Immigration and Swiss House Prices

This study examines the behavior of Swiss house prices to immigration flows for 85 districts from 2001 to 2006. The results show that the nexus between immigration and house prices holds even in an environment of low house price inflation and modest immigration flows. An immigration inflow equal to 1% of an area’s population is coincident with an increase in prices for single-family homes of about 2.7%: a result consistent with previous studies. The overall immigration effect for single-family houses captures almost two-thirds of the total price increase.


Introduction
Recent evidence from country studies on house prices suggests that the impact of immigration on local house prices is a global phenomenon. Saiz (2007) estimates that an immigrant inflow equal to 1% of a city's population results in a 2% increase in house prices for U.S. cities. Gonzalez and Ortega (2009) show that the price effect through immigration is higher for the Spanish housing market. Akbari and Aydede (2009) instead find muted immigration effects for the Canadian housing market. Stillman and Mare (2008) uncover a separation result between migrant groups. They find that the inflows of returning New Zealanders are related to rising house prices but that inflows of new foreign immigrants are not.
A striking feature of these spatial correlations -the correlation between house prices and immigration across local markets -is that they coincide with episodes of high house price inflation and pronounced immigration flows at the national level. Gonzalez  To interpret our short-run estimates that attribute price increases to demand effects through immigration flows, we rely on country specific features of the Swiss housing market. We argue that the structure of the housing 1 The figure 6.3% is from 1983 to 1997 for new single-family homes using the index from the U.S. Department of Housing and Development. market is important for understanding the links between house prices and immigration. On the one hand, nationwide rent control and a low level of home ownership characterize the Swiss housing market. A prior shared by most researchers is that these two market features should lead to moderate house price movements. On the other hand, low vacancy rates and low turnover rates depict the Swiss housing market. These features mean that the tight Swiss housing market is susceptible to local shocks, say through unexpected immigration inflows. This latter channel suggests that the relation between immigration and house prices could be broader than is documented in previous country studies.
Our empirical analysis of the Swiss housing market that exploits the crossregional variation at the annual frequency fits closest to studies by Saiz (2007) and Gonzalez and Ortega (2009). Conditioning on a set of local variables, our estimates find that an immigration inflow equal to 1% of a district's population is coincident with an increase in prices for single-family homes of about 2.7%. The average immigration impact for single-family houses explains almost two-thirds of the total price increase.
The paper is organized as follows. Section 2 outlines the main features of the Swiss housing market. Section 3 presents the empirical methodology. Section 4 discusses the data and descriptive statistics. Section 5 documents the empirical results. Section 6 concludes.

Distinct Features of the Swiss Housing Market
To show that our results are primarily explained by demand shocks in tight local markets, we first outline the main distinguishing features of the Swiss housing market. House price inflation in Switzerland is low by international standards. Table 1 lists the average annual real increase in house prices for 18 OECD countries from 1970 to 2006. The historical record shows that the average real price increase for Swiss housing is 0.34%. This figure is the second lowest among the advanced countries and is seven times lower than the returns for U.S. homes examined in Saiz (2007). 2 Low demand for owner occupancy and nationwide rent control are frequently mentioned as factors explaining the muted growth in Swiss house prices, see Werczberger (1997 Nationwide rent control is a further reason for low house price inflation in Switzerland. Rent increases must be justified by the landlord's cost increases, see Stalder (2003). As such, rent increases do not fully reflect market pressures. Figure 1 shows  A tight housing market is often the consequence of pro-tenant laws.
Tightness of the housing market is observed in low vacancy and low turnover rates. For our period of investigation, the average vacancy rate, measured by the Bundesamt für Statistik, is 1.34% for Swiss rental units compared 3 In fact, taxes discourage owner-occupancy in Switzerland. Property is treated as an asset subject to wealth and income taxes for imputed rental income. Further, unlike other financial investments in Switzerland, housing is subject to capital gains taxes. Capital gains are taxed at the cantonal level with rates differing by duration of ownership.
4 A corresponding rent index at the regional level is unavailable for Switzerland.

5
to 9.7% for U.S. rental units. The tightness of the Swiss housing market is also reflected in low occupancy turnover rates. Wüest and Partner estimate the average stay to be 5 to 6 years for rental units, 12 to 14 years for condominiums, and 20 years for single family homes. 5 In the empirical analysis of section 5, only local information from vacancy rates enters our micro specification. Information on turnover and on home ownership rates is unavailable at the annual frequency. Similarly, the market impact from nationwide rent control is only indirectly captured as an explanation for moderate price movements in Swiss house prices.

Econometric Specification
We estimate the impact of immigrant inflows on house prices at the district level. Our empirical baseline specification follows Saiz (2007)   year fixed effect and X i is a set of control variables, capturing region-specific characteristics. The shock to house prices in region i at time t is ε it .
The specification in first differences assumes that regional fixed effects are filtered out. Still, we are interested in regional indicators that capture immigration into a district raises its local population and thereby the demand for housing. The increase in local demand raises prices and results in a positive β. This positive effect of immigration on house prices also assumes that natives are not infinitely sensitive to changes in housing costs and that native displacement from the local housing market is not complete. One interpretation for this effect offered by Saiz (2007) is that immigrants are less sensitive to housing costs, because local immigrant-specific amenities and networks are more important to them.
An empirical shortcoming of the baseline equation (1)   Establishing causality through an exogenous source of fluctuations in immigration inflows represents an additional concern for OLS estimation of β   (2001), is constructed as follows: (2) The share of immigrants from country c settling in district i in 1997 is denoted by λ 1997 ci . 9 The variable, ∆I ct = I ct − I ct−1 , is the year-to-year change in the national level of immigrants from country c. By summing λ 1997 ci ∆I ct over origin countries, we hope to obtain a predicted measure of total immigrant inflows in district i at time t that is orthogonal to local demand conditions.
9 Munshi (2003) shows that settlement patterns of previous immigrants determine location choices of arriving immigrants from the same country of origin. We construct the instrument with 11 countries of origin: Austria, France, Germany, Italy, the Netherlands, Portugal, Serbia, Spain, Turkey, the United Kingdom, and other.

12
Finally, the instrument is normalized by the population in district i at t − 1.

Data and Descriptive Statistics
The

Estimation Results
In this section, we show that immigration flows are coincident with increases in house prices using price indexes of three different home types. This result is surprising given the low level of house price inflation. We first present baseline estimates based on equation (1) in Tables 3 and 4  In addition to our baseline specification with five regional controls shown in columns 1 to 3, separate regressions are also estimated without regional controls in columns 4 to 6 and with regional fixed effects in columns 7 to 9.
The coefficients of the regional and time controls are not reported in the tables. Heteroskedasticity-robust standard errors are reported in parentheses, while robust standard errors controlling for district clustering are reported in brackets.
The OLS regressions for the three house prices show that the coefficients for immigrant flows lie between 0.361 and 0.914. The price impact from immigration is highest for multi-family homes, followed by single-family homes, then condominiums. This ordering is consistent with the average price increases for the three house types. The regressions show that controlling for regional factors matters. The estimated price impact from immigration is 14 16 highest for the specification without regional controls, followed by the specification with regional fixed effects, and then the specification with regional indicators. Apart from the specification without regional controls, no clear pattern of significance emerges for ( ∆I it P OP it−1 ). Table 4 presents IV regressions for the same specifications shown in Table   3. For all IV specifications, the price effects through immigration are larger than the OLS estimates. This result suggests that the OLS estimates are biased downward due to measurement and endogeneity problems, a finding consistent with Saiz (2007) and Gonzalez and Ortega (2009). The regressions of the baseline specification with regional indicators are in columns 1 to 3.
The coefficient estimates of the immigrant-price effect are significant and range between 1.456 and 3.485, depending on house type. More specifically, an immigration inflow equal to 1% of an area's population is associated with an increase in single-family house prices is 2.7%.
The regressions without regional controls are shown in columns 4 to 6. As in the OLS regressions, regional controls matter in the IV regressions. The significant coefficient estimates tend to be larger than those in the specification with regional controls. This result suggests that our regional controls capture common information across districts, absent in the regressions in 15 17 columns 4 to 6.
Next in columns 7 to 9, we present regressions with fixed effects. Although our specification in first differences should eliminate regional fixed effects, including them reduces concerns about the validity of the instrument in that it allows districts to experience specific shocks. The coefficient estimates are slightly lower with respect to our preferred specification with regional controls in columns 1 to 3. As expected with fixed effects, the standard errors increase such that only multi-family homes remain significant at the 10% level. districts has a p-value of 0.063. An explanation for these results is that the immigration effect is partly compensated by high income growth.
As a further robustness check, we consider whether the 11 largest districts with a population greater than 150,000 influence our estimates. 12 Column 6 shows that the coefficient estimate for ( ∆I it P OP it−1 ) falls to 2.1 in the restricted sample that excludes the 11 largest cities compared to the baseline estimate of 2.7 in the full sample. A χ 2 (6) test with a p-value of 0.017 rejects the null that the immigration effect from the sample without large cities is the same as the baseline estimate. We interpret this result to mean that our baseline 20 estimates are driven by large city dynamics. An explanation for this large city effect is simply that immigrants are more likely to reside in larger districts because these regions offer better job opportunities and amenities. Indeed, over 40% of the total immigrants live in the 11 districts with populations larger than 150,000.
A final check examines whether local tightness in the housing market influences the baseline estimate. Column 7 shows the regression of the baseline specification with local vacancy rates. This added variable is insignificant and has no influence on the baseline estimate of 2.7 for ( ∆I it P OP it−1 ). We interpret this result to mean that the housing market is tight throughout Switzerland and therefore cannot help explain local differences in house prices.
To better understand the price effect from immigration of 2.7 for singlefamily homes, we calculate the average impact from immigration on house prices. First, we consider the average immigrant flows over the 85 districts from 2001 to 2006. This annual average is 0.33% of a district's population.
The overall immigration effect for single-family houses in our sample is 0.33% * 2.7 ≈ 0.99%. This means that almost two-thirds (0.99%/1.52% ≈ 0.60) of the total price increase is attributed to demand effects of immigration. 13

21
This average impact effect from immigration flows is higher for Switzerland than the average estimate of one-third for Spain's boom episode examined by Gonzalez and Ortega (2009).

Conclusions
We find that an increase in immigrant flows equal to 1% of the total population in each district is coincident with a 2.7% price increase in Swiss house prices for single-family homes. The short-run estimates for an environment of low house price inflation are consistent with international evidence found for episodes with higher house price inflation. The results show that rent control and low home ownership rates, distinct features of the Swiss housing market, do not mitigate the house price effect associated with immigration.      Notes: Table 3  unemployed divided by population in region i and time t-1. All estimations include fixed effects by year. Columns 1 to 3 estimate the baseline specification with 5 regional indicators. Columns 4 to 6 show the estimates without regional control variables, and columns 7 to 9 account for regional differences by including regional FE. Heteroskedasticity-robust standard errors in parentheses; clustered standard errors (by region) in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%.    Table 4 displays the 2nd stage of the instrumental variables (IV) relations between changes of immigration and the Swiss house price index. The dependent variables are the annual change in the logarithm of the house price indices, Δ p it , for single-family homes (sfh), multi-family homes (mfh), and condominiums (con). ΔI it / Pop it-1 is the y/y change in immigrants relative to the population in region i at time t-1. Δu it-1 denotes the change in unemployed divided by population in region i and time t-1. In Panel B the first-stage relation is displayed. The instrument SP it is the estimated immigrant change, based on the settlement patterns of immigrants in 1997. All estimations include fixed effects by year. Columns 1 to 3 estimate the baseline specification with 5 regional indicators. Columns 4 to 6 show the estimates without regional control variables, and columns 7 to 9 account for regional differences by including regional FE. Heteroskedasticity-robust standard errors in parentheses; clustered standard errors (by region) in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%.

Sample: 85 MS Regions from 2001-2006
Notes: Panel A of Table 5 displays the baseline relation between changes of immigration and the Swiss house price index. The dependent variable is the annual change in the logarithm of the house price index, ∆ p it , for single-family homes (sfh). ∆I it / Pop it-1 is the y/y change in immigrants relative to the population in region i at time t-1.
(∆I it / Pop it-1 ) high is the y/y change in immigrants in high income regions. ∆u it-1 denotes the change in unemployed divided by population region i and time t-1, ∆ ln y it-1 is the change in the log of per capita income, and ∆ ν it-1 is the change in the home vacancy rate. In Panel B the first-stage relation is displayed. The instruments are the estimated immigrant changes, based on the settlement patterns of immigrants in 1997, SP it , and (SP it ) high respectively. All estimations include fixed effects by year and control for regional effects. Heteroskedasticity-robust standard errors in parentheses; clustered standard errors (by region) in brackets; * significant at 10%; ** significant at 5%; *** significant at 1%. a First stage regression of (SP it ) ; b First stage regression of (SP