Saturday, March 26, 2011

Lawler: Census 2010 and Excess Vacant Housing Units

by Bill McBride on 3/26/2011 05:54:00 PM

CR Note: This long and detailed note on the 2010 Census data is from economist Tom Lawler. Here is a spreadsheet for the 50 states (and D.C.) including the 2000 and 1990 Census data.

This starts with a brief excerpt (click read more for the full post). For those not interested in why some data drives demographers to drink, here is the Summary for Week ending March 25th.

Census 2010: Households, Housing Stock, and Vacant Housing Units: Understanding Why Demographers Drink by Tom Lawler

Census has now released the final Census 2010 counts for state and local population and housing units – occupied and vacant. Here are some national totals, as well as a comparison to the “official” Census 2000 counts – which are “known” to be off, but by uncertain amounts.

Census 2010Census 2000Change% Change
Total Housing Units131,704,730115,904,64115,800,08913.6%
Gross Vacancy Rate11.38%8.99%2.39%

For those who follow housing production, one of the “striking” things about the Census 2010 vs. the Census 2000 data is the apparent growth in the housing stock – 15.8 million units.
According to Census data on housing completions and manufactured housing placements for residential use, the there were about 16.735 million housing completions plus MH placements from April 2000 to March 2010. That would come close to implying that the net loss in the housing stock associated with demolitions, conversions, etc. was only around 935,000, or 93.500 a year – a fraction of the “guesstimate” based on admittedly suspect data from the AHS CINCH studies. (I say “come close” because housing stock estimates include RVs, boats, campers, etc. used as places to live). The number seems especially low given estimates of the housing stock loss just to Katrina and other “natural disasters,” and quite frankly they don’t pass a “sniff test” of reasonableness.

As I have noted before, however, subsequent analysis of Census 2000 by analysts suggested that the Census 2000 housing unit counts were “too low,” though the methodology used to estimate net undercounts was deemed imprecise enough to incorporate into “official” Census data1 . On the population front, subsequent analysis suggested that the Census 2000 population numbers may have been overstated by about 1.3 million, though these new estimates were not incorporated into official Census data2.

If one were to compare Census 2010 with the “unofficial” revised Census 2000 estimates, the comparison table would look as follows:

Census 2010Census 2000 (unofficial revised)Change% Change
Total Housing Units131,704,730116,586,45815,118,27213.0%
Gross Vacancy Rate11.38%9.24%2.14%

Even using these “revised” estimates, the “net” loss in the housing stock from demolitions, conversions, etc. would still have been about 1.62 million units, of 162,000 per year – well south of “AHS-derived” estimates. Interestingly, using the “better” 2000 estimates, population growth over the decade ended April 1, 2010 was just a shade under household growth, suggesting little change in the average size of households (I say “suggesting” because the household population excludes people living in “group quarters,” and that data has not yet been released).

But wait (of course!), there’s more! A Census 2000 supplemental survey (C2SS), using the same “methods and instruments” of the ACS and based on a sample of about 700,000 housing units, “found” a higher vacancy rate than did Census 2000. This result was a bit surprising. Some analysts feel that Census 2000 may have “missed” a significant number of vacant housing units, and that the C2SS may be more “representative” of gross vacancy rates. Others, however, note that the C2SS sample, while large, is dwarfed by that of the official Census.

There is also a “challenge” in using the ACS/C2SS data on Census’ website – the housing units shown in the ACS/C2SS data are “controlled” to equal that of the official Census data (known to be understated). As the paper “Effect of Housing Unit Controls on Survey Estimates” (Albright, Census Bureau)3, the ACS/C2SS estimates of households and housing units varies considerably depending on whether one uses Census 2000 housing units controls, or instead one uses other “weighting” schemes (e.g., Principal Person Factor).

E.g., if one were to assume that the C2SS vacancy rate numbers for 2000 were correct, but if one were also to assume that the Census 2000 REVISED household count numbers were correct, then ACS/C2SS estimates of the 2000 housing units might look something like this. (HU control means Housing Unit Control, while HH Control means Household Control).

Census 2000 OfficialCensus 2000 UnofficialACS/C2SS HU ControlACS/C2SS HH Control
Total Housing Units115,904,641116,586,458115,904,641116,999,235
Gross Vacancy Rate8.99%9.24%9.56%9.56%

If one were to instead use the ACS/C2SS Household Unit Control (this is NOT derived by anyone at Census, by the way), then one would get the following 2010 vs. 2000 comparison.

Census 2010ACS/C2SS HH ControlChange% Change
Total Housing Units131,704,730116,999,23514,705,49512.6%
Gross Vacancy Rate11.38%9.56%1.82%

If the above numbers were correct, and if the Census numbers on housing production were correct, then the net loss in the housing stock resulting from demolitions, conversions, etc. would be about 2.03 million, or 203,000 per year. Even this number is on the low side of “estimates,” but the Census 2010 data do suggest that previous estimates were too high. (Based on AHS data, Census “guesstimated” that the “scrappage rate” for the housing stock would be about 0.025% a year, which for last decade would have meant an average of just over 300,000 per year).

Of course, all of the above comparisons assume that the Census 2010 figures are “correct,” and in fact subsequent analysis will find “issues.” However, early results presented by Census folks on such metrics as % usable responses, etc. suggest that the Census 2010 “error rates” are likely to be lower than those in previous Censuses.

Confused yet? Ah, then mission accomplished!

Any way you “slice” the data, it seems pretty clear that the number of vacant housing units increased dramatically from 2000 to 2010 relative to total housing units, but by how much is not crystal clear. Moreover, there are no “official” data yet on year round vs. “seasonal” vacant units, vacant units for sale or for rent, etc., making it tricky to assess the “excess” supply of housing as of last April.

IF one were to assume that the housing market were “in balance” in 2000, and IF one were to assume that the increase in the “gross” vacancy rate from 2000 to 2010 was a reasonable proxy for the extent of “overbuilding” (relative to household formations) – both assumptions of which are subject to serious debate – then a “reasonable” range for the “excess” supply of housing as of April 1,2010 would be somewhere between 2.14 million units (using Census 2010 and ACS/C2SS controlled for Census 2000 adjusted households) and 3.15 million units (using Census 2010 and “official” Census 2000 data) – with the high end of that range almost certainly being “too high.”

These assumptions, however, do not take into account the likely faster growth in “seasonal/second” vacant homes that would have occurred without any “housing bubble”/overbuilding. Both household an population in the 45+ year old age groups was materially faster than that of younger folks/household last decade, and these older age groups have a much higher likelihood of owning a seasonal/occasional use/second home than is the case for younger households. Trying to figure out how fast “vacant seasonal” homes would have grown relative to the overall housing stock in the absence of the “bubble/bust,” however, is not a simple task, and is hampered by controversy over the MEASUREMENT of the total number of seasonal/second homes!4

Doing a crude adjustment for this “seasonal/second” effect, I estimate that the “range” of possible “excess” supply for April 1, 2010 would shift to the two to three million range, with the latter almost certainly “too high.”

Since April 1st of last year, of course, housing production has been subdued – based on my estimates for housing completions in March and for manufactured housing placements from January to March, I’d estimate that housing completions plus MH placements totaled about 675,000 units. If one conservatively assumed that about 225,000 housing units were “lost”, then the housing stock on April 1, 2011 would be about 450,000 higher than last April 1st. The number of VACANT homes, however, would depend on household growth, and sadly there are no good, reliable, and/or timely data on households – though the HVS data SUGGEST (but just that and no more) that household growth was “decent” in the last three quarters of 2010. If one were to assume that household growth over the 12 months ended in April were about a million, then estimates of the “excess” supply of housing as of (almost) right now would be somewhere in the range of 1.45 to 2.45 million units – with the latter (again) almost certainly too high.

If one were to assume that the current “excess supply” of housing was around 1.75 million, and that the net housing stock lost to demolitions is about 225,000 per year, then here is a “matrix” of how many years it would take to absorb that excess supply for different combinations of housing production and household formations.

Years to Absorb 1.75 million "excess supply" of housing units
Household Growth (thousands)
Housing Production (thousands)9001,0001,1001,2001,3001,4001,5001,600

E.g., if housing production (completions plus MH placements) ran at around 600,000 per year, and household formations increased to 1.2 million per year, it would take a little over two years to absorb a “reasonable but possibly low” estimate of the current “excess” supply of homes – unless, of course, the number of homes “lost” to demolition, conversion, etc., ran above 225,000 per year.

It is easy to see why many housing economists view the current “low” level of housing production as a plus for the overall health of the housing market, even before allowing for the current high number of residential mortgage loans either seriously delinquent or in some stage of foreclosure. It is also easy to understand why competent housing analysts believe that any government policies designed to encourage additional construction of housing units would be “dumber than dishwater,” and that the only government policies designed to encourage increased “AD&C” activity would NOT be acquisition, development, and construction lending, but instead “acquisition, destruction, and or conversion” of existing vacant housing units.

Of course, the “matrix” above was for the nation as a whole, and estimates of the “excess supply” of housing would vary massively by states. Using the “official” Census gross vacancy rates, there was an increase of over three percentage points from 2000 to 2010 in eight states: (in order) Nevada, Florida, Michigan, Georgia, Rhode Island, South Carolina, Ohio, and Arizona. And, of course, within states with overall “modest” vacancy rate increases there were huge increases in some counties – California being one of many examples.) Some state data are shown in the table at the end of this report. In addition, the matrix doesn’t account for the “types” of homes that are vacant versus the “types” of households that might be formed.

Flipping back to household growth, the Census 2010 data suggest that the number of households grew by about 1.1 million per year in the decade ended April 1, 2010, down from what appears to have been around 1.34 million per year in the previous decade (there are lots of issues with the Census 1990 data!). While the multiple and conflicting government data on inter-censal households are extremely unreliable, all suggest that household growth slowed massively in the last few years of last decade. Household growth appears to have been somewhat slower than what one would have expected given population growth by age cohort combined with 2000 headship rates by age cohort, with apparent “headship rate” declines occurring in the last few years of last decade – suggesting the potential for an above-trend rebound in household formations once the economy and jobs market picks up steam. (Official 2010 census data by age of population and by age of householder are not yet available).

For those who don’t remember, the Joint Center for Housing Studies in 2006 upped its forecast for household growth to what at the time seemed an inanely high 1.456 million per year for the 2005 to 2010 period, partly because it increased its assumption for net immigration growth to 1.2 million a year (which was right around the time net immigration per year began to slow), and partly because of its use of highly suspect “headship rates” from the CPS/ASEC. That forecast, combined with what proved to be excessively high estimates of net housing loss estimates and a few other factors, led to widely-quoted (often by home builders!) “projections” that “demographic” demand in the latter half of last decade would “support” close to 2 million units of housing production per year! Available (though not great) data suggest that household growth began to slow shortly after the JCHS upped its forecast. While many believe that the JCHS’ analysis was criminally bad, to the best of my knowledge no litigation has been filed against it.

2 See, e.g., Comparison of A.C.E. Revision II Results with Demographic Analysis
3 Available at Effect of Housing Unit Controls on Survey Estimates