Appraisal Using Generalized Additive Models

Many of the results from real estate empirical studies depend upon using a correct functional form for their validity. Unfortunately, common parametric statistical tools cannot easily control for the possibility of misspecification. Recently, semiparametric estimators such as generalized additive models (GAMs) have arisen which can automatically control for additive (in price) or multiplicative (in ln(price)) nonlinear relations among the independent and dependent variables. As the paper shows, GAMs can empirically outperform naive parametric and polynomial models in ex-sample predictive behavior. Moreover, GAMs have well-developed statistical properties and can suggest useful transformations in parametric settings.