Kernel-based geographically and temporally weighted autoregressive model for house price estimation

Area of Science:

• Geographical Information Systems (GIS)
• Spatial Statistics
• Econometrics

Background:

• Geographical data modeling requires addressing spatiotemporal nonstationarity and autocorrelation.
• The geographically and temporally weighted autoregressive (GTWAR) model estimates house prices by considering these factors simultaneously.

Purpose of the Study:

• To propose a novel kernel-based GTWAR (KBGTWAR) model.
• To enhance the modeling of spatiotemporal nonstationarity and autocorrelation in house price estimation.

Main Methods:

• Incorporating support vector machine regression principles into a spatially and temporally varying coefficients model.
• Applying the KBGTWAR model to a case study of housing prices in Shenzhen, China (2004-2008).

Main Results:

• The KBGTWAR model achieved the lowest Residual Sum of Squares (RSS) and highest R-squared (R2).
• Significant improvements were observed compared to the existing GTWAR model: RSS decreased from 12.0 to 4.5, R2 increased from 0.914 to 0.968, and F-statistic rose from 3.84 to 4.45.
• The model demonstrated high goodness of fit and explanatory power for spatiotemporal dynamics.

Conclusions:

• The KBGTWAR model effectively captures spatiotemporal nonstationarity and autocorrelation in house price data.
• The proposed model offers a robust tool for real estate management policy formulation.