|
...reduces problems caused the infrequent trading of individual commercial properties. However, for those office properties that are located outside the CBD and also for those less frequently transacted, the power of the model in capturing these particular office buildings' price dynamics is limited. The significant differences of the office prices across the various office buildings and submarkets show that the model can capture the variation in office prices and track the timing of capital gains and losses that investors may accrue on spatially distributed office properties more accurately than hedonic or weighted least squares estimates.
**********
Real estate price indexes are used in valuing property, understanding real estate market dynamics, making investment decisions and pricing mortgage-backed securities (Clapp and Giaccotto 1995). Therefore, it is important that these price indexes be as accurate as possible. Previous literature has shown that transaction-based indexes more accurately reflect the timing and magnitude of changes in property value (Downs and Slade 1999) because they avoid the potential problems of appraisal smoothing (Miles, Cole and Guilkey 1990, Webb, Miles and Guilkey 1992, Fisher, Miles and Webb 1999). The degree of smoothing implies an average lag of about three quarters at the individual appraisal level (Clayton, Geltner and Hamilton 2001). In addition, Hendershott and Kane (1995) find the continued excess of the appraisal-based American Russell-NCREIT property index over the fundamental value, indicating clear evidence of overappraisal. However, both Miles. Cole and Guilkey (1990) and Downs and Slade (1999) point out that the most significant problem in pursuing transaction-based commercial property price indexes is the lack of adequate data because individual commercial properties are traded infrequently.
Earlier studies on the construction of a transaction-based office price index focused on the selection of independent variables. The independent variables for the office property hedonic function include three sets of variables: physical or locational attributes associated with a transacted office unit (Miles. Cole and Guilkey 1990, Miles et al. 1991, Mills 1992, Shilton and Zaccaria 1994, Colwell, Munneke and Trefzger 1998, Munneke and Slade 2000, 2001), attributes associated with the office unit as an income producing property (factors related to the discounted cash flow; see Miles, Cole and Guilkey 1990, Webb, Miles and Guilkey 1992, Fisher, Geltner and Webb 1994) and attributes associated with the lease structure (Miles, Cole and Guilkey 1990, Downs and Slade 1999, Munneke and Slade 2000). Time dummy variables are commonly included to indicate the common price movement driven by general economic cycles. The coefficients of these time dummy variables are used to derive indexes that indicate the price changes.
Fisher, Geltner and Webb (1994) examine five alternative price indexes based on different assumptions, methods and data. They conclude that academicians and investors should not rely on any single index methodology, as each has its strengths and weaknesses. By examining different indexes, an in-depth insight on the functioning of the commercial property market can be obtained. More recent work expressed concerns about using different methodologies to construct transaction-based commercial property indexes. Gatzlaff and Geltner (1998) and Geltner and Goetzmann (2000) adopt the repeat sales method to construct indexes. Ling, Naranjo and Nimalendran (2000) use the latent variable model to estimate commercial real estate returns. Munneke and Slade (2000, 2001) study sample-selection bias in the construction of transaction-based commercial property price indexes and examine three time-varying parameter techniques for constructing reliable transaction-based commercial price indexes by considering the potential for sample selection bias.
Three potential problems are observed in the previous literature. First, different proxies have been used to measure the location attributes, the factors related to the discounted cash flow and/or the lease structure. The choice of independent variables affects both the model fit and the residual spatial and temporal correlations because the uncaptured information is reflected in the residuals. Second, methodologically, the spatial and temporal correlations among the residuals can impair the power of the traditional hedonic model that is typically used in constructing a transaction-based office price index and estimated by the OLS method. While this issue has received substantial attention in the residential housing market (Pace, Barry and Sirmans 1998), there is limited empirical research in the commercial property market. Third, individual office properties are less frequently traded than residential properties. This limits the ability of the traditional hedonic model to construct disaggregate transaction-based office price indexes or even aggregate price indexes over short time intervals.
This study addresses the above three problems by adopting a spatiotemporal autoregressive modeling approach. The approach (STAR model) was first employed by Pace, Barry, Clapp and Rodriquez (PBCR 1998) and Pace, Barry, Gilley and Sirmans (PBGS 2000). It then was further developed into a two-order spatiotemporal autoregressive model with Bayesian heteroskedasticity correction (2BSTAR) by Sun, Tu and Yu (STY 2004).
Compared with the previous models of office transaction prices, the 2BSTAR model has the following advantages. First, the model includes a set of spatial and spatiotemporal lags generated by purposely designed two-order spatial and temporal weight matrixes as explanatory variables. These variables can effectively capture the information associated with those immeasurable independent variables. Therefore, the 2BSTAR model should have a higher explanatory ability as it incorporates more information than the hedonic function. Second, including these lags as explanatory variables in a hedonic function can significantly reduce the spatiotemporal autocorrelations among the residuals. And after removing the autocorrelations, the Bayesian estimation approach can effectively detect and correct heteroskedasticity among the residuals. Both can improve the estimation efficiency and, hence, produce more robust estimates. Finally, weight matrixes are constructed using the comparable sales of a subject property. Against these matrixes, transaction-based office price indexes at disaggregate level can be derived for any time intervals, although the power of the 2BSTAR model in capturing the price dynamics at building level may be impaired by the reduction in the number of comparable sales (STY 2004).
This paper is organized as follows. In the next section, we elaborate the causes of spatial autocorrelations among office transaction prices. We then introduce the estimation methodology. The subsequent section gives an overview of the Singapore office market as well as the descriptive statistics of the data set. The empirical results are then presented. The paper closes with a summary of the principal findings and policy implications. An appendix illustrates the computation and estimation methodology.
Causes of Spatial Autocorrelations among Office Transaction Prices
Spatial autocorrelations refer to the possible occurrence of interdependence among the observations from a geographical area. Basu and Thidodeau (1998) explicitly state that neighborhood properties not only have similar structural characteristics but also share similar amenities. These neighborhood attributes partly determine the neighborhood property value, but they are rarely included in publicly available data sets. Besides, identifying relevant neighborhood boundaries are also difficult. Therefore, the impacts of neighborhood attributes on housing prices are difficult to measure in hedonic analysis. Although proxies can be used, the uncaptured neighborhood information flows into the hedonic residuals, resulting in residual spatial autocorrelation. STY (2004) further argue that in the multiunit residential housing market, the causes of spatial autocorrelation are complicated by both building and neighborhood effects. The spatial interdependence among the properties within one building is different from the one among the properties in the neighborhood. Therefore a two-order spatiotemporal filtering process is needed to remove the autocorrelations among the residuals.
In the office market, the causes of spatial dependence among office unit transaction prices can be identified from three aspects. First, strata-titled office units situated in one office building may share a building-specific spatial process. The quality of the building, for example, the common view, the layout, some common facilities (e.g., elevator) and the interior or exterior design features may contribute to the transaction prices of the office units located in the same building, which lead to spatial dependence among these prices. Such impacts are difficult to measure precisely, contributing to the spatial autocorrelations among the hedonic residuals. This is termed the building effect by STY (2004). However, unlike the residential market, which is generally divided into submarkets by location, the office market is typically subdivided into different grades of office buildings within the same location. This is largely due to the different user requirements of these two types of properties. Office users need to be in a particular location, such as the Central Business District (CBD), for complementary reasons, while residential users choose their location according to personal reasons. As office buildings are subject to changing technology more than residential buildings, office buildings will become obsolete when new technology is introduced.
Second, similar to the residential housing market, office units share the same neighborhood business environment and amenities (Colwell, Munneke and Trefzger 1998, Shilton and Zaccaria 1994). Such a spatial process is similar to the neighborhood effect in the residential housing market. It has an impact on office value but cannot be measured precisely. The inaccuracy in using proxies to measure these effects generates spatial autocorrelations among hedonic residuals.
Third, unlike the residential housing market where housing is predominantly bought for owner occupation, office properties are bought for investment as well as occupation. The variables indicating the ability to produce income partly determine the value of the property. These variables are the discount rates, future rental growth rates and future economic performance. However, the previous studies use different proxies to capture these effects; Miles, Cole and Guilkey (1990) point out that...
NOTE: All illustrations and photos
have been removed from this article.
More articles from Real Estate Economics
Sale before completion of development: pricing and strategy., June 22, 2004
Looking for additional articles?
Search our database of over 3 million articles.
Looking for more in-depth information on this industry?
Search our complete database of Industry & Market reports by text, subject, publication
name or publication date.
About Goliath
Whether you're looking for sales prospects, competitive information, company
analysis or best practices in managing your organization,
Goliath can help you meet your business needs.
Our extensive business information databases empower business
professionals with both the breadth and depth of credible,
authoritative information they need to support their business
goals. Whether it be strategic planning, sales prospecting,
company research or defining management best practices -
Goliath is your leading source for accurate information.
|
%20%20&position=1)