Measuring the tail risks to Spanish house prices
The rapid rise in house prices in many European countries during the pandemic has raised concerns about the possibility of a price correction in the coming quarters. Should we be worried in the case of Spain? Given the current macroeconomic scenario, we argue that there is no need for concern. This conclusion is largely due to the good financial health of households as a whole and to reasonable housing affordability in aggregate terms. Neither do we expect an upward spiral in prices: prices may pick up while the economy moves back to its pre-pandemic levels but, in the medium term, we expect house prices to grow in line with household income. We have confirmed this using CaixaBank Research’s new risk model (HaR).
Spain’s real estate market is facing both downside and upside risks. The emergence of a new, more contagious and vaccine-resistant variant of the virus or more severe and persistent bottlenecks in global supply chains than we currently anticipate could cause another slump in activity, with the consequent impact on the housing market. Conversely, greater growth in household credit accompanied by sustained high economic growth could lead to larger house price rises. The questions we attempt to answer in this article are the following: how can we quantify these risks and, should the downside (or upside) risks materialise, how much would house prices fall (or rise) next year?
A traditional way of answering these questions is to examine the historic distribution of real house price growth in Spain (see the first chart).6 Historically, in half the quarters, growth has been equal to or greater than 1.8% (mathematically, this value corresponds to the median of the distribution). If we look at the tails of the distribution, we can see that there has only been a fall in the real price of housing equal to or greater than 8.9% in 5% of the quarters (5th percentile of the distribution), while only in 5% of the quarters is there growth equal to or greater than 13.5% (95th percentile). Assuming this was a typical year, we can conclude that price growth in the following year would most likely be between –8.9% and +13.5%. Depending on which risks materialise, we would move closer to one threshold or the other (and, except in a truly exceptional situation, we would not expect to exceed these boundaries).
- 6. We have used the appraised value of the total national house price published by the Ministry of Transport, Mobility and Urban Agenda and have deflated the nominal house price using the CPI data provided by the National Statistics Institute.
The main limitation of this approach is that it gives us a very wide range of possible values because it does not take into account the current state of the economy. If we were to use this additional information, we could project a more accurate lower and upper threshold for next year’s price growth. In this article, we set out to do just that: using a novel methodology developed by the IMF,7 we have modelled the distribution of house prices at 1 year and 3 years in the future as a function of various key economic indicators (real GDP growth, household credit growth as a percentage of GDP, and housing affordability). In this way, we can narrow down the most probable house prices for a specific timeframe.
- 7. See the «Global Financial Stability Report», International Monetary Fund, April 2019.
that measures the downside and upside risks for the real estate market according to the projected economic scenario
CaixaBank Research’s HaR (House Prices at Risk) model is based on quantile regressions8 that enable us to analyse the impact of different factors on the 5th percentile and 95th percentile of the distribution for the real change in house prices in 1 year and in 3 years’ time. As explanatory variables, we have included a financial factor (household credit growth as a percentage of GDP),9 a macroeconomic factor (real GDP growth) and a factor specific to the housing market (housing affordability, defined as the ratio of house prices to the gross income of the median household). This is used to estimate the impact of each factor (the marginal effect) on the distribution of the change in Spanish house prices for a specific percentile and timeframe (see the chart).10
- 8. The advantage of quantile regressions (compared to traditional methods such as ordinary least squares) is that the marginal effect of factors may differ by quantiles. For example, credit growth may be positively correlated with average house price growth (reflecting the real estate cycle). However, it may have a negative correlation for the 5th percentile; i.e. the fall in price is more severe after a credit boom (the left tail of the distribution moves to the left).
- 9. This variable enters the regressions in a binary format, being equal to 1 in the case of a credit boom and 0 otherwise. We define a credit boom as a period when the percentage of credit to GDP increases by more than 5 pp in 1 year.
- 10. We have standardised the explanatory variables so that the coefficients are comparable with each other.
Here are three examples of how these results should be interpreted:
- A one standard deviation increase (equivalent to 1.7 additional years of income to buy an average home) in the affordability ratio is associated with a 1.6 pp decrease in the 5th percentile of real house price growth at 1 year, all other factors remaining constant. In other words, greater tension in the affordability ratio accentuates the future fall in house prices in the case of an adverse scenario.
- A one standard deviation increase (2.3 pp) in real GDP growth is associated with a 2.2 pp increase in the 95th percentile of real house price growth at 1 year, all other factors remaining constant. In other words, if the economy grows much faster than usual, the increase in house prices is also greater in the event of a favourable scenario.
- A credit boom leads to a 2.6 pp decline in the 5th percentile of real house price growth at 3 years compared to not having a credit boom, all other factors remaining constant.
As expected, an increase in real GDP growth is associated with greater house price growth, while a higher housing affordability ratio leads to lower house price growth. An interesting result is that, when there is a boom in household credit to purchase housing, the 5th percentile increases at 1 year but decreases at 3 years. One possible explanation is that an increase in credit affects house prices through two channels. On one hand, in the short term the entire distribution moves to the right: the increased flow in credit boosts demand for housing and pushes up prices. On the other hand, a credit boom increases the likelihood of a real estate bubble (as in the 2008 crisis), which means that, in the medium term and if the bubble bursts in the medium term (adverse scenario), the price correction will be much greater.
but increases the risk of a further price correction in the medium term, as seen in the 2008 crisis
The HaR model allows us to see the evolution over time of the 5th percentile and 95th percentile for the distribution of the growth in Spanish house prices at 1 year, conditional on the values of the explanatory factors (economic situation) at each point in time. In the chart below, the grey line and red line mark a range within which we would expect house prices over the coming year to lie.11
- 11. The results of the model are based on two assumptions: (i) the three factors included in the regressions are the main determinants of the tails of the distribution for Spanish house price growth and (ii) the historic relationships between the variables will remain valid in the future.
the HaR model provides us with much more accurate ranges within which the following year’s prices will oscillate.
Some of the results of the HaR model are striking. Firstly, if we compare the percentiles obtained using the model (grey line and red line) with the historic percentiles (dotted lines), we can see that they differ substantially: the model’s percentiles give us a much more precise range within which prices will oscillate over the coming year. For example, during the 2008 crisis, the model tells us that house prices should fall because of the economic situation observed, information not provided by the historic percentiles. Secondly, during the gestation period of the housing bubble prior to the 2008 crisis, the model was continually subject to upside surprises as the economic situation (in particular, the tensions observed in affordability) suggested that growth in prices would moderate. The same thing happens with the COVID-19 crisis: the model found it surprising that house prices were so resilient despite the slump in GDP.
The most interesting information comes when we analyse the current situation (with data up to Q4 2021) and the future prospects of the real estate market. The model tells us that, even if the downside risks were to materialise, it would be highly unlikely that real house prices would fall by more than 3.9%. Although this lower threshold would imply a considerable decrease decrease, if compared with the historic lower threshold (8.9%), we can see that the coming year should bring relatively moderate downside risks. On the other hand, even if the upside risks were to materialise, the model tells us that it would be highly unlikely that real house prices would rise by more than 4.8%. Looking ahead, and based on CaixaBank Research’s central scenario forecasts, the model’s percentiles indicate that downside risks are contained (the 5th percentile even reaches 0 in 2022) due to the significant GDP growth we expect next year. The upside risks are somewhat greater but limited: as GDP growth returns to normal in 2023, and thanks to the absence of a credit boom (unlike in the pre-crisis period of 2008), price rises should also be moderate.
neither a severe price correction nor an upward spiral are likely
This result is consistent with the assessment recently issued by the Bank of Spain12 on the situation of the country’s real estate market. In particular, house price imbalance indicators suggest that prices are above but nevertheless very close to their equilibrium levels. This diagnosis for Spain contrasts with the situation of real estate markets in other European economies, as the ECB has noted in its latest stability report.13 Specifically, the ECB warns that the risks of price corrections in the medium term have increased substantially due to the overvaluation of house prices in some countries and to the fact that credit standards have been relaxed, expressing some concern about the emergence of a debtdriven real estate bubble. Given this situation, the ECB recommends that those countries in which such vulnerabilities are emerging should consider the option of gradually adjusting some of the macroprudential policy instruments at their disposal. However, we believe the situation of the Spanish real estate market does not warrant such instruments to be implemented in our country, at least in the short term.