Weenink, Philibert Lorenz, MSc (2022) Local spatio-temporal regression kriging for property price predictions. Master thesis.
|
Text
Thesis_Real_Estate_Studies.pdf Download (10MB) | Preview |
Abstract
This thesis proposes a hedonic regression model that accounts for spatio-temporal dependence and heterogeneity in property transaction prices. The considered methodology extends upon purely spatial geostatistical methods and applies local spatio-temporal regression kriging (LSTRK). We account for spatio-temporal dependence by specifying dual structure variograms. Since such variograms consider distances between properties in space and time simultaneously, they allow for the estimation of a single covariance structure that controls for spatio-, temporal- and joint dependence. We account for heterogeneity of both attribute prices and dependence structures by performing the analysis per individual submarket. To empirically examine the prediction accuracy of our model, out-of-sample forecasts are made on residential housing prices. We consider a housing transaction dataset of 57,154 properties in Mecklenburg County, North Carolina. Our data is split into 7 spatial zones and 5 temporal zones, providing us with a total of 35 submarkets. For each submarket, in-sample- and out-of-sample subsets are obtained. These allow for model training and prediction assessment respectively. In terms of mean absolute error, median absolute error, mean squared error and root mean squared error, the proposed methodology outperforms traditional hedonic models consistently. Hence, empirical results suggest that LSTRK indeed provides more accurate predictions than traditional OLS models.
| Item Type: | Thesis (Master) |
|---|---|
| Degree programme: | Real Estate Studies |
| Supervisor: | Liu, X. and Vlist, A.J. van der |
| Date Deposited: | 31 Aug 2022 10:32 |
| Last Modified: | 31 Aug 2022 10:32 |
| URI: | https://frw.studenttheses.ub.rug.nl/id/eprint/4009 |
Actions (login required)
 |
View Item |