Developing a model to isolate the effect of location on property value
Introduction
While location is one of the most important factors affecting the financial qualities of a property, quantifying the financial effect of location on property is challenging. Separating the financial effect of location from other variables is difficult due to the complex interactions between these variables and the absence of independence between these variables.
Services like realestate.com.au allow for the financial comparison of locations (suburbs in this case) by average property, broken down by property type and number of bedrooms. This accounts for the dependence between a property’s location, property type, and number of bedrooms but not other interdependent variables, such as the number of bathrooms or quality of a property.
Using the number of bedrooms a property has as an example of these complex interactions, increasing the number of bedrooms that a house has from two to three will have a different effect on the sale value of a property in different suburbs, and at the same time properties in certain suburbs will be more likely to have a greater number of bedrooms than properties in other suburbs.
The sections below discuss the model and data that were used to perform an analysis on the effects of location on property sale value, rental income, and rental yield. For the analysis and its results see the follow up to this article, linked below.
The Estimator
The model built for this analysis estimates a property’s sale value, rental income, and rental yield based on variables related to that property. A linear estimator was chosen because of the limited size of the data set and a desire for interpretable parameters. Polynomial features and combinations of features were created to improve the accuracy of the model and to capture interactions between the variables in the data set.
Variables used by the model:
- Number of bedrooms as a polynomial feature of degree 3;
- Number of bathrooms as a polynomial feature of degree 3;
- Number of garage spaces as a polynomial feature of degree 3;
- Type of property as a categorical variable; and
- Suburb the property resides in as a categorical variable.
Variable interactions used by the model:
- Number of bedrooms and property type as a linear feature;
- Number of bathrooms and property type as a linear feature;
- Number of garage spaces and property type as a linear feature;
- Number of bedrooms and suburb as a linear feature;
- Number of bathrooms and suburb as a linear feature;
- Number of garage spaces and suburb as a linear feature;
Variants of the model were estimated for the following financial quantities:
- Total property sale value, in 2016–17 Australian dollars; and
- Rental income per-week, in 2016–17 Australian dollars.
Neither this set of variables or the variable interactions added to this set of variables account for all of the variables that affect the financial qualities of a property. Because of this and because of the lack of independence between all of the variables that affect the financial qualities of a property, the model will be unable to account for all of the dependencies between variables not included in the model and the location of a property. This means that the financial comparisons of locations that are done in this analysis will actually be a comparison of location and other variables that are not included in this model and which are dependant on the location of a property. A notable example is the quality of a property, which could be correlated with a property’s location and travel times between a property and important areas such as the CBD.
The Data
This analysis is based on properties being advertised as for rent and for sale between late 2016 and early 2017. Property advertisements with incomplete information or information that was clearly incorrect were excluded from the data set, properties that were listed multiple times as parallel or series listings were reduced into a single data sample. The data contains the following breakdown of samples.
Sale samples:
- 3586 houses; and
- 1969 units.
Rentals samples:
- 1745 houses; and
- 2246 units.
These sample sizes are not large, especially when broken down by suburb. This should be remembered when considering the results of this analysis.
Sample number for sales by suburb
Sample numbers for rentals by suburb
Using the model
The model allows for the type of property, and the number of bedrooms, bathrooms, and garage spaces that the property has to remain constant, while the property’s location is varied. This allows for the effect of location, on the financial quantities being studied to be analysed independently to the variables that the model accounts for. This methodology is used to perform an analysis on the effects of location on property sale value, rental income, and rental yield in the ACT in Australia in the follow up to this article, linked below.
