In this study we use Indicio, a forecasting tool developed by Indicio Technologies, to generate forecasts of Swedish existing home sales. We forecast house- and apartments sales in both Sweden and the Stockholm region. Our modelling suggests that sales of one and two dwellings houses are highly interrelated with the development of mortgage rates and the business cycle both on the national level and in the Stockholm region. Sales of tenant owned apartments are closely linked to the stock market development, apartment prices and google searches on housing related issues. Our combination of uni- and multivariate models forecasts slowly declining future sales of both houses and flats in both the Stockholm region and the country as a whole.

Trends, Breaks and Seasonality in Home Sales Data

We have access to two sources of data on home sales. The first is Statistics Sweden, which are basing their data on registration of title deeds and tenant-owner associations control data sent to the tax authorities. That means only existing tenant-owned dwellings are included, not newly produced apartments or converted rentals. Data is available 20-25 years back in time. The second source is Svensk Mäklarstatistik, which are basing their data on sales agents reports. They cover more than 95% of the deals that are done through agents. That means that a sale is reported when a deal is signed, which usually is 2-3 months before it is noted in official title deed data. It also means that sales of new houses and apartments only are included if they are made via a realtor. Monthly data is available from 2005 and onwards, however, coverage was smaller back then.

Both data sources have advantages and disadvantages. We, therefore, choose to construct time series of house and apartment sales based on both. We use the annual sales figures from Statistics Sweden but apply the seasonal pattern from Mäklarstatistik. We then calculate these sales series both for the country as a whole and for the Stockholm region. The calculated time series are shown below.

Yearly sales peaked in 2007 when almost 164 000 homes were sold, 64 000 houses and 100 000 apartments. Then it took until 2014 for apartment sales to reach 100 000 again and annual house sales hasn’t been over 60 000 yet. Both sales series have been leveling out or over the last 2-3 years. Home sales show a strong seasonal pattern with dips in January, February, July and December, and peaks in May, June, September and October. Besides the seasonal pattern we also observe major slumps during 2008, the end of 2011, spring 2016 and spring 2018 which coincide with financial crises, larger stock markets plunges, changes in taxation and/or strengthening of mortgage rules.

Determinants of Home Sales

Supply and demand factors such as stock of dwellings, household income, the business cycle, mortgage interest rates and household preferences on risk, debt and housing, in combination with price levels determine the number of homes sold. Statistics Sweden publish yearly data on the stock of owner-occupied and tenant-owned dwellings both on a national and regional level. They also publish quarterly data on household disposable income and monthly data on average actual mortgage interest rates and unemployment which could be used as a proxies for demand. Regional income and unemployment are available but mortgage rates are national. Monthly house and flat prices from Valueguard (HOX) are tested in the models, both national indexes and Stockholm indexes. We also test if changes in the stock market index (SIXRX), in the Economic Tendency Index (ETI) constructed by the National Institute of Economic and Social Research and in google searches (national and regional) for housing are significant since they are potential indicators of household sentiment.[1] Finally, we also include a survey from the European Commission where Swedish households are asked if they are planning to purchase or build a new home over the next twelve months.

Ranking of variables ability to determine Home Sales in Sweden and Stockholm, 2006 – 2019
  Sales of Houses, Sweden Sales of Apartments, Sweden Sales of houses, Stockholm Sales of Apartments, Stockholm
1 ETI (National) SIXRX Average Mortgage Rate (National) SIXRX
2 Average Mortgage Rate (National) HOX Apartments (National) ETI (National) Real Disposable Income per Capita (Stockholm)
3 SIXRX Real Disposable Income per Capita (National) Unemployment Rate (Stockholm) Disposable Income per capita (Stockholm)
4 Purchase Survey (National) Average Mortgage Rate (National) Purchase Survey (National) Google Searches (Stockholm)
5 Unemployment Rate (National) Google Searches (National) SIXRX HOX Apartments (Stockholm)
6 HOX Houses (National) Disposable Income per capita (National) Google Searches (Stockholm) Stock of Tenant-Owned dwellings per capita (Stockholm)
7 Google Searches (National) Stock of Tenant-Owned dwellings per capita (National) Real Disposable Income per Capita (Stockholm) Average Mortgage Rate (National)
8 Real Disposable Income per Capita (National) Unemployment Rate (National) Disposable Income per capita (Stockholm) Purchase Survey (National)
9 Disposable Income per capita (National) Purchase Survey (National) HOX Houses (Stockholm) ETI (National)
10 Stock of Houses per Capita (National) ETI (National) Stock of Houses per Capita (Stockholm) Unemployment Rate (Stockholm)


The variables that end up highest on the ranking are the the Economic Tendency Index (ETI) and Average Mortgage Rates when it comes to house sales whereas the Stock Market Index (SIXRX) is the best single determinant of apartment sales. Measures of household income and prices (HOX) come rather high up in the apartment sales ranking but quite low in the house sales ranking. Unemployment Rates go the other way, high in house sales ranking but low in apartment sales ranking. The stocks of dwellings have no larger single variable correlation to sales, neither do google searches or the purchase survey.

Both House and Apartment Sales Set to Slowly Decline

Indicio allows us to combine 18 different univariate model and 15 different multivariate model forecasts into one single forecast, thus improving forecasting accuracy considerably.[2] Models are evaluated according to their historic forecasting accuracy and forecasts are combined with weights based on the models respective Stepwise Root Mean Square Error. For Swedish house sales the best multivariate models include Mortgage Rates and ETI as additional variables, the Swedish apartment sales models include HOX Apartments, the Purchase Survey, Stock of Apartments, Google Searches and Real Disposable Household Income. The Stockholm apartment models include SIXRX, Google Searches, Unemployment Rate and ETI while the Stockholm house models include Real Disposable Income, HOX Apartments and Stock of Apartment. The combined forecasts incorporate between 12 (Stockholm house sales) and 28 (Stockholm apartment sales) different models.

All four forecasts show a slow decline in sales over the coming year. In the whole nation, House Sales are set to slow by 2 percent from 4 650 (seasonally adjusted) in July 2019 to 4 630 in July 2020, apartment sales is forecasted to slow down by 5 percent from 9 370 to 9020. In the Stockholm region, the number of sold houses is expected to drop from 320 to 310, equivalent to -3 percent. Finally, apartment sales in Stockholm will fall by almost 4 percent from 3 370 to 3 320 over the coming year.

Home Sales Forecast
Sales of Houses, Sweden


Sales of Apartments, Sweden Sales of houses, Stockholm Sales of Apartments, Stockholm
Calculated Sales in July 2019 (seasonally adjusted) 3862

(4 652)

6 389

(9 374)



1 728

(3 374)

Percentage change, July 2018-July 2019 +3,5% +14,9% -1,6% +10,9%
Forecasted Sales in July 2020 (seasonally adjusted) 3 774

(4 635)

6 058

(9 020)



 1 679

(3 325)

Percentage change, July 2019 -July 2020 -2,3% -5,2% -2,9% -3,5%




[1] Indexes of Google Searches on housing related terms were downloaded from Google Trends.

[2] Indicio contains estimation of ARIMA, ETS, Theta, STL, TBATS, Neural, VAR, VECM, VAR, VAR Lasso, VARMA, ARDL and BVAR models.