Special Studies, June 17, 2008
By Paul Emrath, Ph.D.
Report available to the public as a courtesy of HousingEconomics.com
Unsold inventories of homes are routinely reported each month. Among the statistics typically reported are inventories of new single family homes, existing single family homes, and existing condominiums on the market. Noticeably missing from this list is an inventory of new unsold condos, a statistic that should be of particular interest given recent developments in U.S. housing markets.
Yet an inventory of new condos (not completely equivalent to the other housing inventory measures, but similar) is published by the government each quarter. According to these data, the inventory of new condos that were completed in the previous four quarters but not yet sold grew by over 400 percent since the mid-1990s—from 5,000 units in 1996 to over 20,000 in 2007. Most of the increase occurred after 2005, in the period when single family housing starts had started to tail off.
In addition, a new data series published by the Census Bureau can be used to construct inventories of relatively new condos by state and metro area, although the numbers are only produced annually and become available with a lag. These local condo inventories, which are subject to sampling error and sometimes based on small numbers that may be anomalous, range from an estimated zero in some states and metros, to highs of a nearly three year supply in Tennessee, and a four and a half year supply in the Detroit metro area.
Trends in multifamily condominium production have changed considerably over the past decade and a half. During the historic low point for multifamily construction in 1991 through 1993, a little over 40,000 condos (multifamily units intended for sale) were started per year, accounting for roughly one-fourth of all multifamily starts. From that trough, overall multifamily production increased more or less steadily until it more than doubled by 2003.
Condo starts also more than doubled over that stretch, reaching 87,000 units a year by 2003. Even by then, however, condos were still only accounting for about one-fourth of multifamily production. The fact that the condo share of new construction remained relatively static from 1991 to 2003 seemed somewhat puzzling.
Overall homeownership was increasing (at times supported by explicit government policies), the number of households in the age brackets that typically buy condos was on the rise, and strength of demand was evident in rising prices for existing condos, and fast absorption of new condos. Industry analysts began to speculate that builders were turning some multifamily structures originally classified as rental apartments into condos at some point during the construction process in a way government statistics were not capturing. However, the American Housing Survey (which is conducted in odd-numbered by HUD and the Census Bureau, and is based on a survey of occupants done after units are completed) also showed a condo share of new construction that was flat through 2003 then spiked in 2005 (the latest year for which the data are available).
Then, the production of new condominiums essentially exploded. Starts of new multifamily condos grew until they exceeded 150,000 and accounted for 45 percent of multifamily starts at the peak in 2006, before falling off in 2007. Although the decline in condo production after 2006 was significant, the condo share of new multifamily construction was still over 38 percent in three of four quarters in 2007. This is leading some observers to wonder if builders are now converting structures originally classified as condos into rental apartments during the construction process in a way government statistics are not capturing.
The Census Bureau conducts two surveys that generate information about new condos. The Survey of Construction (SOC) is based on a nationally representative sample of residential building permits (supplemented with estimates of construction in non-permitting places) and follows the sample through until the units are completed. It is the instrument that generates the familiar series on housing starts, completions, and the value of new residential construction. Among the information collected in the SOC is whether the units are intended for rent or sale. Apartments intended for sale should be condos (or co-operatives), as a legal arrangement is needed to share ownership of the common areas.
The Survey of Market Absorption (SOMA) is a follow-on survey conducted by the Census Bureau and sponsored by the US Department of Housing and Urban Development that collects information from a sample of units completed each month in buildings with five-or more apartments. The SOMA then revisits these units 3, 6, 9, and 12 months later to see if they have been occupied or “absorbed”—which, in the case of units classified as condos, is interpreted as meaning they have been sold.
If plans change and an apartment building is transformed from condo to rental (or vice versa) sometime before completion, the SOC is not likely to capture the change. However, the SOMA does not pick up information on intended use from the SOC, but collects this information independently during its initial survey of a newly-completed apartment building. Thus, if condo-to-rental (or rental-to-condo) conversions are occurring systematically between start and completion, it should be possible to detect the trend by comparing SOC and SOMA five-plus completion numbers.
In recent years, however, the SOC and SOMA numbers agree quite closely. The SOC shows 79,000 five-plus condos being completed in 2005 and 104,000 in 2006. The equivalent numbers from the SOMA are 82,000 and 103,000. However, after its initial visit, information collected by the SOMA is largely restricted to the share of units that are vacant and occupied. So if a conversion takes place between completion and absorption, the SOMA is not likely to capture it.
The 2006 surge in condo production occurred against a backdrop of a housing market that was showing signs of weakness. Starts declined while the inventory of new and existing unsold single family homes on the market has increased substantially. Although a large share of the weakness is attributable to turmoil in capital markets, it is generally acknowledged that the presence of speculator/investor buyers at the peak of the boom was a contributing factor.
Speculation and investment may refer to buyers who have no intent of living in the homes, but intend to turn them over, often after a very short holding period, in order to realize a profit. However these terms can also be applied to renters who made the transition to homeownership sooner than they otherwise would have, in order to take advantage of anticipated appreciation. As markets weakened, speculators looking to sell quickly found it difficult to get the prices they were anticipating. Marginal buyers found it difficult to maintain mortgage payments, and without the anticipated house price appreciation lacked the option selling the home to pay off the mortgage. These factors contributed to an excess supply of homes on the market.
Investors have traditionally accounted for a relatively high share of condo apartment purchases. Moreover, because multifamily condos use land more intensively and tend to be smaller, they can often be made available at somewhat lower prices, and for this reason may be attractive to marginal buyers. Hence, condo inventories could be more sensitive than other segments of the market to stagnant or declining prices.
National Inventory Measures
The two often-cited sources of information on the inventory of unsold single family homes are the Census Bureau’s “Sales of New One-Family Homes”, and the existing home sales series from the National Association of REALTORS® (NAR). Both sources compare the number of unsold homes on the market to the number sold and publish a “months’ supply” measure of the inventory. The Census Bureau’s series does not include multifamily construction. NAR does publish separate sales and inventory numbers for existing condos.
Although NAR’s months’ supply of existing homes has been higher than the Census Bureau’s months’ supply of new homes on the market during certain stretches since 1999, the differences are small, and there are many cases where the opposite is true. On balance the new and existing months’ supply numbers track each other quite well, even during the post-2005 period of rising inventories (Figure 1).
NAR’s months’ supply of existing condo series has always been more volatile than its months’ supply of single family homes, but on average the two tracked each other fairly well until 2005. After that, both series started to escalate, but the existing condo inventory rose at a considerably faster rate, eventually climbing to a historic peaks of over 12 months’ supply on the market (Figure 2).
In order to answer questions about how new condo inventories may have fared in this environment, we can look at the SOMA. Although these specific statistics are seldom cited, the quarterly SOMA release reports on the number of (five-plus) condos units that were completed during the last four quarters, the number that were sold during that time, and the number that remain for sale at the end of the period. In short, the SOMA generates an inventory of unsold new condos, based on a rolling sample of four quarters of construction.
The SOMA shows that the rolling four-quarter measures of completions and sales both increased rapidly after 2004, but after 2006 the inventory gap between sales and completions widened appreciably (Figure 3).
To provide a clearer picture of the gap, Figure 4 charts the inventory of new condos that remain on the market unsold after four quarters. This new condo inventory was about 5,000 units at the start of 1996. The inventory remained under 10,000 units—usually well under—until 2006, when it jumped to 12,700 and kept on rising thereafter. In the latest quarter, the inventory of new condos still on the market after one year stood at a record 21,200.
Technically, it’s possible to express the SOMA inventory of newly completed condos in terms of months’ supply, but the resulting numbers are much smaller (well under 3 months even in the last quarter) and not really comparable to other months’ supply measures. Part of the reason is that, after a year, the SOMA quits tracking the condos whether they’re sold or not.
Recently, the continued existence of the SOMA has been called into question. Late in 2006, HUD announced plans to discontinue the survey on the Census Bureau’s web site. A letter writing campaign organized by NAHB was successful in overturning the decision so data on the absorption of condos (as well as rental apartments) will continue to be available from the government in the immediate future. In the longer term, the future of the SOMA is tied to the budget for HUD’s Office of Policy Development & Research, which has been under considerable pressure in recent years.
States and Metros
The supply of condominiums is not spread evenly across the country. It tends to be concentrated in areas where buildable lots are scarce and land costs are consequently high—vacation spots near mountains or beaches, or near employment centers in large metropolitan areas, for instance. Surveys like the SOC and SOMA are based on national samples that are not designed to provide local information. However, it is possible to extract some local information on condos from the American Community Survey (ACS).
The ACS is the Census Bureau’s mechanism for generating housing data previously available only once a decade from the Census on a more continuous basis. The ACS collects information on structure type, year the structure was built, and the ownership and vacancy status across the country on an ongoing basis and releases data sets every year.
From these it’s possible to produce counts of multifamily units that were 1) built recently, and 2) either owner-occupied or vacant for-sale at the time the ACS was conducted. As of this writing, the most recent ACS data available are from 2006.
The 2006 ACS data show 55,462 units in five-plus structures that were built in 2005 or 2006 and either owner-occupied or waiting to be sold at the time of the survey. Because ACS data are gathered over the course of a year, the 2006 ACS does not capture all housing units that were built in 2006. If data collection and condo production both take place at an even pace throughout the year, the 2006 ACS would capture all the units built in 2005 and half those built in 2006.
The ACS number of 55,462 five-plus condos is drastically lower than the number of completions reported by the SOMA in 2005 and 2006. In fact, the SOMA shows about 80,000 five-plus condo completions in 2005 alone. Some of the difference could be due to condo-to-rental conversions that the SOMA misses.
However, there are other reasons why ACS and SOMA numbers are not perfectly equivalent. A major difference between the two surveys is that the SOMA interviews the owners of the apartment buildings, and the ACS (at least, in the cases where the units are occupied), interviews the occupants. If a unit is sold as a condominium, but the owner subsequently rents it out to another tenant, it is not captured in the universe of condominiums in the ACS. The same is true if the vacant units are pulled off the market or otherwise identified in the ACS as something other than “vacant for sale.” It would be interesting to see how the spread between the ACS and SOMA new condo numbers varies over time, but this is not possible at present given limitations of the ACS data.
Of the units built in 2005 or 2006 that can be identified as condos in the ACS, 39,621 were owner-occupied, and 15,841 were still for sale (Table 1). It’s possible to convert the ACS condo inventory to a months’ supply, although timing is an issue. In theory, the 2006 ACS should capture condos that were sold and occupied by their owners during any of the twelve months of 2005. However, it will miss any sales that occurred in 2006 after the survey was conducted. We assume that, on average, that activity for half of the year is missed and base the inventory calculation on 18 months of sales (for all of 2005 and half of 2006). This works out to a 7.2 month’s supply of new condos on the market for the 2006 ACS.
As expected, the units that can be identified as recently built condos in the ACS are concentrated in particular states. About 97 percent of the 55,462 that were built in 2005 or 2006, 96 percent of the 39,621 that were sold, and 100 percent of the 15,841 in the unsold inventory, are located in only 27 states.
These 27 states are shown in Table 1. They are ordered according to the size of their inventories of new unsold condos, measured in months’ supply. There is a tendency for condo construction to be highest in the states with the largest populations, but there are exceptions to this general rule. There are more people in California than in Florida, for example, but the ACS data indicate over twice as many condos being built in Florida.
In terms of the inventory of new condos on the market, Tennessee is an outlier at the high end with a 35 months’ supply. The other three states with over 15 months’ supply of new condos are located in the Midwest, around the great lakes. Although the ACS is based on a relatively large sample of 3 million housing units per year, it is still subject to sampling error, and the construction of a new multifamily condominium is a comparatively rare event. This caveat should be kept in mind when viewing an estimated inventory of zero unsold condos—something reported for seven states in the table, even though the table excludes states with a very small amount of new condo activity. Even without sampling error, the 610 unsold condos reported for Tennessee represents a small number of housing units for an area as large as an entire state, and they could be the result of a very few, possibly atypical, projects.
The ACS condo data by metro area are shown in Table 2. Most of the metro areas shown in the table are Metropolitan Statistical Areas (MSAs), which are aggregations of counties defined by the US Office of Management and Budget (OMB). They are based on local commuting patterns and are therefore designed to capture local labor markets. In some cases, OMB breaks large MSAs into Metropolitan Divisions. The San Francisco-Oakland-Fremont MSA in California, for example, is split into a San Francisco-San Mateo-Redwood City Metropolitan Division and an Oakland-Fremont-Heyward Metropolitan Division. In this case, tabulations for the individual Division are shown. Metro areas tend to have smaller populations than states (although there are exceptions, as 112 of the 377 MSAs and Metropolitan Divisions in the 2006 ACS have larger populations than Wyoming), so the caveats about rare events, sampling error, and timing issues are even more pertinent to the ACS metro data.
Again, the new condo activity shows an unsurprising pattern of concentration. About 77% of the condos built in 2005 or 2006, 75% of the ones that were sold, and 83% of the ones that remain unsold, are located in only 41 metros, and it is these that are shown in the table.
The table shows that over 5,000 five-plus condos were built in the Chicago Metropolitan Division in 2005 and 2006—more than in any other metro area by a wide margin, and more than in any state except for Florida and (obviously) Illinois. The new condo inventories range from zero in nine of the metros to a months’ supply of nearly 54 in Detroit, a metro area known to be experiencing economic hardship with an unemployment rate of over 8%.
In cases where the months’ supply is very high, keep in mind the facts that the total production of condos is a relatively small number in many metro areas, and that the ACS only tells us that the new condos were built sometime in 2005 or 2006. The construction could have taken place any time within that period. In some cases, it’s possible that the unsold new condos captured in the ACS are based on very few projects that may have been on the market only a short time when captured by the survey. In metro areas where the months’ supply number is very high (or very low), it would be desirable to track the inventory over several years in order to help judge if the inventory measure for a particular metro area in a specific year is an anomaly or not.
However, this not possible at present, as the inventory numbers are being generated from a relatively new data source. The ACS has been fully implemented and producing data for all geographic entities with a population of at least 65,000 only since 2005. The housing industry is still early in the process of learning how to mine this data source for relevant information. If there is sufficient interest in this topic, it may eventually be possible to track the unsold inventory of new condos by state and metro area over time and report the year-to-year changes.
 Although a significant number of single family condominiums exist in the U.S., this article focuses on multifamily condos.
 Michael Carliner, “The Neglected Condominium,” Housing Economics, November 2000.
 As of this writing, the annual numbers for 2007 are not yet available.
 According to the 2005 American Housing Survey, the average size of an owner-occupied multifamily unit was only about 1,400 square feet, compared to nearly 2,300 for an owner-occupied, single family detached home.
 The ACS data do not identify the vintage of structures very precisely if they were built before 2005.