Printer Friendly

A detailed look at America's real estate wealth.

As many property management firms evolve toward full-lie asset management companies, new issues become important, creating needs for new types of information. Particularly important for both the management of a client's overall portfolio and for positioning the asset management firm itself is a knowledge of the extent, location, and quality of America's real estate wealth.

The existing literature in this area is well summarized in a 1990 article by Mike Miles appearing in Real Estate Review. (A table from the article is reproduced here as Figure 1.) It concludes that the best estimate for the total value of U.S. commercial real estate in 1988 was $2 trillion. However, the existing literature provides very little breakdown regarding the type, quality, or location of this property.
 Value in Value per
Estimate Source $ Billions Square Foot(1)
Bureau of Economic
Analysis, total
business property $4,692 $176
Roulac 3,492 131
Salomon Brothers 1,295 49
Census of
Governments(2) 815 31
(1)Assuming 26.66 billion square feet of commercial property

Yet, it is the knowledge of just these variables that is essential if a true picture of the country's investment-grade real estate is to be drawn. Current research, by these authors and others, is seeking to extend the body of knowledge with better estimates of the principal components of the nation's real estate wealth.

The specific research techniques described in the first sections of this article may only interest the researcher. But the conclusions as to the location, property type, and value of U.S. real estate are vital to every asset manager.


Our methodology for this research proceeds from the foundation established in the existing literature and then relies on property tax records as the underlying data source. Property tax records are sampled and a regression analysis is the used to extend from a manageable sample to an estimate of national real estate wealth.

In this work we approach the estimation at two levels. We first apply the methodology to state-level property tax records to obtain gross estimates of the nation's real estate wealth, distinguishing only between residential and combined commercial and industrial. We then use county property tax samples to estimate more detailed components of real estate, e.g., retail, office, warehouses, and so forth.

The primary property tax data used in this study is collected by REDI Real Estate Information Service, a company which reproduces and sells property tax records from around the country in computer form. The second primary data source is the U.S. government, which provides various demographic and economic variables through the Census Bureau. As a check of our results, we have used estimated data from F.W. Dodge and Real Estate Information Service (REIS).

Econometric estimates

The total value of different types of real estate per person in each state (and later in each county) is used as the dependent variable. The independent (or explanatory) variables are a series of demographic and economic factors which should logically explain differences among various geographic regions with regard to the value of the different types of real estate.

Clearly, our sample will only include a subset of the total number of geographic regions in the country. However, regression analysis is used to estimate the weights that should be given to the demographic factors in determining the aggregate real estate values. These weights can then be used to project value for nonsample regions.

In other words, if the value of the office stock in a particular set of counties is a function of three demographic variables and we know the same demographic variables for the other counties in the nation, we can use the regression results from the known sample of counties to determine the real estate value by property type in the rest of the counties in the U.S. We then sum over the appropriate counties to determine value by property type for the larger Metropolitan Statistical Areas (MSAs) in the nation.

We begin with estimates from state-level property tax records. These records are less detailed than county records, and we are forced to rely on assumptions of other researchers regarding consolidations. Further, certain states are obviously not appropriate for this type of methodology. (For example, California with its Proposition 13 cannot be used as a sample state.)

The same general methodology is applied to a sample of counties where there is far more detail regarding property types. The use of county data, however, has several drawbacks. First, given the number of counties in the nations, it is practically impossible to sample as high a portion of the total population as is the case when state data are used. Further, most real estate markets extend across several counties, indicating that the county data must be used with caution. Finally, different counties use different land-use codes, requiring us to construct a consistent land-use code scheme.

By comparing the state-and county-level estimates, we obtain a more accurate view of the components of the total stock, and we can assess a degree of reliability for our results.

State-level data

The dependent variables for the state-level analyses were obtained by telephone survey from the various states. During the initial stages of the research project, phone calls were made to each state's Department of Revenue, Tax Equalization Board, or other state-level department responsible for aggregating county-level data to state totals. A questionnaire was used to determine the type of data collected by each state.

In general, most states were very cooperative with our data collection efforts. Unfortunately, very few states collected square footage data.

Approximately half of the states collected data that was broken out by five basic property types: commercial, industrial, agricultural, residential, and vacant land. However, several states did not break out by even these broad property types and could only provide total assessed values for all real property.

We tried to obtain two observations for each state: 1989 and 1985. A few states did not have data available for 1989; therefore, data for 1988 was used for the study. Likewise, some states did not have 1985 numbers so data for 1986 was used for the research. Whenever the observed year varied for the dependent variable, the observations on the independent variables were taken from that same year, i.e., either 1988 or 1986. Once data was received from the states, the data was entered into a spreadsheet.

County-level data

The county-level dependent variables were obtained from REDI. The sample counties were chosen based on the detail in their land use codes, the quality of the REDI assessor's data for the market, and the possibility to get a balance across MSAs from a size and age perspective. As metropolitan statistical areas were built up from county data, the eight areas (Figure 2) were provided county by county.


MSAs and Counties Included in Sample

Miami--Fort Lauderdale CMSA:

Dade, Broward Phoenix MSA:

Maricopa Philadelphia PMSA:

Chester, Bucks, Philadelphia, Montgomery,

Burlington, Camden, Gloucester

(Missing:(*) Delaware) Dayton--Springfield MSA:

Greene, Montgomery, Clark

(Missing: Miami) Cleveland PMSA:

Cuyahoga, Lake

(Missing: Medina Geauga) Denver PMSA:k

Arapahoe, Denver, Douglas, Jefferson, Adams Dallas--Forth Worth CMSA:

Dallas, Tarrant, Collin, Denton

(Missing: Ellis, Kaufman, Rockwall,

Johnson, Parker) Seattle PMSA:

King, Pierce, Snohomish (*)REDI does not collect property tax information on these smaller counties

Significant transformation were necessary to put all information in a standard land-use-code classification format. This is particularly true for data collected at the county level. The typical property types by which states provide data are commercial, industrial, agricultural, and residential. In contrast, the REDI county data usually has very detailed breakdowns, but the various categories are inconsistent across counties.

Therefore, it was necessary to devise a consistent property-type format and to reclassify the extensive REDI data for all counties to that format. This format is shown in Figure 3.


Aggregation of Major Land Uses

Symbol Description Residential Total:
R-L Vacant and zoned residential
R-S Single family
R-M Mobile homes
R-C Condominium, town houses, cooperatives,

and attached (duplex, four-plex, etc.)
R-A Apartments--all over four units
R-H Hotel
R-R Retirement housing
R-U Undesignated, Public Housing Authority

Commercial Total:
C-L Vacant zoned commercial
C-RM Regional mall
C-R Other retail--gas stations, restaurants,

parking garages
C-O Office (financial institutions)
C-U Undesignated

Industrial Total:
I-L Vacant land zoned industrial
I-W Warehouse/industrial--bulk warehouse

through high-grade industrial park

I-M Manufacturing/special purpose,

heavy industry

I-U Undesignated

Other Property Types:

A Agricultural--nurseries, farms, cropland

pasture, ranches, mines, extractive forest

V Vacant--private lakes, flood plain,

unzoned, unused

G Government--local, state and federal O/T.

schools and public parks where separated,

post offices, military prisons

M Municipal utility--railroads, airports,

public roads (where designated), landfills
H Hospitals, veterans, and retirement housing
I Indian
S Schools
C Cemeteries and mortuaries
R Religious and charitable
RPR Resort, parks, recreation--race tracks,

golf, bowling, museums, library

O Other

Note: Mixed use is listed with first use except office/warehouse, which is with warehouse

As can be imagined, a faulty, rearrangement of the data can lead to serious biases in the results. Great caution was therefore exercised in order to achieve results which would be as reliable as possible.

When a type of property could not be classified in any of the categories or when an informed allocation was not possible, that property was allocated, depending on its nature, to one of the following categories: other, residential-undesignated, commercial-undesignated, and industrial-undesignated. The undesignated categories were then reallocated to the appropriate categories, prorated on the relative value of each type of property in all counties.

Once all of the different county data was refined to the standard format, it was then necessary to adjust all of the assessed values to supposed market values. Across the states and across the counties, local official assess at different percentages of market values, so it was necessary to contact every county in the sample, find out the percentage of market value used, and then adjust all the assessed values to get an estimate of the true market value.

Regression equations

Working at both levels of aggregation, state and county, we estimated the relationship between the value of real estate per capita (dependent variable) and a series of demographic and economic factors (independent variables) which should logically explain differences among various geographic regions with regard to the value of commercial real estate.

What happens in Figure 4 are the final best equations, i.e., those equations with the highest [R.sup.2] and all logical signs on the independent variables.


Regression Equations

Commercial and Industrial Property Combined: (COM + IND)/POP =

31.9 + 0.000966 x CAPINC - 0.0517 x POPM - 0.258 x OWND +

0.631 x UNEMP + 0.252 x POP65 - 0.22 x MNFP 0.893 x TRDP

([R.sup.2] = 0.85)

Fractions of variability(*)
 Partial [R.sup.2] 0.42 0.17 0.12 0.06 0.03 0.03 0.02

Commercial Property: COM/POP =

27.79 + 0.000369 x CAPINC - 0.133 x OWND - 0.183

x MNFP - 0.677 x TRDP

([R.sup.2] = 0.82)

Fractions of variability:
 Partial [R.sup.2] 0.38 0.20 0.15 0.09

Industrial Property: IND/POP =

3.313 + 0.000471 x CAPINC - 0.043 x POPM + 0.465 x UNEMP - 0.091

x MNFP - 0.281 x TRDP + 0.00165 x POPD

([R.sup.2] = 0.74)

Fractions of variability
 Partial [R.sup.2] 0.18 0.17 0.15 0.09 0.08 0.07

Residential Property. RES/POP =

-14.47 + 0.00227 x CAPINC + 0.00664 x POPD - 0.4913 x FDSTMP

([R.sup.2] = 0.75)

Fractions of variability:
 Partial [R.sup.2] 0.71 0.02 0.02

(*)Fraction of the dependent variable's variability explained by each independent variable

State-level regression equations

The data was first analyzed with all available sample points (47 observations, two-period data for twenty-three states and one-period data for New Hampshire), and regression estimates obtained Alabama, Arizona, and New Hampshire were found to be outliers and were dropped from the sample. Regression estimates were then generated from the remaining 42 sample points. An explanation of the independent variables used is given in Figure 5.


State-Level Independent Variables
Variable Brief Discussions
CAPINC Per capita income
OWND Percent of owner-occupied housing units
UNEMP Unemployment rate
POPM Percent of population residing in
 metropolitan areas
POPD Population density
FDSTMP Food stamp recipients as a
 percent of population
POP65 Percent of population aged
 65 years or more
MNFP Percent of nonagricultural
 employees in trade

We do not report t-statistics for the coefficient estimates. All variables that were retained were significant at least at the 15-percent level. The choice of this rather unusual level of significance was motivated by the fact that our equations are used primarily for predictive purposes and variables with t-statistics whose absolute values are greater than one add to predictive power based on the adjusted [R.sup.2] criterion.

A t-statistic of one would actually imply an even lower significance level, but no adjustment was made to the standard errors to correct for the fact that we have multiple observations on some states which would tend to deflate coefficient standard errors and increase t-statistics. Therefore, we chose 15 percent as a fairly conservative significance level that indicates increased predictive power.

Per capita income, CAPINC, is found to be positively related to property value per capita in all four regressions. Looking in more detail first at the combined nonresidential equation, the next most important variable is the percent in manufacturing (MNFP), which should have a negative impact on the larger commercial component. The percentage in trade (TRDP) is negative as would be expected given the high negative correlation of the percentage in trade with the percentage in service, which is on the average a higher income component.

The percentage in metro areas (POPM) is also negative, which is consistent with a higher percentage of welfare recipients in cities. With the welfare factor accounted for, unemployment (UNEMP) should be positive as it now proxies for available worm force. Percentage of housing units owned is negative due to the well-documented increase in rental units in more densely populated areas. Finally, the elderly factor is positive consistent with the heavy shift in national resources to this age segment over the last decade.

The residential equation is simpler with income per capita (CAPINC) and population density (POPD) both having the expected positive impact on value per person. The higher the percentage on food stamps, the lower the value of residential property.

The extension to a national estimate from the state equations is rather straightforward with state demographics multiplied by the equation coefficients to get the state estimates. We predict commercial and industrial both jointly and independently, then add the commercial and industrial estimates and average that number with the results of the joint estimation to derive the totals shown in Figure 6.

All independent variables are in the sample period except per capita income. As this is the only dollar denominated variable, it is adjusted (for each state) to a 1990 figure. Thus the total estimate of $2.66 trillion for commercial and industrial property shown in Figure 6 is for 1990. This figure is slightly higher than the Miles estimate for 1988, thus substantiating the methodology and setting up the more interesting county-level analysis.

County-level regression equations

The county-level methodology was identical. While the equations contained slightly different sets of independent variables, the signs were again as expected, and the percentage of variability explained by each independent variable was appropriate.

The regression model was specified using three functional forms: linear, log linear, and log. It is somewhat surprising that the linear model appears to be the most stable model across property types in our runs, even though one would expect that it would be more susceptible to problems with outliers. Our results clearly indicate that the outliers are still a problem, especially when we do out-of-sample predictions.

Figure 7 presents the final county estimates aggregated to the MSA level. The total for the 36 largest MSAs (those with a population of over 1 million) is appropriately smaller than the national total from the state regressions. Now, however, there are more details on property-type distinctions.

In looking at more property-type distinctions at great levels of disaggregation (county vs. state), the degree of confidence in each cell (e.g., retail in New Orleans) is proportionately lower when making out-of-sample predictions. However, the geographic and property-type distinctions shown in Figure 7 should prove useful to institutional investors seeking a benchmark market portfolio.

Since per capita income does not appear in all regressions, the figures in Figure 7 are not directly comparable to the 1990 national estimate in Figure 6. Because population is up since 1986 and the purchasing power of the dollar is down, the figures slightly understate the current situation. We have not made such adjustments as our intent is to show the relative size of all the major components of real estate wealth with the state regressions providing the current overall estimate value.

Part B of Figure 7 shows the breakdown of the two major types of commercial property by geographic region. While this breakdown is only for the larger MSAs, the percentages provide useful guidance to those seeking a diversified real estate investment portfolio.

Conclusions after early research

The initial $2 trillion estimate of the value of all commercial and industrial real estate (from prior work) has been refined to 42.7 trillion with a corresponding estimate of the value of residential real estate equal to $6.1 trillion.

Due to the nature of the data sources and the underlying methodology, these are inherently long-term estimates and do not reflect such cyclic factors as the current credit crunch. Hence, the reader may wish to reduce these figures by 10 to 15 percent to account for current capital market conditions.

Of more interest is the finding that this commercial and industrial total can be broken down as follows in the 36 largest MSAs--42 percent retail, 38 percent office buildings, and 20 percent industrial (42 percent distribution warehouses, and 58 percent special purpose/manufacturing).

Several important features of the gross estimates warrant mention: * While these residential numbers are from state property tax aggregations, they are confirmed by the various county-level specifications as the total of all residential real estate in the nation's 281 metropolitan areas is in the same range regardless of model specifications. * The research finds a higher percentage of retail properties than was previously estimated. * The percentage of the nation's commercial and industrial value in the largest metropolitan areas is on the order of 75 percent with almost all of the rest in the remaining 245 metropolitan areas. * Within a portfolio of the largest metropolitan areas, roughly two-thirds of the commercial and industrial properties are found in the traditional Northeast and Pacific Rim regions.

The sample metropolitan areas which underlie these estimates were tested with the Dodge and REIS databases. From the Dodge perspective, our value estimates, when combined with their stock estimates, suggest that retail properties are priced at approximately $77 per square foot, office building at $70 per square foot, warehouses at about 440 per square foot, and manufacturing properties at approximately $34 per square foot.

Looking to the REIS data, which deals only with the higher grade portion of the stock, we find a proportionately higher value given the lower stock estimate. Using the REIS stock estimates, retail properties are priced at about $200 per square foot, and office buildings at approximately $181 per square foot.

Continuing efforts

Given the value of accurate estimates of the components of real estate wealth to major investor groups as well as to all the professional organizations associated with the real estate industry, we are working to extend and refine this analysis. The research done for the Institute of Real Estate Management Foundation by Arthur Andersen & Company and Hoyt Advisory Services (see the preceding article) has certainly advanced the body of knowledge.

The National Council of Real Estate Investment Fiduciaries (NCREIF) an the Homer Hoyt Institute have jointly funded work to extend this research in several interesting ways. Institutional investors are especially concerned with the quality of some of the distinctions, particularly the distinction between investment-grade and noninvestment-grade real estate.

By sampling individual properties from key MSAs used here and then interviewing professionals in the market, the authors are obtaining a more accurate breakdown of each of the property types shown in this report by investment grade and noninvestment grade.

In addition, there are several biases inherent in the approach used above which must be reduced. While we have adjusted all assessed values to what the assessment officer would call true market value, there still may be differences among geographic areas as to how this estimate relates to most likely sales price. By using properties actually sold from the NCREIF database and comparing them to previous assessed values, the authors are obtaining a more rigorous adjustment ratio for each property type and area, which would then make the overall econometrics more accurate.

Finally, it would be interesting to compare the investment-grade properties held by NCREIF with the properties held by the Resolution Trust Corporation (RTC). This end of the spectrum is also an important component of the total real estate wealth. By sampling from eight metropolitan areas and comparing the sample to the RTC portfolio, one could obtain an idea of the percentage of the stock which is distressed. Mike Miles is Foundation Professor of Urban Development at the University of North Carolina at Chapel Hill. He received a bachelor's degree from Washington and Lee University, an M.B.A. degree from Stanford University, and a Ph.D . degree from the University of Texas at Austin with concentrations in real property, economics, and finance. After graduation, Mr. Miles worked for Peat, Marwick, Mitchell and Company; and subsequent, as vice president-controller of a large Dallas- and Atlanta-based real estate development company. Since 1976, he has been on the faculty at UNC with teaching and research interests in real property investment and development and business policy. He has also taught at the University of Texas, the University of Wisconsin, and the University of Hawaii. Mr. Miles is a certified public accountant, licensed real estate broker, and senior real property analyst. Along with Modern Real Estate, a textbook now entering its fourth edition, he is the author of over 80 journal articles, cases, and monographs on various aspects of the real estate industry. Robert Pittman is vice president of Hoyt Advisory Services (HAS), a real estate counseling and investment firm based in North Palm Beach, Florida, and Washington, D.C. HAS is a subsidiary of the Homer Hoyt Institute (HHI), an independent non-profit research and education-support organization dedicated to improving real estate decision making and land use. HHI's grants support real estate research at colleges across the nation. Mr. Pittman holds a Ph.D. degree in economics from Northwestern University. Pankaj Bhatnager is a Ph.D. candidate at the University of North Carolina at Chapel Hill. His dissertation is on the influence of options trading on the underlying stocks when leverage constraints exist in equity markets. Pankaj has a bachelor of technology degree from the India Institute of Technology, Kanpur, and an M.B.A. degree from Kent State University. Martin Hoesli is an assistant in finance at the University of Geneva (Switzerland) and currently is a visiting scholar at the Graduate School of Business of the University of North Carolina at Chapel Hill. His research interests lie in the analysis of real estate investments. He is completing his doctoral studies with a dissertation entitled "An Analysis of Swiss Real Estate in a Modern Portfolio Theory Framework," which examines the benefits for institutional investors of including real estate in their portfolios. David Guilkey is an econometrician with a microeconomics focus. His current theoretical work in econometrics involves development of estimation methods that can be used to analyze large survey data sets with limited dependent variables. His recent work involves the development of a series of estimation programs that can be used to estimate a structural model of contraceptive choice. These are being used to analyze method choice in Tunisia, Zimbabwe, and Colombia.
COPYRIGHT 1991 National Association of Realtors
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1991 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Miles, Mike; Pittman, Robert; Hoesli, Martin; Bhatnager, Pankaj; Guilkey, David
Publication:Journal of Property Management
Date:Jul 1, 1991
Previous Article:Who owns America and how should it be managed?
Next Article:Interviewing techniques.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters