Academic journal article Cityscape

Evaluating Spatial Model Accuracy in Mass Real Estate Appraisal: A Comparison of Geographically Weighted Regression and the Spatial Lag Model

Academic journal article Cityscape

Evaluating Spatial Model Accuracy in Mass Real Estate Appraisal: A Comparison of Geographically Weighted Regression and the Spatial Lag Model

Article excerpt

(ProQuest: ... denotes formulae omitted.)

Introduction

Ad valorem property taxes are a prominent source of government revenue in jurisdictions around the world. Taxing authorities are held accountable to ensure that these valuations are fair and equitable. In such roles, the optimization of the accuracy of mass real estate valuation approaches is critical.

Because of their precision and time- and cost-saving advantages, real estate mass appraisal methods that employ multiple regression-based models, known as automated valuation models (AVMs), are becoming increasingly prominent in industry practice and have received attention from the academic community. AVMs are used in a host of industries-both public and private- including loan origination, fraud detection, and portfolio valuation (Downie and Robson, 2007), and are promoted and advanced by such organizations as the International Association of Assessing Officers (IAAO). Statistical standards of equity established by such organizations give additional benchmarks by which modelers may test various approaches and methodologies.

Academic research has expanded regression models using geographically specific dummy variables and distance coefficients, and, although this approach has been shown to improve ordinary least squares (OLS)-based regression models, they often still suffer from biased coefficients and t-scores (Berry and Bednarz, 1975; Fotheringham, Brunsdon, and Charlton, 2002; McMillen and Redfearn, 2010). Some researchers (Fotheringham, Brunsdon, and Charlton, 2002) have used geographically weighted regression (GWR), a locally weighted regression technique, which has improved model performance by employing a spatial weighting function and allowing for coefficients to fluctuate across geographic space (Huang, Wu, and Barry, 2010; LeSage, 2004). Similarly, the spatial lag model (SLM)-a spatial autoregressive (SAR) model- addresses spatial heterogeneity by including an autocorrelation coefficient and spatial weights matrix (Anselin, 1988).

Because real estate markets behave differently across geographic space, AVMs free of spatial consideration often produce inaccurate, misleading results (Anselin and Griffith, 1988; Ball, 1973; Berry and Bednarz, 1975). GWR is prominently demonstrated throughout literature as a more accurate alternative to multiple regression analysis (MRA) AVMs (for example, Borst and McCluskey, 2008; Lockwood and Rossini, 2011; McCluskey et al., 2013; Moore, 2009; Moore and Myers, 2010). Similarly, SAR models have been sufficiently demonstrated to increase the predictive accuracy of such models (Borst and McCluskey, 2007; Conway et ah, 2010; Quintos, 2013; Wilhelmsson, 2002). Descriptions of their methods and findings are summarized in exhibit 1.

Despite the popularity of both GWR and SLM models in housing research, to our knowledge, a study that simultaneously compares the performance of GWR and SLM using industry-accepted IAAO standards and that extrapolates each model's performance to aggregate and subaggregate levels has yet to be published. Färber and Yeates (2006) found GWR to have more accuracy and produce less spatially biased coefficients than SAR models, but no comparison has been made of how each performs against the other in the context of mass appraisal for tax assessments. A major finding of Bidanset and Lombard (2013)1 is that traditional measures of hedonic model performance (for example, the Akaike Information Criterion [AIC],R2) do not necessarily indicate which model will perform the best given the assessment industry standards of uniformity (that is, coefficient of dispersion [COD]).2 This article compares spatial regression techniques of the SLM and GWR and compares not only their prediction accuracy ability but also their attainment of IAAO equity and uniformity standards. Given the increasing availability of Geographic Information System, or GIS, data and advances incomputational ability to perform spatial AVMs, the understanding of the capability that each method lends to governments in reaching more accurate value estimations is critical. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.