Abstract
The Danish housing market saw a decline as a result of the financial crisis beginning in 2008. In Denmark the housing prices have been on the rise for the past several years. This situation is expected to have been brought about from the general economic developments. The purpose of this paper is to attempt to create a vector autoregressive model, founded in a series of macro-economic factors which are expected to affect the Danish housing market. The low interest rates constitute part of the explanation for the rise in housing prices. International and domestic interest rates have been declining for the past several years, which has resulted in lower user costs for property owners. Part of these user costs is property tax, and in order to pay this tax, the property owner must have a disposable income. Therefore, the model-estimation will be built on four variables: Housing prices, disposable income, real estate tax revenue and interest rate on mortgage bonds.
The selected data is examined for the presence of unit root through a Dickey-Fuller test, and the model-estimation is hence conducted. An estimation of simple autoregressive models and vector autoregressive models is carried out. The estimated models are used in forecasts for Denmark as a whole, and for the Danish regions, such as to form six VAR models and six AR models. The cause for necessity of estimating different models for the separate regions, is that the housing prices have risen faster in some parts of the country than others. The estimated models are thus applied, in order to create forecasts with 2 different forecast intervals.
The forecast values are then closely analysed, such as to facilitate an evaluation of the forecast accuracy of the models. Denmark as a whole, and the Danish regions – with the exception of Region Sjælland – get the most precise forecasts when the parsimonious AR models is applied. The best forecast models are subjected to analysis, to determine the future values for development in housing prices.
That which can be deduced, is that the generated VAR models have not been more successful in creating forecasts than the simple AR models. It is discussed whether the cause for the low accuracy in the VAR models can be found in the data from the period of observation, which was influenced by a housing bubble and a financial crisis.