Speed of adjustment in commercial real estate markets.

• Author: Eppli, Mark J.

1. Introduction

   The 1980s were less than favorable to the U.S. real estate industry. Vast amounts of overbuilding occurred in most areas and in most property types. Commercial vacancy rates increased dramatically and rents fell both in nominal and real terms. Untoward events (tax laws changes, deregulation of financial institutions, and a real estate depression in many areas of the country) virtually halted speculative development and caused a freefall in real estate prices. Real estate credit markets were crippled by the decade-long crisis in the S&L industry and the federal government's handling of that crisis, while large pools of government-owned real estate sold at distressed prices. Employment decreases in the finance, insurance, and real estate industry, and corporate consolidations in the late 1980s reduced the number of jobs for the industries who would otherwise be tenanting empty office buildings. Finally, environmental issues significantly increased the uncertainties in real estate development and added to overall construction delays. Today, we begin the 1990s with an environment of soft real estate markets in most domestic locations, and among real estate developers the widespread belief is that this is not just another down cycle.

   Given this recent past history, it seems like the perfect time to ask the question, How quickly do commercial real estate markets respond to changes in excess demand or supply? Traditionally, this problem has been known as the "speed of adjustment" problem, and it has been typically discussed in terms of stock-adjustment models of inventory investment [1; 6; 21]. Adjustment speeds are important for understanding future effects of real estate price changes and for interpreting recent events. The higher the adjustment speed, the greater the fraction of long-run adjustment experienced so far, the smaller the adjustments still to be expected, and the lower the implied long-run demand elasticity.

   Adjustment speeds also have important implications for macroeconomic policy. There is reason to believe, for example, that the slow economic growth in the U.S. during the four years of the Bush administration was attributable, in large part, to the massive overbuilding that occurred in commercial property markets during the 1980s and the relatively slow process by which unemployed resources have been absorbed into productive new endeavors [16].

   In what follows, we propose and test a stock-adjustment model of real estate investment. The model is tested with quarterly data on the stock of retail, office, and industrial real estate in the U.S. during the 1980:I-1990:II period. Then, the results are compared and contrasted to adjustment speeds prevailing in Canadian and U.K. real estate markets during approximately the same time period. Canadian and U.K. commercial property markets provide an interesting juxtaposition to U.S. commercial property markets given their strikingly different institutional relationships. Barkham and Geltner [2] note that, owing to greater homogeneity in England, U.K. commercial real estate markets are far more informetionally efficient than in the U.S. Byrne and Goldberg [8] argue that differences between U.S. and U.K. commercial real estate cycles may arise as a result of dissimilarities in market size. We attribute the greater stability in Canadian and U.K. commercial property markets to a variety of factors.

2. Market Adjustment Process for Commercial Real Estate

   There is a tendency to regard commercial real estate prices and rents as being relatively slow to adjust, and to assume that the quantity of space will adjust to equilibrate the market. Rosen and Smith [24], for example, have asserted that residential landlords react to fluctuations in demand primarily by building up or drawing down inventories of unlet space. Residential prices and rents are then affected but only after a lag.

   A similar price-adjustment process for commercial real estate has been suggested by Shilling, Sirmans, and Corgel [27], and Voith and Crone [29]. They find that price adjustments are the strongest when the gap between the normal, long-run vacancy rate is the largest, and weakest when vacancies exceed the normal rate.

   Wheaton [30], and Wheaton and Torto [32] have argued that commercial rents are determined by bargaining between landlords and tenants. Landlords set the minimum rent that they will accept based on an expected vacant time on the market. The expected vacant time decreases with the flow of prospective tenants and increases with the amount of competitive space available. Tenants set the maximum rent that they will pay based on the opportunities they have available to rent other space and on the competition they perceive from other prospective tenants. Actual rents are determined somewhere between the maximum that the tenant will pay and the minimum that the landlord will accept. Wheaton and Torto [32] further assume that, because landlords and tenants may be slow to perceive the flow of perspective tenants or the amount of competitive space on the market, the actual level of rents will adjust gradually to the desired rent level.

   There is room for argument whether quantity adjustments clear the commercial property markets. Studies by Barth et al. [3], and Kling and McCue [19] find that there is a considerable lag in the supply response of new U.S. office construction to changing demand conditions. There is also some evidence of a weighty lag in the supply response of both industrial and retail real estate to changes in demand [31; 20; 4].

   This work, however, can be criticized for various reasons. Perhaps the most damning criticism has to do with the failure of these studies to correct for autocorrelation in estimating adjustment speeds. It has long been realized that slow adjustment speeds can result when stock-adjustment models are estimated without correcting for autocorrelation [31; 20; 4].

3. A Stock-Adjustment Model of Commercial Real Estate

   To examine the speed of adjustment in commercial real estate markets, we adopt the following specification. We begin by writing the desired stock of retail, office, or industrial properties as

   [Mathematical Expression Omitted]

   or

   [Mathematical Expression Omitted]

   where

   [Mathematical Expression Omitted] = the desired stock of capital,

   [E.sub.i] = employment in industry i,

   [Phi] = average square feet per employee, and

   [[Epsilon].sub.it] = random error term.

   The subscript i in this case refers to the retail, office, or industrial sector.(1)

   Next, we write a stock-adjustment model for retail, office, and industrial properties as

   [Mathematical Expression Omitted]

   where

   [K.sub.it] = actual stock of capital, and

   [[Delta].sub.i] = the speed of adjustment parameter.

   Equation (3) specifies that the change in [K.sub.it] will respond only partially to the difference between the desired level of [Mathematical Expression Omitted] and the past value of [K.sub.it-1]. The rate at which the market responds is the adjustment coefficient [[Delta].sub.i]. If [[Delta].sub.i] equals one, then [Mathematical Expression Omitted] equals [K.sub.it] in each period and markets fully adjust each period (which would be indicative of a build-to-suit real estate market with no speculative construction). As [[Delta].sub.i] becomes closer to zero, however, the longer it takes for real estate markets to equilibrate. This framework is developed in Blinder [6], and Maccini and Rossana [21], and many other papers.

   Substituting equation (2) for [Mathematical Expression Omitted] in (3) yields

   [K.sub.it] = [[Delta].sub.i][[Beta].sub.0i] + [[Delta].sub.i][[Beta].sub.1i][E.sub.it] + (1 - [[Delta].sub.i]) [K.sub.it-1] + [[Delta].sub.i][[Epsilon].sub.it] (4)

   or

   [Mathematical Expression Omitted]

   where [Mathematical Expression Omitted], [Mathematical Expression Omitted], and [[Xi].sub.it] = [[Delta].sub.i][[Epsilon].sub.it].

   Finally, and most important, it might be thought that the error term [[Epsilon].sub.it] follows an AR(1) process. Hence, to estimate (5) we use the following two-step procedure. First, we estimate

   [K.sub.it] = [[Delta].sub.i](1 - [[Rho].sub.i])[[Beta].sub.0i] + [[Delta].sub.i][[Beta].sub.1i][E.sub.it] - [[Rho].sub.i][[Delta].sub.i][[Beta].sub.1i][E.sub.it-1] + ([[Rho].sub.i] - [[Delta].sub.i] + 1)[K.sub.it-1] - [[Rho].sub.i](1 - [[Delta].sub.i])[K.sub.it-1] + [e.sub.it] (6)

   or

   [K.sub.it] = [[Mu].sub.0i] + [[Mu].sub.1i][E.sub.it] + [[Mu].sub.2i][E.sub.it-1] +...