Predicting apartment rental prices in Finland

Thumbnail Image
Journal Title
Journal ISSN
Volume Title
School of Business | Master's thesis
Degree programme
Information and Service Management (ISM)
The affordability of housing is a significant challenge for many households, especially those with lower incomes. Inflation, various cost-of-living crises, and stagnating wages have made renting difficult to afford. As such, understanding the drivers of rental prices is an important step in finding maximum value in the market. This phenomenon also applies to Finland, where low-income households are disproportionately burdened by rental prices, and even the cost of owning a home can be cheaper than renting in the long run. While previous research has attempted to understand and predict rental prices in other places, none have yet done so for the Finnish market. The expectation is that local differences could potentially lead to different results. Machine learning can help in the identification of the most important drivers of rent, as well as in predicting average prices based on these factors. The thesis uses publicly available data from Statistics Finland to gather information on average rent prices per postcode, as well as a number of other potential features for the different models. Due to the complex nature of this data, linear regression, linear mixed model regression, polynomial regression, random forest regression, and support vector regression were all used to analyze different aspects of this problem. Despite limitations in the availability of data, the findings show that apartment prices, the number of rooms, and location are some of the most important factors in predicting rental prices. On the other hand, while housing benefits were initially believed to be among the greatest drivers of rent due to their direct price transfer effects, they turned out to be a relatively poor predictor of rents, perhaps due to the relatively small local differences in average subsidies. This was also the case for other benefits, such as student benefits or old-age security payments. The effects of local population demographics also proved relatively insignificant in the prediction of rental prices, except for average household incomes. Ultimately, random forest regression turned out to be the best model for the prediction of Finnish rental prices, although both linear models also outperformed expectations. These results show how useful machine learning applications can be in predicting such complex issues.
Thesis advisor
Malo, Pekka
apartment rent, cost-of-living, machine learning, predictive modeling
Other note