Forecasting House Prices in Norway: A Univariate Time Series Approach

Abstract

The main objective of this thesis is to forecast residential house prices in Norway from April 2013 to March 2014. Three univariate time series models are employed in an attempt to find an appropriate fit. The three are an AR-, an ARIMA- and an exponential smoothing state space (ETS) model. The forecast from the three models are also combined, in an effort to improve upon the accuracy of the single “best” forecast. This study implements a weighting scheme based on inverse out-of-sample mean square errors (MSEs). Weights of 0.29, 0.21 and 0.50 are assigned to the AR-, ARIMA- and ETS-models, respectively.

The analysis identifies the forecast from the ETS-model as the most accurate among the individual models based on both out-of-sample root mean square error (RMSE) and mean absolute scaled error (MASE). The weighted forecast has a higher RMSE (less accurate), but a lower (more accurate) MASE compared to the ETS. Thus, we cannot reject the idea that a combination of forecast can in fact improve upon the accuracy of the single best forecast, since the two measures give conflicting results.