Rental Housing Stock: Addition...and Subtraction
Every month, the housing starts numbers are the subject of press releases, articles in the business sections of newspapers, and roundtable discussions on cable news programs, where they’re used as a shorthand measure of growth in the country’s housing stock. But starts represent only one way the housing stock changes over time, and starts themselves depend on other changes.

Houses That Go Away

There are basically two reasons new apartments are built: 1.To accommodate additional renter households, and 2. To replace units that are removed from the housing stock. Previous articles have discussed the impact of new renter households. In particular, we saw how underestimated immigration helped explain why multifamily production exceeded expectations in the late 1990s, and why the long-term outlook for multifamily production depends critically on continued immigration as a source of new renter households.

Removals are another piece of the rental demand puzzle Market Outlook hasn’t looked at before.  Removal rates may be especially important if they vary over time. One speculation, for example, is that the well-documented economic prosperity of the late 1990s enabled renters, on average, to afford newer and better units — and that in turn enabled the economy as a whole to demolish and replace older units at a faster rate.

Speculation like that has been difficult to document, however. One article that did investigate removals was published by NAHB in Housing Economics in 2003.¹ That article compared the housing stock in the 1990 and 2000 Censuses and computed a net removal rate as a residual. The resulting estimate was about 0.2% of the existing stock being removed each year. Although that may seem a small number (and, in fact, it implies that the median age of the housing stock is increasing), it’s large enough to mean that one in every six new units built during the decade was built to replace an existing unit rather than to accommodate a new household.

More recently, two papers funded by HUD’s Office of Policy Development and Research have sought to trace what happens to specific housing units between installments of the American Housing Survey (AHS). The AHS, conducted by HUD and the U.S. Census Bureau, visits the same housing units (plus new ones that have been added to the stock since the last survey) every two years.

Getting a Handle on Demolition Rates

The first paper, “The Destruction of Housing Capital: A Preliminary Exploration Into Demolitions and Disasters,” was released in late November and authored by Gregory J. Watson and Frederick J. Eggers. Watson works at ICF Consulting, where he maintains and updates the AHS data base under a HUD contract. Eggers, currently of Econometrica, Inc., is former Deputy Assistant Secretary for Economic Affairs at HUD.

The main contribution of the paper is a statistical model that relates probability of demolition to unit characteristics. We can use the model to estimate the probability that a rented multifamily unit will be demolished sometime during a two-year period. The paper also generated a separate model for each AHS from 1987 through 2001, so we can use those results to analyze trends over time. 

An important industry question is why production of new multifamily units continually exceeded predictions during the late 1990s. We know part of the reason: Census-based projections of the number of renters were too low, primarily due to underestimated immigration. But economic growth was also very strong and steady during that period — at least 3.7% each year from 1996-2000 — and that also could be a contributing factor. Favorable economic conditions may enable people to afford better quality housing units, enabling older, lower-quality units to be retired at a faster rate. But since the Census Bureau quit tracking demolition permits there hasn’t been a direct way to investigate that intuitively plausible theory.

The Watson-Eggers model provides a way. We used their model and results to generate a predicted demolition rate for a standard multifamily rented unit from 1987 through 2001. This standard unit has four rooms, is built between 1940 and 1960, is rated as by its occupant as a “5” in terms of overall quality (on a scale of 1 to 10). It has no structural inadequacies according to the conventional AHS definition,  and is located in a suburban neighborhood without any of the following characteristics: abandoned structures, buildings that have bars on the windows, schools, or shopping centers.

The results are shown in Figures 1 through 5. In each graph, the vertical axis is the probability that a multifamily unit occupied by a renter two years prior to the current survey date has been demolished. One caveat: because estimated coefficients in the model sometimes differ substantially from year to year, you could get a different pattern of demolition rates if you began with a standard unit that had different characteristics. 

Partly for that reason, the figures show separate results for each of the four principal Census regions. In the Northeast, the estimated demolition rate started at 0.37% in 1987, dropped precipitously to less than 0.10%, and then increased steadily — rising to 0.65% by 2001 (Figure 1).

Source: simulation of logit regression results published by Watson and Eggers under a contract issued by the HUD Office of Policy Development and Research

In the Midwest, the demolition rate started at 0.35%. It declined (but more gradually than in the Northeast, never dipping much below 0.25%), then rose dramatically in the late 1990’s—to a spike of over 1.2% in 1999 — before falling back below 0.6% in 2001 (Figure 2).

Source: simulation of logit regression results published by Watson and Eggers under a contract issued by the HUD Office of Policy Development and Research

In the South, the demolition rate started at 0.62% in 1987, then declined through the early 1990s until it reached a low of 0.16% in 1995. After that, the rate shot up to a plateau, staying in the 0.90%-0.95% range from 1997 through 2001 (Figure 3).

Source: simulation of logit regression results published by Watson and Eggers under contract from the HUD Office of Policy Development and Research

The West started with a demolition rate of 0.79% in 1987. After that, the rate  declined to a low of 0.18% in 1995, increased to 0.68% in 1999, then fell back to 0.54% in 2001 (Figure 4).

Source: simulation of logit regression results published by Watson and Eggers under a contract issued by the HUD Office of Policy Development and Research

In all four regions, demolition rates declined immediately after 1987. A relatively high peak in 1987 (capturing units that were rented in 1985 but demolished by 1987) could be partly supply-driven, as a very large number of multifamily units were still coming on line as a result of tax policies adopted in the early 1980s.   

Also, demolition rates in all four regions were higher in 1997-2001 than they were in 1989-1995. Although that’s consistent with the theory that a strong overall economy in 1996-2000 increased demolitions, it’s important to note that HUD made  substantial changes to the AHS between the 1995 and 1997 surveys. Figure 5 shows a weighted average of results for all four regions (the weights being based on the distribution of units across regions in 2001).

Source: simulation of logit regression results published by Watson and Eggers under a contract issued by the HUD Office of Policy Development and Research

Conversions and Other Changes

There are other ways the housing stock can evolve, of course, than by demolition and new construction. Housing units can be lost through conversion to non-residential use, or by merging two units together. They can be gained through similar processes. Another significant category in the AHS is “temporary losses.”  These are units that have become unoccupiable but potentially could still be reclaimed as part of the housing stock. Particular units may fall in and out of that category as well. In addition, if we are looking at the rental housing stock, units may be lost or added by conversion from renter to owner-occupied or vice versa.

The second HUD paper, “Rental Market Dynamics: Is Affordable Housing for the Poor and Endangered Species?” written by the same authors and released a month after the first, looks at these and as well as other issues.

Although this paper looks at a broader range of issues, its scope is narrower in that it uses AHS data for only six metropolitan areas (New York, Los Angeles, Chicago, Philadelphia, Detroit, and Northern New Jersey) and two points in time (1995 and 1999).

In addition, its primary focus is on movements within the rental stock (from one rent tier to another) but it considers movements in and out of the rental stock as well. Aggregate results for the six metros are converted into rates and shown in Figures 6 (for removals from the rental stock) and 7 (additions).

The most obvious feature in the figures is the way that tenure (conversions from renter to owner-occupied and vice versa) account for a very large share of rental tock changes.  Not only are some of the other changes comparatively small, in several cases a movement into the rental stock is almost perfectly offset by corresponding movement out of it. For example the number of rented units lost through conversion to nonresidential uses is nearly the same as the number gained through conversion from nonresidential.

Renter-to-owner and owner-to-renter conversions also come close to offsetting each other, although the renter-to-owner rate is a bit higher (8.5% vs. 7.4%), as you would expect during a period of rising homeownership rates.  If anything, we might wonder why the difference isn’t greater.

The share of the rental stock permanently destroyed (1.10%) is roughly consistent with the demolition rates shown in Figures 1 through 5, given that the former is based on a four-year period and the latter on a two-year period. The rate at which units were added through new construction (1.41%) seems suspiciously low, however. Nationally, multifamily starts were roughly equal to 1% of the existing stock in each of the four years during the study period. Moreover, the 1.41% is based on an estimate of 66,400 rental units being built in the six metro areas between late 1995 and late 1999. We would expect more than that based on multifamily starts in the New York and Chicago metro areas alone.
 
That illustrates one of the difficulties involved in this type of research. Although new construction in the AHS is ultimately derived from the same data source used to generate housing starts (the Survey of Construction), the new units are not weighted so that AHS totals equal starts (or completion) totals in a given year — one of many severe challenges involved in trying to trace what happens to the housing stock one from year to the next. 

Info on Past Demolitions Help Predict the Future

Together, the two recently published HUD studies provide significant new information about the stock of rental housing in the U.S. and how it evolves.⁴ Although not completely conclusive, the information suggests that demolition rates increased somewhat between 1996 and 2000, a period when economic growth in the U.S. was close to (or over) 4% per year. Thus, replacement demand would have been higher than usual, contributing to the strong rates of multifamily production (well over 300,000 units a year) seen during that period.

This has implications for the near-term future, as NAHB is forecasting more than 4% growth in real GDP for 2004, and close to 4% for 2005. If that materializes, increased removals could help prop up multifamily production, which seems almost certain to experience some kind of slowdown in response to the current very high vacancy rates. NAHB’s 2004-2005 housing forecast calls for only modest reductions (roughly 5 percent each year) in multifamily starts.

¹Drew A. Mitchum, “Housing Removal Rates” in Housing Economics, February 2003, published by the NAHB Economics Group.

²NAHB research indicates that at least one reasonable alternative to the conventional definition exists, and that alternative flags a much greater share of the existing housing stock as inadequate, and in need of repair or replacement.  See “The Need for New Housing Revisited” in Multifamily Market Outlook, March 2002.

³The number of units changing back and forth between “rented” and “vacant” status is even larger than the number changing between “owned” and “rented.” Vacant units are excluded from the figures due to confusion over whether they are purely “vacant-for-rent” units or not. If so, they should still be counted as belonging to the rental stock.

⁴Both are available free of charge from the Huduser web site at http://www.huduser.org/datasets/ahs/ahsReports.html#1 and http://www.huduser.org/datasets/ahs/ahsReports.html#2 [ return to top ]