Incipient Housing Demand

For the housing market to come back into some approximation of balance will require less supply and more demand.  Additions to supply from new construction have been reduced to a trickle, but demand growth, until recently, has been far below underlying demographic potential, prolonging the imbalance.  There are finally signs that demand is recovering.

Although demand is frequently assessed based on home sales, the more fundamental expression of demand is the net formation of households.  A household is defined as an individual or group occupying a housing unit, so the number of households is, by definition, equal to the number of occupied units.  If the adult population in each age bracket maintained a consistent tendency to form independent households, the growth in the adult population would translate into about 1.2 million households per year.  Beginning in 2007, as the economy headed toward recession, however, the rate of increase in the number of households fell sharply, to about 500,000 per year.  That slowdown was partly caused by lower net immigration, but it was mainly due to various forms of doubling-up.  Doubling-up included more people living with roommates, partners, other families, etc., but the most common manifestation was an increase in the share of young (and not so young) adults living with their parents.

Among the population aged 25 to 29, for example, the share living in their parents' homes rose from about 16 percent in March 2007 to 18 percent in March 2010, while the share in that age group who were household heads declined from 47 percent to 44 percent.  With about 20 million people in that age bracket, the decline in the "headship rate" of 3 percentage points translated into 600,000 fewer households.  Across all age groups, changes in headship rates from 2007 to 2010 reduced growth in the number of households by about 2 million during 2007 to 2010.

There are a number of conflicting measures of the number of households, none of which is very precise, and many of which are only available after a substantial lag.  The most current information is contained in the monthly Current Population Survey, which is conducted primarily to measure unemployment.  The household totals from the monthly CPS are not shown in any standard reports, although a variant, from the CPS-related Housing Vacancy Survey, is available with a few months lag.


I tabulated some of the CPS monthly data through May 2011, and the data indicate a rebound in the rate of increase in the number of households in recent months.  From May 2010 to May 2011, it appears that there was an increase of about 1.6 million, and the average number of households for the first 5 months of 2011 represented an increase of 1.3 million from a year earlier.

The CPS data do not show that there has been a decline yet in the share of young adults living with their parents.  The increase in headship comes instead from more people living alone rather than in couples or groups.

In the 1970s, a surge in divorces contributed to an especially large increase in the number of households, and the recent pattern is somewhat analogous.  The divorce rate hit a peak in 1979, however, and is now lower.  You can't get divorced without first being married, and that prerequisite has been less frequently met.  Over time, but especially in the last few years, the share of people who have never been married has grown, and recent growth in single-person households has come largely from never-married young adults.

There's a lot of noise in the monthly CPS numbers, but if the recent data are not just driven by random sampling error, by deficiencies in Census Bureau population estimates and weighting schemes, or by programming mistakes on my part, it suggests continued declines in vacancy rates and stabilization in the overall housing market.  If growth in the number of households comes mainly from single-person households, however, that implies demand for smaller, multifamily rental housing, in urban settings, rather than a near-term revival in demand for suburban detached homes.

Technical Note:
Household surveys do not actually count the number of households—or the number of employed or unemployed people.  The surveys measure proportions of those surveyed and those proportions are multiplied by estimates of the number of housing units or of the population in different categories.  Most estimates of the number of households, including those from the ACS, AHS, and HVS, are calculated by multiplying an occupancy rate by the estimated total housing stock.  The CPS data discussed here are instead tied to estimates of the population.  If the survey finds that 45 percent of non-Hispanic white women aged 30 to 34 are household heads, and the estimated population for that category is 10 million, that means 4.5 million households.  Adding up the various population categories then produces an estimate of total households. (That's an oversimplified, but hopefully instructive, description.) Whether estimates based on population controls or housing stock controls are superior is debatable.  The population-based estimates are closer to the household total in the 2010 census, but the decennial census data tend to count some vacant units as occupied.

The monthly CPS is not identical to the annual March CPS supplement that collects more detailed demographic data. The latest annual data are for March 2010 (actually collected February to April).  The annual data for March 2011 won't be available until September.

June 24, 2011 - Comments (1)

Housing Preservation Gone Wild

A comparison of the change in the housing stock between 2000 and 2010 , as reported in the decennial census, with the number of units built over that period indicates losses from the stock averaged 94,201 housing units, or 0.08 percent of the stock, per year.  If only 0.08 percent of the housing stock were lost each year in the future, the majority of today's homes would still be standing and habitable 850 years from now.

The total number of housing units as of April 1, 2010, was reported as 131,704,730, compared to 115,904,640 on April 1, 2000.  The increase of 15,800,090 compares to 16,742,100 new units produced (15,424,800 conventional units completed and 1,317,300 mobile homes placed).  The residual, 942,010, is the implied number of net losses.  For the period from 1990 to 2000, the comparable numbers (115,904,640-102,263,678 vs 13,359,200+2,854,029) translate into an implied net loss of 2,572,267.

The small number of apparent net losses is probably due, at least in part, to measurement problems.  There could have been fewer units missed, or more units double-counted or erroneously included, in 2010 than in 2000.  An analysis by the Census Bureau of housing unit coverage in the 2000 Census estimated that there was a net undercount of 0.61 percent, or about 700,000 units, in 2000.  There were some refinements in the address lists and procedures for the latest count.  If that meant that the net undercount  in 2010 was zero, the implied number of net removals would be higher by about 70,000 per year.

If the reported volume of new construction was less than actually occurred, that would also bias the implied net removal value downward.  In general, data for housing starts and completions are among the more reliable economic statistics.  There was, however, a change in the methodology for estimating the number of new units that produced lower numbers during the latest decade than would have been reported without the change.  The most important change, applied to estimates of single-family construction in permit-issuing areas beginning in January 1999, was to remove a 3.3 percent upward adjustment that had been incorporated to account for the possibility of units built (illegally) without permits.  If the previous methodology had remained in use, the estimate of construction during the latest decade might have been higher by about 35,000 units per year.

It should be noted that on a previous occasion the inconsistency between the growth in the stock, as measured by the decennial census, and the estimates of construction led to a massive revision in the construction estimates.  In 1964, the data for housing starts from 1945 to 1958 were restated, raising estimates for the years from  1945 to 1949 by 48 percent, with estimates for other years subject to smaller revisions.  If there is an underestimate in recent data, however, the magnitude is almost certainly small.

Net losses reflect a number of factors beyond the destruction of houses by demolition and natural disasters.  Vacant units that are no longer characterized as habitable are not counted in the stock, even though they haven't actually been destroyed.  Some will be demolished later, but others may return to the housing stock.  Other losses from, and additions to, the stock occur as structures move between residential and nonresidential use, or are subdivided or merged.  Most studies in the past have indicated that changes in use tend to cancel out at the national level, and that net losses are largely driven by permanent removals through demolition and disaster, but that may not have been the recent experience.  There are no good, direct measures of either net losses or of demolitions.

One type of addition to the stock that is particularly overlooked in the available data consists of the conversion to residential use of structures that had never previously been used for housing.  Although conversions of schools, factories, or other nonresidential structures to housing often require extensive construction work, they are not counted in housing starts or completions or in most other measures of residential construction.  For California, where the state reports data from the Construction Industry Research Board that includes permits for such conversions, total permits for 2001 to 2010 were 2.6 percent higher than the Census Bureau permit data show.  For states like New York the relative share of such reconstruction may be even greater.

Even taking all the possible measurement and definition issues into account, it appears that the rate at which units have been removed from the housing stock in recent years has been remarkably low.  Along with a rate of new construction that is smaller, relative to the size of the stock, than during most periods in the past, this has meant relatively more older housing.  The median age of the stock, as reported in the 2009 American Housing Survey, was 35 years.  Although that's a long way from the extrapolated value of 850 years suggested above, it's probably more than at any previous time in U.S. history.  As recently as 1985, the median age of the stock was about 23 years, down from 28 years in 1950.

With the current large number of vacancies and the lack of maintenance provided to many distressed properties caught up in the foreclosure crisis, as well as the seeming proliferation and severity of natural disasters, losses from the stock might be expected to rise.  Indeed, that may be a key part of the process by which supply and demand come back into balance.  For assessing the economy, planning business activities, and formulating public policy, it would help if we had some way of tracking this.

It has been almost 3 years since I have written anything for this blog.  I actually forgot how to do so, and managed to get caught up in various projects.  At the moment I have fewer commitments.  I only mention this so you know that the reason why the previous post was in 2008 is not a problem with your computer.

June 8, 2011 - Comments (1)