How will house prices in the Spanish provinces evolve in 2021?
CaixaBank Research has developed new models for forecasting house prices at the level of province using large amounts of information (big data) and applying machine learning techniques. According to these models, house prices will fall in 7 out of 10 Spanish provinces in 2021 and grow only very moderately in the rest. Comparing current forecasts with those projected by the models before the pandemic, a notable correction can be seen in the expected growth of house prices in one year’s time, approximately 4 pp on average. This correction has been more pronounced in provinces with a higher urban concentration and greater dependence on foreign tourism, although they are still the most dynamic in spite of this.
In this article we examine the forecasts of house prices for 2021 at the level of province, obtained from the machine learning models of CaixaBank’s new real estate big data tool. The tool combines millions of internal CaixaBank data with reliable external sources of information. This enables the application of machine learning algorithms, which improve the forecasts as more information is available. The Ministry of Transport, Mobility and Urban Agenda’s historical series of house prices has been used to train the models, based on free housing appraisal prices. Most of these data come from appraisals of properties more than 5 years old, so the house price forecasts in this article largely reflect trends in the second-hand property market.
Expectations regarding the trend in Spanish house prices have altered extensively as a result of COVID-19. If we compare the distribution of house prices projected by the models one year ago (October 2019) with the most recent projections (October 2020), we can see a shift in the price distribution to the left. In other words, the models are currently projecting significantly lower house prices than those forecast a year ago.
Specifically, in October 2019 the models predicted that house prices would rise in one year’s time in almost all provinces, with an average increase of 3.5%. The highest rises were expected in the provinces of Valencia (+7.4%), Navarra (+6.5%), Zaragoza (+6.5%) and Madrid (+6.4%). And only four provinces were projected to see a decline: Ciudad Real (–1.4%), Zamora (–1.3%), Palencia (–0.8%) and Ávila (–0.3%).
while an increase was expected in 9 out of 10 provinces a year ago, today a decrease is expected in 7 out of 10.
In October 2020, the models forecast that, in one year’s time, house prices will have fallen in 7 out of 10 provinces. However, the models still project price increases in almost one third of Spain’s provinces, albeit a much more moderate rise than predicted a year ago, particularly in Malaga (+2.0%) and Madrid (+0.8%).
It is therefore evident that the downward adjustment of forecasts over the past year has been widespread and considerable. In particular, the house price growth forecast over a one-year horizon has been lowered by about 4 pp between October 2019 and October 2020.
The greatest adjustment in forecasts between October 2019 and October 2020 has been in the urban provinces.1 While, one year ago, the models predicted a 5% revaluation on average in house prices over a one-year horizon for the urban provinces, they now predict a slight fall in prices between Q3 2020 and Q3 2021 (–0.2% on average). In contrast, rural provinces have seen a significantly smaller adjustment in their house price projections (about 3 pp in the past year). However, as these real estate markets were already much less dynamic before the pandemic (one year ago, an average increase in house prices of 1.2% was projected for rural provinces), the change in expectations has led to notable decreases in house price forecasts for most rural areas (–2.0% on average between Q3 2020 and Q3 2021).
A similar pattern can be observed if we divide the provinces into two groups according to their degree of dependence on foreign tourism. Specifically, we consider a province to be dependent on foreign tourism if more than 10% of CaixaBank’s POS terminal transactions in the province in 2019 were carried out with foreign cards.2 The result is that the house price forecast for tourism-dependent provinces has altered considerably in the past year: from an average one-year growth rate of 4.7% projected in October 2019 to the current forecast of –0.4%. On the other hand, less tourism-dependent provinces have seen a smaller adjustment in their house price forecasts over one year, although these are precisely the provinces with the largest fall in prices projected for the coming year.
- 1. A province is considered rural if more than 50% of its population lives in a rural municipality (municipalities with fewer than 30,000 inhabitants or with a population density under 100 inhabitants/km2). A province is considered urban if fewer than 15% of the population lives in a rural municipality. The rest of the provinces are classed as intermediate.
- 2. The provinces dependent on foreign tourism according to this criterion are Alicante, Balearic Islands, Barcelona, Girona, Las Palmas, Malaga, Santa Cruz de Tenerife and Tarragona.
whose forecasts have seen the greatest adjustment; nevertheless, they are still the most dynamic.
Barcelona province and the Community of Madrid concentrate a large part of real estate activity and it is therefore important to look at these in greater detail. The following chart shows the trend in house prices since 2005, together with the model’s forecasts made in October 2019 and October 2020. The correction is obvious, as summarised below:
- In Barcelona province: in October 2019, a 5.5% price increase was projected. Currently, the model predicts a price drop of 1.1% over one year. Expectations have therefore adjusted considerably (6.6 pp).
- In the Community of Madrid: in October 2019, a 6.4% price increase was projected. Currently, the model predicts a very moderate increase in price over one year, namely 0.8%. Expectations have therefore adjusted less than for Barcelona but the change is still significant (5.6 pp).
Finally, it is also useful to look at the trend in the forecasts month by month as this reveals how the model learns and recalibrates its forecasts as new data become available. The following chart shows an initial shift in forecasts in March 2020 with the arrival of COVID-19 and a second shift in June 2020 with the incorporation of Q2 2020 data, indicating the first year-on-year fall in house prices (the target variable for our forecasting models).
These results show the great potential offered by models combining big data with machine learning to predict future trends in Spain’s real estate market. This information is particularly important in a situation such as the present, when high uncertainty regarding developments in the pandemic and its impact on the economy requires us to continually re-evaluate our forecast scenario in order to be able to make more informed and accurate decisions.