COMPARISON OF SPATIAL HEDONIC HOUSE PRICE MODELS: APPLICATION TO REAL ESTATE TRANSACTIONS IN VANCOUVER WEST

This study compares hedonic house price models for single family properties in Vancouver West, Canada. The real estate literature has shown that traditional hedonic models based on OLS are unable to handle spatial effects inherent in housing markets, prompting the application of spatial econometric methods. This study compares four hedonic house price models: (i) classical OLS model, (ii) OLS model with neighborhood code dummies, (iii) Spatial Durbin Model, and (iv)Geographically Weighted Regression. The latter two models are common spatial econometric techniques that researchers have used. Models are compared based on model 2 R , out-of-sample prediction error, and ability to remove spatial effects from the data. Results indicate that Geographically Weighted Regression is the best performing model. In addition, classical OLS overestimates effects and is unable to address spatial effects. All four models predict a similarimpact of property attributes on sale price.