ABSTRACT

Log linear hedonic models are ubiquitous in econometric real estate research even though functional form assumptions are often not satisfied and the nested structure of homes in suburbs is not captured adequately. This study focuses on appraising different residential property types located throughout South Africa, investigating a flexible approach which does not assume some explicit functional form. The objective of this paper was to fit and compare two hierarchical generalised additive models to 412 500 property listings from 2013 to 2017. A gamma hierarchical model with random intercepts for the suburb provided the best fit and generalisability, while accounting for the spatial dependency in the data. The results show that hierarchical generalised additive models capture complex shapes between listing prices and structural property characteristics, and further reveal that partial pooling is useful to capture between suburb variability.

摘要

对数线性特征模型在计量经济学房地产研究中无处不在，即使函数形式假设通常不满足并且郊区房屋的嵌套结构没有得到充分捕捉。本研究的重点是评估位于南非各地的不同住宅物业类型，研究一种灵活的方法，该方法不假定某些明确的功能形式。本文的目的是将两个分层广义加性模型与 2013 年至 2017 年的 412 500 个房产列表进行拟合和比较。 具有郊区随机截距的 gamma 分层模型提供了最佳拟合和普遍性，同时考虑了数据。