Recently I attended the New Zealand Association of Economists (NZAE) conference and presented on the New Zealand Land Value Model we developed here at MRCagney with the Ministry of Business, Innovation, and Employment (MBIE) in New Zealand. The abstract for the paper is perhaps the best introduction to NZLVM:
Understanding the effects of present land use policy on land values is fundamental to understanding housing markets and informing policy. Historically, attempts to understand the impact of land use policy have been one-off, and not easily generalizable to other locations. This paper introduces the New Zealand Land Value Model (NZLVM). NZLVM is an automated model which makes use of CoreLogic ratings data and LINZ parcels to analyse land values at the parcel level. It focuses on finding discontinuities in land values at (a) rural / urban boundaries and (b) between rural, residential, industrial, commercial and retail zones. The model can be applied in various contexts and has already been applied to 20 different urban areas in New Zealand.
The results suggest that the nature and impact of zoning varies greatly by location throughout New Zealand, with large differences in estimated discontinuities. Furthermore, it suggests that land demands within an urban area also spatially heterogeneous. The outputs from the model can be used to provide an understanding of the spatial demand for land use and therefore support more informed decision-making in zoning and infrastructure supply.
I thoroughly enjoyed giving the presentation, and a lot of the feedback was extremely positive. The figure below presents some of the outputs from the model. In this case, it is a simple scatter plot of land values inside and outside of the rural urban boundary in Auckland.
I will upload the presentation shortly; however, in this blog post I wanted to discuss some key lessons learnt from NZLVM project.
Firstly, R provided a fantastic environment to conduct statistical analysis in, and I should make a special mention of the sf (short for simple features) package developed by Edzer Pebesma for geospatial analysis. sf provided a modern, easy, and tidy way to analyse the millions of property parcels that the model used. Furthermore, as a GIS user, providing an integrated environment for both normal and spatial analysis meant we didn’t need to reach outside of R, and in particular to any third party proprietary tool which would have made reproducibility extremely difficult.
Secondly, the importance of the research to be reproducible. In this, case, MBIE wanted to be able to run the NZLVM model’s code and analysis themselves. To ensure reproducibility, MRCagney developed an R package which could be loaded on an MBIE analysts machine and the entire analysis could be re-run, with original and new updated data.
Finally, and this is connected to the previous lesson, was that the project is re-usable and shareable. By being shareable, it meant that the codebase could be easily shared with other developers and built upon. For this, MRCagney set up a private Gitlab repo – which MBIE can share with who they please – from which the library could be installed from. This taught me that Git is not only a helpful internal tool for development, but it is also important external tools for sharing work outside of the company with clients.
In summary, NZLVM is a model or codebase, which can be easily re-run and expanded upon. Whilst models had been constructed looking at boundaries in Auckland, New Zealand, none of them were easily reproducible and could be consistently re-applied.
I should also make a special mention to my colleagues at MRCagney who helped develop the package: Peter Nunns, who really led the economic analysis and Alex Raichev who taught me how to be a proper developer, and who’s knowledge was instrumental in developing the package and using Git successfully.
A lot more information on the project can be found on the MBIE website here.