The historical relationship, as estimated by a variety of statistical models, between Swedish house prices and their driving forces suggests that prices will increase at an annual rate of two to three percent over the next one to two years. Large price falls would only occur if equity markets fell significantly, household income fell, interest rates rose steeply and/or housing construction accelerated markedly.

Housing markets are not like most other markets. Housing is a necessity, everybody needs to live somewhere, there are really no substitutes to it – very few are voluntarily homeless. But housing is not a staple since almost every dwelling has its own unique qualities. Most dwellings are hard to move without effort, making location important determinant of market value. This makes it hard to measure and compare prices in an aggregate way. The market is dual since it is both a consumption and an asset market. People consume housing when they live in their rented or owned dwelling. But those who own an apartment or a house are also hoping to get a return on their investment. For most people, buying a house and taking out a mortgage are the biggest transactions in their life. One other interesting aspect is that there is not much of international integration to be seen. Regulations are still very much national and they can and do differ quite significantly between countries.

All this means that the housing market often is in the limelight of people’s minds and that changes in supply, demand and prices get a lot of media coverage. Headlines about price trends and future price developments attracts big audiences. In Sweden a quite steep price fall during autumn 2017 has been followed by more or less stationary house prices during 2018. The market is uncertain. There are forecasters who claim that prices are about to resume their fall and there are those who forecast unchanged or slowly rising prices. Although there are some forecasters who put much effort and knowledge into their forecasts there are plenty who seem to base their predictions on quite loose grounds. Forecasts are often intuitive, judgmental and/or subjective and not really based on empirical or theoretical foundations.

In this study we will use econometric techniques to forecast Swedish housing prices. We will use the forecast tool Indicio, developed by F&A Forecasting specially to facilitate advanced forecasting without requiring deep statistical knowledge. Our goal is to find the best models for predicting Swedish house prices and we will model the Statistics Sweden (SCB) Price Index, which is available on a quarterly basis back to 1986, as well as the Valueguard Housing Index (HOX) for flats, available on a monthly basis starting from 2005.

Univariate forecast

Indicio can calculate thirteen different univariate models. These models are based only on the historical data of the variable of interest itself, and they capture eventual trends, cycles, seasonality etc. Forecasts based on univariate models can be quite useful, especially when it comes to short term predictions. They are also useful as benchmark models for the multivariate models. Below is a two year ahead forecast for the SCB index, where the univariate model forecasts have been combined to one, weighted according to their historical accuracy. The index is expected to remain unchanged in the first quarter of 2019 and then return to a stable annual rate slightly above 2 percent in the coming quarters. The two year ahead confidence intervals gives us a 50 percent probability that prices will be ±2.9 percent from the point forecast, 75 percent probability that they will be within ±4.5 percent and 95 percent probability that they will be within ±8.6 percent at the end of the forecast horizon respectively. This is not bad at all for a benchmark model but hopefully we will be able to improve it further by incorporating other driving variables.

 

The drivers of house prices

There are a vast amount of factors that are affecting aggregate house prices which we test.[1] At the supply side we are incorporating the number of existing dwellings, housing construction, construction costs and existing home sales as possible driving factors. On the demand side we are looking at population growth, household disposable income, GDP growth, employment growth, unemployment, household financial wealth as well as mortgage rates and household debt. Finally we also include consumer price inflation (CPI) and apartment rents as potential drivers. All in all we let Indicio test 18 variables as drivers of house prices. Indicio analyses the dynamics based upon each indicator’s effect on out-of-sample accuracy and ranks the indicators according to this. Since there have been large changes in the institutional and regulatory setting, e.g. taxes, subsidies, credit conditions, over the last 30 years we also check whether the ranking differs over time. We do this by varying the time span.

The table below shows the ranking of the indicators during three different time periods. The supply factors existing home sales and dwellings started end up at the low ranges in the ranking, independent of time period chosen. Dwellings completed seem to climb in the ranking over time while constructions costs consistently ranks quite high. The demand indicators total and younger population ends up rather high in the rankings together with household income and financial wealth while GDP is in the middle and employment and unemployment are in the bottom. The ranking of the financial indicators vary quite a bit over time, household debt and debt service ratio come high in the overall ranking but their ranking is sinking over time while mortgage and bond rates are climbing. Apartment rents and CPI inflation are also climbing in the rankings over time.

 

Indicators ranked according to their influence on house prices (top ten)
1986-2002 2001-2018 1986-2017
Household Debt Ratio Household Financial Wealth Household Debt
Disposable Household Income Disposable Household Income Debt Service Ratio
Population 20-34 years Population Disposable Household Income
Household Debt CPI Population
Construction Costs Construction Costs Household Financial Wealth
GDP Household Debt GDP
CPI Dwellings Completed Population 20-34 years
Household Financial Wealth Mortgage Bond Rate Mortgage Rate
Population Population 20-34 years Construction Costs
Mortgage Bond Rate Unemployment Rate Mortgage Bond Rate

 

Multivariate forecast

Indicio can estimates twelve different multivariate models and then combine their forecasts into one single by weighting them according to their historical accuracy. Since there seems to be quite large structural changes in the statistical relationships, we choose to focus on models estimated over the period 2001-2018. Household disposable income and financial wealth, population growth and household debt were the factors that were singled out as significant drivers of house prices. The single forecast, constructed by combining the top univariate and multivariate model forecasts, according to their stepwise historical accuracy, shows that house prices are expected to remain unchanged in the first quarter of 2019 and then settle around an annual rate around 2 percent in 2019 and 2020. The two year ahead confidence intervals translates into ±1.8 percent (50% interval), ±2.8 percent (75% interval) and ±5.4 percent (95% interval) at the end of the forecast horizon respectively. A major improvement from the univariate forecast.

Apartment prices

Public data over apartment prices with sufficient quality are only available from 2005 onwards. That means we have a quite short time span to use in our estimations. So instead of elaborating with different time periods we try different combinations of explaining variables. We find two different set of variables that can explain the development of apartment prices. The first set is mortgage bond rates, household debt, household disposable income and dwellings completed. The second set is household financial wealth, GDP and house prices (as measured by Valueguard). Both models give a forecast of rising apartment prices in the short run (2019q1) and an average annual rate of around 2-3 percent in the long run (one to two years).

 

 

Scenarios where there’s a significant fall in prices

Our models give a fairly optimistic picture of future house and flat prices. But we know for sure that political and economic factors can change in unpredictable ways. Thus one important question is how robust house and apartment prices are against unexpected events? We therefore check what it would take in terms of slower income growth, higher interest rates, falling stock market etc to really hurt the housing market, i.e. trigger a price fall of at least ten percent. To investigate this we single out the best of the models and impose restrictions on the driving forces to see what happens to house prices. In the SCB House Price Index Model, we have to assume that annual growth in disposable income per person and household debt goes from 4 respectively 6 percent during 2018 to -1 respectively 0 percent in 2019 and 2020. That is quite a strong cycle downturn. In the apartment price models the biggest effects comes from household financial wealth and the number of newly built dwellings. Here we have to assume such a big fall in the equity market that household financial wealth falls mote than it di during the great recession 2008/09 in the first model. Or that housing construction accelerates so that the yearly number of new dwellings lands around 85 000, (instead of the originally forecasted rate of 50 – 55 000) at the same time as interest rates rise significantly in the second model.Thus we can conclude that, judging by our econometric exercises, the Swedish housing market seems to be quite robust. Prices are expected to rise by between two percent (houses) and four percent (apartments) on a yearly basis over the next two years. Possible external chocks would have to be rather big to trigger larger price falls.


[1] Examples of more academic studies where Swedish housing prices are modeled are : Claussen C.A., Jonsson M., Lagerwall B. (2011), “A macroeconomic analysis of houseprices in Sweden”, Chapter 2 in The Riksbank’s inquiry into the risks in the Swedish housing market, Turk R. A. (2015), “Housing Price and Household Debt Interactions in Sweden, IMF Working Paper 15/276, Sörensen P.B. (2013) “The Swedish housing market: Trends and risks”, Report 2013/5 Swedish Fiscal Policy Council