J R E R � V o l . 2 9 � N o . 1 – 2 0 0 7

T h e S e c u l a r a n d C y c l i c B e h a v i o r o f ‘ ‘ T r u e ’ ’ C o n s t r u c t i o n C o s t s

A u t h o r s Wi l l i a m C . W h e a t o n a n d Wi l l i a m E . S i m o n t o n

A b s t r a c t Current construction cost indices typically are derived by applying national weights to local costs for materials and labor. In this study, construction cost indices are developed that are based on actual contractor tenders for projects. As such, they incorporate full variation in factor proportions, as well as factor costs, contractor overhead, and profit. Cost indices are produced for two product types, office and multi-family residential, in six different MSAs using F.W. Dodge project cost data from 1967 through the first half of 2004. Standard ‘‘hedonic’’ analysis is applied to control for variation in project scale and features to extract the true time trends in costs for each market. The findings indicated that real construction costs generally have fallen slightly over the last 35 years. In addition, no correlation is found between costs and building activity. Causal (IV) analysis implies that the construction industry is elastically supplied to local real estate markets, with any ‘‘excess’’ profits going to land and developer entrepreneurship. This is consistent with the traditional ‘‘urban land economics’’ literature.

This paper develops construction cost indices based upon the actual construction tenders for a very large number of building projects over the 1967–2004 period. Using these indices, two questions are examined. First, there has been much discussion over why the appreciation of most commercial real estate over the 35 year period has been actually slightly less than inflation (Geltner and Miller, 2004). The current paper finds that construction costs have behaved similarly during this time—generally increasing slightly less than overall economic inflation. Real estate markets are widely thought to be mean-reverting around replacement costs and thus to the extent that replacement costs are based mainly on the construction (as opposed to land) component—this paper provides an answer. The second question is to explore if there is a tendency for construction costs to rise cyclically during periods of major building activity. If this is the case, it could suggest that the construction industry is less than elastically supplied to local real estate markets. Whether this is due to inelasticity in the supply of materials and labor, or the whether contractors and construction firms are able to extract short run

 

 

2 � W h e a t o n a n d S i m o n t o n

profits, cannot be distinguished. In either case, however, inelasticity would cast some doubt on the time-honored axiom of modern urban economics that any excess returns go to land or entrepreneurship (Clemhout, 1981; and Blackley, 1999).

The analysis is made possible by using the records from the FW Dodge company covering 67,000 building projects in six chosen metropolitan areas, for two major types of buildings: low- to mid-rise office building and garden-type apartments. This is the first time this data has been used to assess cost trends, although Coleman and Gentile (2001) present interesting results about the construction process. With this database, hedonic cost equations are estimated for each market and property type that include several variables related to project scale, building features, and density of development. Yearly fixed effects are used to extract the time tends in these costs.

This approach is far broader than that used by several commercial suppliers of construction cost ‘‘indices’’ (ERN, 2004; and RS Means, 2004). These indices generally apply fixed weights to local variation in the costs of the major factors of production: labor and materials. In theory, this approach suffers from two omissions: first, factor proportions may and in fact should vary both cross section and over time with differences in factor costs and secondly, construction profit or overhead may vary cyclically to absorb some short run development profits.

The results of the analysis are quite pronounced. First, the yearly movement in construction costs is quite small—rarely more than 5%–10% when adjusting for inflation. In fact, the indices are quite smooth. In this regard they look quite similar over the 35 year study period to the factor cost indices of RS Means and ERN.

Secondly, over the longer run, there is a gradual trend wherein constant dollar costs have declined for both studied property types, in all but one of the six markets. This too is similar to the long run trends in the indices of the cost vendors. This may help to explain the observation that constant dollar office and apartment rents show no upward trend over a similar 35 year period (DiPasquale and Wheaton, 1992; and Wheaton and Torto, 1994).

Finally and perhaps most importantly, virtually no evidence is found that the construction cost indices vary with the level of building activity. A wide range of correlation tests is performed with different lags and there simply is no significant pattern. However, this might be the result of the complicated simultaneity issue suggested by Sommerville (1999). If building activity is negatively related to costs (construction demand) but costs are positively related to building activity (construction supply), then reduced form correlations could yield little. This idea is explored with an IV approach using local job growth, national interest rates, and inflation as demand instruments for building activity. The results are little different and this as a strong indication that the construction industry is supplied with almost perfect elasticity to local commercial real estate markets.

The paper is organized as follows. First, there is a discussion of the data and the selection of cities and building types. Second, the hedonic equation estimates are

 

 

S e c u l a r a n d C y c l i c B e h a v i o r � 3

J R E R � V o l . 2 9 � N o . 1 – 2 0 0 7

provided along with the time trends in the data. These are then compared with other construction cost indices. Third, there is a comparison of the movement in costs to the yearly level of building activity data for each of the chosen property types. Finally, the paper closes with concluding remarks.

� F. W . D o d g e C o n s t r u c t i o n D a t a The data used in this analysis was made available by F.W. Dodge, a subsidiary of McGraw-Hill Construction. This proprietary data was used with their permission and is derived from the company’s business of providing a ‘‘matching process’’ between developers and contractors. The former post requests for bids (from architectural plans to final interior finishing), the latter respond, and after the project is complete, F.W. Dodge assembles all the contracts into a ‘‘project file.’’ From this, total project costs are derived. It should be mentioned that these total ‘‘bid’’ costs are then later followed up by a survey of ‘‘all in’’ costs that presumably covers cost adjustments or over runs.

For this study, six MSAs were selected: Chicago, IL; Phoenix, AZ; Denver, CO; Washington, DC; San Diego, CA; and Dallas, TX. Chicago and Washington, DC are traditionally considered more mature cities, while the others have exhibited consistently higher growth over the last 35 years. The Dodge company classifies projects into 25 categories. This paper examines two that were both very prevalent in the data, and in addition represent fairly uniform types of structures. This was low-rise office buildings (2–4 stories) and garden-type apartment properties. The data for these two types included total construction costs for over 80,000 projects. In addition to this project ‘‘filtering,’’ limited descriptive information regarding frame or construction technology, number of stories, and floor area was also included. For apartments, the number of units was also known. The data for the Chicago, Dallas, and Washington, DC MSAs begins in 1967. Data for Phoenix and San Diego was available back to 1968, and data for the Denver MSA was only available back to 1969.

Of the 80,836 data points collected, approximately 4% were incomplete. Projects that lacked area, cost, or story information were removed from the sample. Data from projects under 2,000 square feet for multi-family and 10,000 square feet for office buildings or over .5 million square feet were also eliminated from the data set. The intent of the study is to develop a construction cost index representative of typical projects, so outliers at the extreme ends of the size spectrum were removed from the sample. The data was then segregated at the MSA and property level, providing twelve different sets of data. After this screening, the office sample size was reduced to 18,469 observations ranging between 1,937 and 3,857 observations per MSA. There were 42,340 apartment observations ranging between 3,777 and 12,259 observations per MSA (Exhibit 1).

It certainly is to be expected that construction costs per square foot will vary by project size, and in the case of apartments, project density (units / square foot).

 

 

4 � W h e a t o n a n d S i m o n t o n

E x h i b i t 1 � Number of Observations by Product Type and Location

Chicago Dallas Denver Phoenix San Diego Washington, D.C. Total

Apartment 12,259 3,777 3,356 5,943 9,301 7,704 42,340

Office 3,857 3,154 1,937 3,344 2,383 3,794 18,469

Number of stories is also often associated with greater foundation and structure costs. The frame type of the property was also included as potentially impacting costs. The F.W. Dodge data included twelve different frame classifications.

� C o d e F r a m e T y p e 0 Alterations, non-building, etc. without framing A Load or Wall Bearing (no further description) B Steel C Wood D Concrete E Pre-Fabricated or Pre-Engineered F Other Described Framing Types G Unknown Framing Type (no description) H Steel and Concrete I Load or Wall Bearing and Steel J Load or Wall Bearing and Wood K Load or Wall Bearing and Concrete

For this study, records with the ‘0’ code were removed because they are alterations, and not new construction. Records that had unknown or hybrid frame types, classified as codes E–K, also were combined into one single ‘‘other’’ category. This was done because most of the structures fall into one of the first four classifications—load bearing, steel, wood, or concrete. Exhibit 2 displays the distribution of frame types by location and product type.

The construction cost data does not account for soft costs such as time delays, development, or legal fees. The cost data does cover engineering and architectural fees as those are part of the Dodge matching service. Of course it must be remembered that project specifications and building standards have changed over the years—in many ways not captured by the data. Properties today have improved HVAC systems, Internet wiring, and must follow generally more stringent building codes. All of these omissions generally mean that more recently built projects are ‘‘better’’ in many unmeasured ways. As such, the indices will be biased upwards over time. To build a property in 2003 that is actually comparable to one in 1967

 

 

S e c u l a r a n d C y c l i c B e h a v i o r � 5

J R E R � V o l . 2 9 � N o . 1 – 2 0 0 7

E x h i b i t 2 � Frame Type by Location and Product Type

MSA Product Type