increased by a similar amount. In fact, of course, it did not, and as a result there will be sizable downward revisions in the HVS’ estimated household growth in 2009 and so far in 2010.

CR Note: The above was from Tom Lawler. I've tried to figure out when the excess supply will be absorbed, but as Tom points out, it is difficult since there is no reliable time series of household growth.