Forecasting UK Housing Market Using Artificial Neural Networks

Abstract

In this dissertation, we examine the use of Artificial Neural Networks (ANN) to predict trends within the residential property market by estimating the future average house price, while using the national economic, employment, social and residential property time-series data for the training purposes. The selection of input variables is based on meaningful and deliberate grounds. In particular, statistical correlation analysis, study and assessment of economical theory and of previous literature, and in-person interviews with the industry experts are conducted. This project is a part of an ongoing project at the Oxford Man Institute of Quantitative Finance, which aims to construct an innovative data extraction model using advance data extraction tools and algorithms, and subsequently apply this methodology on the housing sector. We also demonstrate the practical application of ANNs in the housing market, by undertaking an extensive experimentation and analysis of over 65 forecasting models with varying architectural design and configuration, while using a dataset of over 209 monthly observations (Jan-92 to May-09). The model with the best performance is selected and further empirically and comparatively analysed, in terms of predictive accuracy and reliability of the results. The findings certainly substantiate the validity of our selected forecasting model, and suggest that it is suitable to forecast the housing market trends. Our model obtained 0.86% mean squared error, while 7.7% absolute error, which is better result as compared to those, claimed by most of the earlier studies. The model also successfully computed the future housing market trend as estimated by the existing HPIs (namely Nationwide, Halifax and Land Registry).

Author-supplied keywords

Cite this document ^(BETA)