The Economic Impacts of Immigration: A Look at the Housing Market

Transcription

1 The Economic Impacts of Immigration: A Look at the Housing Market Honors Senior Thesis Moises Yi Advisor: Prof. David Card Department of Economics University of California-Berkeley May 2008 Abstract This paper uses data from the 1990 and 2000 Decennial Censuses to study the impacts of immigration on local housing markets. Specifically, I use matched tract-level data from the two Censuses to study the effects of changes in the number of immigrants on rental prices in a tract. Without other controls, the results indicate that tracts with larger inflows of immigrants between 1990 and 2000 had slower rental price growth, with evidence of differential impacts for Hispanic and non-hispanic immigrants. However, once controls are added for the changing socioeconomic composition of tracts, the direct effects of immigrants fall in magnitude and become economically and statistically insignificant. These findings are consistent with previous research which argues that immigrants effects on the local housing market have to do more with their socioeconomic characteristics than with the fact that they are immigrants. Acknowledgements: I would like to thank Professor Card for all his guidance and support throughout these past few years. I would also like to thank my friend Monica Deza for her advice and help with STATA. I would not have been able to do any of this without their help, patience and encouragement. I would also like to acknowledge the financial support from the Summer Undergraduate Research Fellowship, the Pergo Foundation, and the McNair Scholar Program.

2 I. Introduction The potential economic impacts of immigration on the native population have been a contentious and much-debated issue in the U.S. for over a century. 1 Concerns over immigration have been heightened in the past decade by the rise in levels of legal and illegal immigration to rates not seen since the beginning of the 20 th century (e.g., Borjas, 1999). Most of the economic literature on immigration focuses on measuring its impact on the labor market. Relatively few papers discuss how immigrant influxes affect the housing market, even though housing values are important determinants of individuals consumption patterns. The few papers that study the relationship between housing and immigration (Saiz 2001, 2003; Saiz & Wachter 2006) have proceeded by classifying immigrants into a single homogenous group, regardless of nationality. It is well-established that immigrant skills largely depend on the immigrants national origin (Borjas 1987), and that because of this the economic impact of immigrant influxes will vary depending on where immigrants come from. Using tract-level data from the 1990 and 2000 U.S. Censuses, this paper tries to differentiate immigrants into groups based on their national origin, and focuses on how these distinct immigrant groups have different effects on the housing market. In particular, ordinary least squares (OLS) regressions are used to estimate the effect of immigrant inflows over the period on renter-occupied housing 2 prices in U.S. metropolitan areas. 1 For example, the economist and statistician Francis Walker (president of the American Economic Association from 1885 to 1992) wrote a famous article arguing that the U.S. should restrict immigration in (Francis A. Walker, Restriction of Immigration, Atlantic Monthly, June 1896). Throughout this paper, the term natives refers to American born individuals, regardless of their race or ethnicity. The U.S. Census Bureau uses the term native to refer to anyone born in the United States or a U.S. Island Area such as Puerto Rico, or born abroad of a U.S. citizen parent. 2 Renter-occupied housing is defined by the U.S. Census Bureau as all occupied units which are not owner occupied, whether they are rented for cash rent or occupied without payment of cash rent. For the rest of this paper, the terms renter-occupied housing prices and rental prices will be used interchangeably. 2

3 Simple univariate models indicate that Hispanic immigrants have a raw negative and significant impact on rental prices. These effects vary among different regional groups of Hispanic immigrants, and are different from those other (non-hispanic) immigrants. However, once controls are added for some of the immigrants socioeconomic characteristics (such as education and income level) the negative effects of immigration become statistically insignificant. These results suggest that immigration by itself does not affect housing or rental prices, but rather that it is the socioeconomic characteristics of certain groups of immigrants that make neighborhoods less attractive and hence cause a decrease in the value of housing. This paper is organized as follows: Section 2 contains a review of previous literature on the subject of immigration and the importance of housing prices in consumption patterns. It also describes the previous research done on the links between housing prices and immigration. Section 3 is a description of the data, methods and the models used in this paper. Section 4 is a discussion of the results of the regressions at a national level (all MSAs in the U.S.). Section 5 describes the regression results for five selected cities. Section 6 talks briefly about the Owneroccupied housing market. Section 7 presents the conclusion. II. Literature Review Previous economic studies of the effects of immigration have mainly focused on labor markets. The existing literature on immigration consists of papers dealing with topics ranging from assimilation rates (Borjas 1990, 1995; Card 2005) and the geographic diffusion of immigrants (Card & Lewis 2005) to the demographic characteristics of immigrants (Passel 2005) and the fiscal effects of immigration (Lee & Miller 2000). A relatively large number of studies have been devoted to the labor market effects of immigration, specifically in terms of 3

4 employment and wages for natives (e.g., Borjas 1995, Card 1990, Orrenius & Zavodny 2006, Toussaint-Comeau 2007). Some authors argue that immigrants economic impact is detrimental to native workers (Borjas 1985, 2004) because immigrants increase the relative supply of less skilled workers in the economy, driving down the wages of less skilled natives. Yet other studies, such as Peri (2006), argue that immigration produces no harm to most groups of native workers, and that in fact immigrants have helped bolster natives wages over the past decade and a half. In examining how immigrant inflows affect natives, however, few studies have focused on the relation between immigration and housing market fluctuations, even though the importance of the housing prices to natives and aggregate welfare is considerable. Case, Quigley and Shiller (2006) found that at a macro level there was a strong correlation between housing wealth and consumption in developed nations. They also concluded that housing market variations were more important than stock market variations in determining consumption. Studies conducted by Iacoviello (2004) and Campbell and Cocco (2005) arrived to similar results. Iacoviello (2004) used quarterly U.S. data from the Federal Reserve Economic Data (FRED) to show that, because they are correlated with borrowing capacity, housing prices act as a driving force of consumption fluctuations. Similarly, Campbell and Cocco (2005) conducted a study using micro-level data in the U.K. and found empirical evidence suggesting that an increase in housing prices had a positive effect in the consumption patterns of specific demographic groups, with the largest effect observed for older homeowners and a negligible effect observed for young renters. The conclusions of Campbell and Cocco (2005) differ from those of Li and Yao (2004) and Bajari et al (2003) who found a negative to negligible effect of housing prices increases to younger individuals and at an aggregate level. 4

5 In a later study, Tsharakyan (2007) showed that when binding credit constraints are added to the Bajari et al model housing prices again have a positive effect in welfare. Tsharakyan concluded that the housing price appreciation in led to per household improvement of aggregate welfare by an amount equivalent to about 40% of mean household income in Despite the evidence suggesting an important role of housing prices in aggregate welfare, relatively few studies on immigration take the housing market into account. Dowell and Seong (1996, 1998) conducted studies on immigrants homeownership rates and residential overcrowding in immigrant communities. A more recent study by Ottaviano and Peri (2006b) focused on the housing market as a new approach to measure the overall impact of immigration on the economy. In their study, Ottaviano and Peri conducted a study using data for different U.S. states and concluded that housing prices act as transfer mechanism between immigrants (who tend to rent) and natives (who own the vast majority of the housing stock). When taken into account, these transfers produce an overall increase in the wellbeing of natives, even if wages for unskilled native workers are negatively affected by immigration. Ottaviano and Peri s results underscore the importance of housing values in determining the economic effects of immigration. It is for this reason that this paper will focus on the relation between immigrant inflows and housing prices. Susin (2001) and Saiz (2000, 2003) are among the few studies that deal with direct links between immigrant inflows and housing market fluctuations. Susin (2001) and Saiz (2000) focused on the impact of the Mariel Boatlift on rental prices in Miami, relative to trends in other metropolitan areas. Their studies concluded that, in the short run, the immigrant influx in 1980 resulting from the Mariel Boatlift had a positive impact on the rental market of Miami relative to 5

6 other metropolitan areas of similar characteristics. Similarly, using more extensive data from the 1970s to 2000, Saiz (2003) showed that there is a positive impact of immigration on rental and housing prices in metropolitan areas across the U.S. Greulich, Quigley and Raphael (2004) arrived at a different conclusion. They point out that even if housing prices increase as a result of immigration rental burdens the actual share of an individual s income allocated to housing are not affected because of wage increases for natives and because immigrants consume different types of housing. These results are reconciled by Card (2007) who finds that the effect of immigration on rental prices found by Saiz (2003) is comparable to his estimates of the positive impact of immigration on average wages. According to Card, this provides an explanation as to why the rental burdens remain unchanged among natives after immigrant inflows. In a more recent paper, Saiz and Wachter (2006) argue that the growth in housing prices experienced by cities that receive immigrants (such as documented in Saiz, 2003) is simply the consequence of an upward slopping housing supply curve. Saiz and Wachter argue that a more relevant issue is whether, within metropolitan areas, prices in the neighborhoods where immigrants settle grow faster than they would have otherwise. Using localized data (at the Census tract level) and controlling for MSA-by-year fixed effects, they ran regressions using the model: to estimate the impact of immigration (measured as the percentage of foreign born in each tract: 1-φ i,m,t ) on average housing prices P i,m,t, controlling for housing traits (Z i,m,t ) and socioeconomic characteristics of the neighborhood (X i,m,t ). They also controlled for many other variables such as environmental quality and home value growth in the previous decade, showing 6

7 these variables did not affect the results. More importantly, by using a Gravity Model of Immigrant Residential Choice, Saiz and Wachter were able to determine that reverse causality (immigrants moving to already cheap neighborhoods instead of immigrants causing housing prices to decrease) accounted only for 15% of the impact they cause. Even when accounting for reverse causality, Saiz and Wachter found evidence that within metropolitan areas, housing values grow relatively more slowly in neighborhoods with larger numbers of immigrants. These negative effects were larger in non-hispanic white neighborhoods, than in neighborhoods with high concentrations of minorities. Saiz and Wachter point out that this effect might have to do more with certain socioeconomic characteristics of the immigrant population than with the fact that the individuals moving in are immigrants. In their paper, Saiz and Wachter argue that immigrants education and race, among other factors, are likely to make predominantly immigrant neighborhoods less attractive to natives. These characteristics, however, are directly correlated with the national origin of the immigrants in question. Borjas (1987) documented that national origin has an important role in determining the impacts of immigrants to the U.S., given that their socioeconomic characteristics greatly vary depending on where they come from. Borjas found immigrants from developed Western countries tend to fare much better than immigrants from developing countries in the labor market. Saiz and Wachter (2006) found that this applies to housing markets too, noting that the effect on the housing market of immigrant groups from different regions is heterogeneous. Their discussion of how national origin affected immigrants impact on housing markets is very limited, mainly due to the lack of national origin information in their data. In 7

8 order to address this issue, this paper will focus on how the housing market in American cities 3 is affected by immigrants of different nationalities. In particular, this paper will focus on how the impact on rental prices differ among recent Hispanic immigrants of different regions (South Americans, Central Americans and Mexicans), and how this impact compares to the market impact of other groups of non-hispanic immigrants. III. Methods 3.1 Data Two separate cases were studied in this paper. The first case deals with all cities 4 in the U.S. and only aims to distinguish immigrants into two groups: Recent Hispanic Immigrants and Other immigrants. The second case is more focused. It studies 5 medium-sized cities. These cities were picked because their boundaries did not change dramatically during the 1990 s, which is important for reasons that will be explained later in this paper. The cities in question are: Fresno, CA; McAllen-Edinburg-Mission, TX; Naples, FL; Raleigh-Durham-Chapel Hill, NC; and Sioux City in Iowa. In this case, specific nationalities are allocated for all immigrants. As with Saiz and Wachter (2006), Census data for 1990 and 2000 from the Neighborhood Change Database (NCDB) form the primary source of data for this paper. The NCDB is constructed from U.S. Census data and divides the U.S. in small geographical units called Census Tracts. Each Census tract has a population of approximately 4000, and they have been frequently interpreted by researchers as geographical measures of neighborhoods. Individual microdata from the Integrated Public Use Microdata Series (IPUMS) is used as a secondary 3 According to Card (2007), metropolitan areas are disproportionately affected by immigrant inflows. The average share of immigrants in the 17 largest metropolitan areas is about twice the level as the nation as a whole. 4 Cities in this paper refer to Metropolitan Statistical Areas (MSA) as defined by the U.S. Census Bureau 8

9 source, mainly to obtain national origin information for immigrants in each PUMA (a larger geographical division containing about 20 census tracts). Additional data from the Census 2000 Summary File 3 (SF3) Sample Data is employed to obtain tract-level national origin figures for immigrants (only for the year 2000). The NCDB dataset contains information about the number of immigrants living in each census tract, but it does not feature information regarding their national origin. This was a major problem for Saiz and Wachter (2006) and one of the reasons why national origin was not discussed in depth in their paper. To account for this problem and better estimate the impact of different groups of immigrants on housing prices this paper uses a different approach. First, a different variable (from the NCDB dataset) is used as a proxy for number of Hispanic immigrants in each tract. The NCDB features a language variable for each census tract that shows how many people in such tract speak Spanish and speak English not well or not at all. In this study, this variable is used as an indicator of the number of recent Hispanic immigrants in each tract. To check how well this proxy measures the number of recent Hispanic immigrants, I conducted three evaluation exercises. In the first exercise, I used 2000 Census microdata from the IPUMS to estimate the fractions of recent Hispanic immigrants in each of 55 randomly selected public use microdata areas (PUMA s) the smallest geographic unit that is identified in the public use files. I then combined Census Tract data from the NCDB to estimate the fractions of people in each PUMA who Speak Spanish and speak English not well or not at all (using the available key that maps Tracts to PUMA s). A comparison between these two estimates is shown in Figure 1 in Appendix A. The figure suggests that the two estimates are highly correlated. 9

10 In the second exercise, I used the IPUMS data to estimate the fraction of recent Hispanic immigrants in 45 randomly selected MSA s/pmsa s. Again, I compared this estimate to the fractions of people in each MSA/PMSA who Speak Spanish and speak English not well or not at all, derived by totally all the Tracts in the MSA/PMSA. As shown in Figure 2 of Appendix A, the two series are very highly related. Finally, using tabulations at the tract level from another data source the Census SF-3 file, I compared the fractions of recent Hispanic immigrant in each of 502 tracts used in the 5- city case study against the fraction of residents of the tract who Speak Spanish and speak English not well or not at all from the NCDB. Again, the plot shows that the estimated and actual numbers of Hispanic immigrants are highly correlated. Taken as a whole, I conclude that the evidence supports the conclusion that the NCDB-based count of people who Speak Spanish and speak English not well or not at all is a good approximation, for each specific tract normalized to 2000 boundaries, of the number of recent Hispanic immigrants living in each tract 5. Based on this conclusion, for the national case study (which is based on all MSAs in the U.S.), I construct an estimate of the fraction of recent Hispanic immigrants for each tract based on the number of people whose first language is Spanish and who have limited English ability. I also use direct counts of the number of total immigrants in the tract (available in the NCDB). This study is based on a total of 50,819 tract-level observations in this case. For the 5-city case study I use a more detailed immigrant classification, based on PUMAlevel data on immigrant origin groups. The 5 cities selected for this part of the paper were chosen specifically because they did not experience substantial tract boundary changes from 5 Appendix-A illustrates some of the work done to verify these assertions. 10

11 1990 to This allowed the allocation of specific nationalities to the population of Hispanic immigrants using IPUMS data at the PUMA level. Each PUMA contains about 20 tracts. For the 5-MSAs case, national origin distribution of Hispanic immigrants was obtained for each PUMA and then allocated to all the tracts within the PUMA s boundaries using the MABLE geo-correlation engine. Because the PUMA data was not normalized to 2000 tract boundaries, while all the other data from the NCDB was, it was important that the tract boundaries in the MSAs studied did not change dramatically. Otherwise, nationality allocation for tracts in 1990 would not have been possible. After allocating specific nationalities for immigrants in each tract, national groups with relatively few members were assigned to regional groups (i.e. South Americans, Central Americans and Mexicans). There were 502 tract-level observations for this case. 3.2 Models a. Nationwide Case For the first case, which includes all cities in the U.S., the number of observations is substantial (50,819 neighborhoods). As stated before, immigrants are not assigned specific nationalities and instead are divided in two broad groups recent immigrants from Latin America 6 and Other Immigrants. The original number of tracts (50,819) was reduced to 41,187 by eliminating tracts at the upper and lower ends of the distribution with respect to median rental growth. These cases are considered to be exceptional and as such could affect the final results in undesirable ways. Thus, tracts below the 5 th percentile and above the 95 th percentile were eliminated. In the case of median housing prices, tracts where median prices 6 For the rest of this paper, the terms Immigrants from Latin America and Hispanic Immigrants will be used interchangeably. 11

12 grew by less than 2.39% or more than 92.54% were thrown out as well. Descriptive statistics are shown in Appendix Table 4. The data was then used to estimate the impact of both population growth and immigrant share growth on the growth of median rental prices. The following models were employed to carry on the appropriate regressions: (1) (2) (3) (4) i and T are subscripts for census tract and year. Rent i,t represents Median 7 Rental prices and Pop i,t is the population for each tract i in year T. Imm i,t is number of immigrants (of all nationalities) living in each neighborhood. MSA i are dummies for each city. The independent variable in the four models above is, therefore, the percentage change in median rental prices in each tract. % Imm i,t is given by the equation so it represents the percentage growth of immigrant population with respect to the 1990 tract population. Because they way the model is constructed, α will capture the effect of overall population growth, while β will capture the effect of increased share of immigrants in the population net effect of actual increases in overall population. This will allow the model to 7 Median prices were chosen instead of average prices to avoid having the upper and lower tails of the housing price distribution affect the results. 12

13 single out the effect immigrants have on rental prices irrespective of growth in population (and hence in housing demand). The X i,t variable in Model 3 represents three socioeconomic controls: average family income, poverty rate and number of children in the tract. These were chosen for the Model 3 because they consistently appeared to have large effects on rental prices. The Z i,t variable in Model 4 includes additional socioeconomic controls for tract-time specific data on the share of retired people living in the tract, unemployment rate, vacancy rate, and numerous variables on education level of the tract. Because Saiz and Wachter also found that adding controls for environmental quality and home value growth in the previous decade does not significantly change the results, controls for these variables were omitted. To differentiate the impact of Hispanic immigrants and Other immigrants, the following models were used: (1) (2) (3) These three models feature the same variables from the first set of models. The only difference lies on the definition of the immigrant variables. Instead of having an Imm i,t variable for all immigrants living in each tract i, now immigrants are separated into two groups. Thus, HispImm i,t represents the number of Hispanic Immigrants, while OtherImm i,t represent the number of non-recent Hispanic immigrants living in in each tract. X i,t and Z i,t are socioeconomic variables and MSA i,t are dummy variables for each MSA. 13

14 b. Five-MSA Study For the second case (of only five medium sized MSAs), the number of observations was 502. As in the nationwide case study, observations in the lower and upper tails of the distribution of growth in rental prices (below the 5 th percentile and above the 95 th percentile) were removed to prevent them from affecting the results. 82 observations were removed, leaving 412 tracts. Tracts where the price increase (from 1990 to 2000) was less than 6.4% or more than 100%, were thrown out. Descriptive statistics for this case can be found on Appendix Table 5. Models similar to those used in the nationwide case were employed here, with two important adjustments. First, in this case immigrants are actually assigned specific nationalities. Because the limited number of observations, immigrants where clustered in regional groups (i.e. South Americans, Central Americans). Secondly, because the number of observations is relatively small (compared to the nationwide case), a smaller number of control variables was used in model 3. These are the three models used: (1) (2) (3) As in the nationwide case models, in this case i is a subscript for tracts and t is a subscript for year. Model 1 and Model 2 are the same models used in the nationwide case, except that immigrants are now divided into 4 groups: Mexican immigrants ( M i,t ), Central American immigrants ( C i,t ), South American Immigrants ( S i,t ), and Other Immigrants 14

15 ( O i,t ). Model 3 is different from the one in the nationwide case in two ways: immigrant group variables are divided according to national origin, and the fact that the Z i,t variable now includes less control variables 8 to account for the smaller number of observations. As in the nationwide case, controls for environmental quality and previous home value growth are omitted. IV. DATA ANALYSIS All U.S. Cities The sample in this case consisted of the observations from all tracts in metropolitan areas in the U.S. Table 1 shows the results obtained in a regression where the dependent variable was Growth in Median Rental Prices and the only independent variable was population growth. The results show that with no controls, population growth is positively (and significantly) correlated to positive growth on rental prices. Intuitively, this makes perfect sense, since increases in population will lead to an increase in demand for housing units (which include rental units), and this increase in demand will in turn push housing prices up. Model 2 adds an Immigrant variable to Model 1. Besides a control for population growth and MSA dummies, no other control variables are included. The population growth control in this model is important as it captures the effect that higher population, (which can be the result of inflows of either natives or immigrants) and thus allows the immigrant coefficient to capture the net effect of immigration. Without controlling for overall population growth, the immigrant coefficient would be certainly positive, since immigrant inflows would lead to higher population and hence higher demand for housing. The results obtained in Model 2 indicate that immigrants (as a whole) have a significant negative effect on the growth of rental prices. The coefficient of indicates that a 1% 8 Specifically, several controls for education were left out. 15

16 growth in the immigrant population (with respect to the 1990 tract population) is correlated with a decrease in the growth of rental prices by %. Though this coefficient is small in magnitude, it is highly significant. Adding three socioeconomic controls in Model 3 (average income, poverty rate and share of children) produces interesting results. Population growth continues to have a positive and statistically significant impact on rental prices growth. The All Immigrants coefficient, however, while still negative becomes statistically insignificant. The fact that the All Immigrants coefficient is insignificant is evidence that, if income and poverty rate are held constant, immigrants and natives are no different in the way they affect rental prices. Model 4 was used to run a regression of growth in rental prices against population and immigrant share growth, while controlling for several socioeconomic characteristics (in addition to the controls of Model 3, Model 4 included controls for education, retirees, and unemployment rate in each tract). The results of Model 4 closely follow those from Model 3. The main difference here is that the coefficient for population growth in this case is statistically insignificant. The All Immigrants coefficient in this case remains statistically insignificant. As a result of the way Model 4 was constructed, the population growth variable captured the effects of population growth, leaving the immigrant variable to estimate the effect that immigrants moving into a tract (net of how much population changed in such tract) affected the growth in rental prices. In short, β is supposed to capture how immigrant presence in certain tracts affected rental prices. The results obtained with this model (shown on Table 1) indicate that, when controlling for several socioeconomic variables, the impact of population growth caused by natives and the impact of population growth caused by immigrants are not significantly different from each other. An F-test carried to check if the coefficient for 16

17 population growth and immigrant growth were statistically equal failed to reject the hypothesis that they were equal to each other (F(1, 40842)=0.69, Prob>F =0.4078). These results are consistent with those of Model 3. Immigrants moving into a neighborhood affect rental prices the same way that natives moving into a neighborhood would. Two important conclusions can be derived from the results of Table 1. First, controlling for socioeconomic characteristics of immigrant populations greatly affects the estimates of how immigrant inflows affect rental prices. The second important conclusion is that growth in immigrant populations does not seem to affect growth in rental prices when the socioeconomic characteristics of immigrants are controlled for. While useful, these insights highlight the problem with grouping all immigrants in a single category. If socioeconomic characteristic controls affect our estimates for the immigrant coefficients, at the very least this shows that these characteristics are important determinants of how different groups of people (in this case immigrants and natives) affect rental prices. The problem is that clustering all immigrants into a single group obviates the fact that, in terms of socioeconomic characteristics, immigrants to the U.S. are not a homogenous group. To account for this, the All Immigrants group was divided into two groups in the following regressions. More elaborate grouping was not possible because of limited data on the nationality of immigrants at the national level. Table 2 shows the results for regressions aimed to estimate the impact of Hispanic Immigrants and Other Immigrants on housing prices growth, as well as how these two groups impacts on rental prices compare to one another. The results of these regressions are shown on Table 2. As in Table 1, Model 1 is employed to conduct a regression with growth in rental prices as the independent variable and growth in the share of Hispanic and Other immigrants as the dependent variables, while 17

18 controlling for population growth and including MSA dummies. Model 2 is similar to Model 1, except that it adds controls for 3 socioeconomic variables: average income, poverty rate, and share of children in each tract. Model 3 includes more controls than Model 2: education, unemployment rate and retirees, among others. With no socioeconomic controls, Model 1 shows that Hispanic immigrants have a raw negative and statistically significant effect on rental prices. The Hispanic immigrant coefficient of indicates that a 1% growth in the Hispanic immigrant share of the (1990) population is correlated with a decrease in the growth of rental prices by %. The coefficient for Other Immigrants is negative but statistically insignificant. F-tests carried out to test the coefficients failed to reject that hypothesis that the Other Immigrants coefficient and the coefficient for overall population growth were the same (F(1, 40853)=2.41, Prob>F =0.1206). The same F-test rejected the hypothesis that the Hispanic Immigrant and the population growth coefficients were equal. These results seem to indicate that Other Immigrants are no different from natives in terms of the impact they have on rental prices, while Hispanic Immigrants do have a different impact on rental markets (compared to natives) that is both negative and strongly significant. The results also seem to indicate that the negative coefficient found using Model 2 in Table 1 (where immigrants were treated as a single group) is driven mostly by Hispanic immigration, and not by the Other Immigrants group whose coefficient is insignificant. Adding the three controls in Model 2, does not change the Other Immigrants coefficient drastically, as it remains negative and statistically insignificant. The coefficient for Hispanic Immigrants, on the other hand, becomes statistically insignificant. This result is consistent to that of Table 1. When controlling for income and poverty rate (and number of children) in each tract, immigrant inflows (both Hispanic and non-hispanic) have a negligible effect on the rental 18

19 market. Moreover, the coefficients for the two immigrant groups are not statistically different from one another (an F-test rejected the hypothesis that the coefficients were equal to one another (F(1, 40850)=1.76, Prob>F =0.1848)). Of particular interest is what occurs in Model 3. Once more controls are added (for characteristics such as education and unemployment rate), the coefficients for Hispanic immigrants becomes positive and significant. This coefficient of indicates that a change of 1% in the growth of the share of immigrants in a neighborhood is related to a % increase in the growth median rental prices. While this effect is small in magnitude, it represents a dramatic turnaround from the statistically significant and negative coefficient for the same variable in Model 1. While the coefficients for Hispanic Immigrants changes drastically when socioeconomic controls are added, the same cannot be said of the Other Immigrants coefficient. The coefficient for this group of immigrants is consistently statistically insignificant, regardless of the added socioeconomic controls of Model 2 and Model 3. It could be the case that non- Hispanic ( Other ) immigrants closely resemble natives and so their impact on the rental market is captured by the population growth coefficient. The relatively greater variation in the coefficient of Hispanic Immigrants that occurs when socioeconomic controls are added may point to the fact that socioeconomic factors might have more to do with the effect of Hispanic Immigrants on the growth of housing prices than they do with that of Other Immigrants. This could happen because Hispanic immigrants socioeconomic characteristics (such as low level of education, for example) might be viewed more negatively by natives than characteristics of Other Immigrants. 19

20 This case study at the national level produced several important findings. First, the effect of immigrants as a whole on rental prices (without controlling for several socioeconomic factors) is statistically significant and negative. Dividing immigrants in two groups shows that this raw effect on rental prices varies among Hispanic Immigrants and Other Immigrants. Moreover, from the results it can be concluded that the negative impact on rental prices that immigrants have when treated as a single group is caused by Hispanic Immigrants, given that the Other Immigrants coefficient is insignificant. Secondly, when controlling for socioeconomic variables, immigrant inflows have a negligible effect on the rental market which is not different from that caused by general population growth. Third, when differentiating between Hispanic and non-hispanic immigrants, the coefficients for these two groups are not statistically different from one another once socioeconomic factors are taken into consideration. These results indicate that it is not immigration per se that affects rental markets, but rather the socioeconomic characteristics of people who move into neighborhoods. The findings are also evidence that Other (non-hispanic) immigrants are less likely to affect rental prices differently from natives, possibly pointing to the fact that non-hispanic immigrants more closely resemble natives in terms of socioeconomic characteristics than Hispanic immigrants. However, the broad definition of these two immigrant groups makes it difficult to assess the validity of this theory. This problem will be dealt with in the next section of this paper, when more specific data on individual nationalities is available at a smaller scale. V. DATA ANALYSIS The 5-MSAs Case Using the models described in section 3.2.2, regressions were carried using the percentage change Median Rental Prices as the dependent variables. As discussed in the 20

21 methods section, 5 medium size cities were chosen: Fresno, CA; McAllen-Edinburg-Mission, TX; Naples, FL; Raleigh-Durham-Chapel Hill, NC; and Sioux City in Iowa; and immigrants were clustered in five groups: Mexican, Central American, South American and Other immigrants. Regressions were carried out both for each city individually and for the five cities together. The discussion of the results follows. Table 3 shows the results for regressions carried for the five cities together. When using no controls (other than MSA dummies) in Model 1, there is large heterogeneity with respect to the coefficients of the different immigrant groups. Central American immigrants have a negative but insignificant coefficient. South Americans have a positive but again insignificant coefficient. Mexican immigrants have a negative coefficient that is also statistically significant (-.3794, t- stat=-3.04). And the Other Immigrants group has a positive statistically significant coefficient. As in the nationwide case study, when socioeconomic controls are added in Model 2 and Model 3, the coefficients for all groups become insignificant (at a 5% confidence level). Similar regressions were carried with each city individually. The results of such regressions yielded similar results. In every case, once socioeconomic controls were added in Model 2 and Model 3, the coefficients of all immigrant groups became insignificant. These results are consistent with Saiz and Wachter findings about the importance of immigrants characteristics (such as education or income level) when determining their impact on housing markets. Another important question is whether the impact that immigrants have on rental prices varies across different groups of immigrants. Carrying an F-test using the information of Model 1 to check whether all immigrant coefficients were the same produced the following results: F( 3,403)=3.96, Prob>F =0.0084, which indicate that the coefficients of the four immigrant groups are 21

22 indeed statistically different from each other. This is not surprising since Model 1 does not include any controls for socioeconomic variables. Using the information obtained through this Model 3, an additional F-test was used to determine whether the impact of different groups of immigrants varied according to where they came from (this time controlling for socioeconomic characteristics). An F-Test to check if the coefficients for the different Hispanic immigrant groups were the same, failed to reject the hypothesis that they are all equal to each other (F( 2,395)=1.00, Prob>F =0.3681). Similarly, an F- test to check whether the coefficients for all immigrant groups (including non-hispanic immigrants) were equal to each other produced results that indicate that the coefficients for all four immigrant groups are not statistically different from one another (F( 3,395)=0.74, Prob>F =0.5307). An F-test of joint significance for all the immigrant variables failed to reject the hypothesis that the immigrant variables are jointly insignificant ( F( 4,395) =1.42, Prob > F=0.2265). This is consistent with the results of Model 4 in Table 1 of the nationwide case. Immigrants (as a whole) appear to have an insignificant effect on rental housing markets (when controlling for several socioeconomic factors). Two important findings are worth highlighting. First, the results in this localized casestudy seem to be consistent with the results obtained in the nationwide case. Immigrants as a whole have statistically insignificant effects on the rental market once their socioeconomic characteristics are taken into account. The second important result is that, while different groups of immigrants have different effects on the rental market when no controls are added; these effects become statistically similar once socioeconomic factors are controlled for. 22

23 Intuitively, these results can be explained as follows: a European immigrant and a Salvadoran immigrant moving into a neighborhood will have different impacts on the housing market. However, this difference arises not from the fact that one immigrant is from Europe while the other one is Hispanic, but from the fact that the European immigrant will more likely be better educated or wealthier, and possesses certain characteristics that are attractive to other people in the neighborhood. Thus, an influx of European immigrants would likely make rental prices go up as the neighborhood they move to becomes more desirable for prospective renters. This would explain the results of Model 1. However, if these socioeconomic factors are held constant, then being an immigrant from Europe or from El Salvador does not really make a difference in terms of how rental prices are affected. This would explain the results from Model 3. While the results obtained here are more difficult to generalize than those from the nationwide case, many new things can be learned from this five-msa case. The main advantage of this case with respect to the study at the national level is the ability to further differentiate immigrants by nationality. While the results in the nationwide case study show that when controlling for socioeconomic variables the impact of Hispanic immigrants are positive and statistically significant, the five-msa results show that the effects of different groups of Hispanic immigrants on rental prices vary depending on the country/region they come from. For example, Hispanic Immigrants have a positive and statistically significant effect on rental prices (Model 3 of Table 2 ). However, when Mexican immigrants are treated as a separate group in Table 3, they have a negative (although insignificant) coefficient. The results obtained in this section indicate that Hispanic immigrants, far from being homogeneous, are a diverse group with respect to the effects they have on the housing market. 23

24 The relatively small number of observations in this case did not allow for a more in depth study of different specific nationalities. It would have been nice to add more controls and compare more diverse nationalities (Argentineans against Bolivians for example) instead of grouping them in the same regional group. Despite this, the results obtained in this case are helpful as they point to the importance of studying separate groups of immigrants when trying to estimate their economic impact here in the U.S. VI. Owner-Occupied Housing Market The previous section presents the results of regressions where the dependent variable is growth in Renter-Occupied housing prices. Because a large share of immigrants do not own property, but instead rent, it was important to understand the effect of immigrants on this market. The logical next step of this study would then be to analyze the impact the immigrant inflows have on Owner-Occupied 9 housing prices. The same models used to study the rental market were used to study the owner-occupied housing market. The results can be found on Appendix C. This section will present a brief explanation as to why these results were not included in this study. The results of the regressions in this case were significantly different from those of the rental market case. Even when controlling for several socioeconomic variables, immigrants as a whole had negative and statistically significant coefficients. Oddly, when differentiating between Hispanic and Other immigrants, their coefficients were different from one another even after controlling for socioeconomic factors. The coefficient for Hispanic immigrants was positive, while that of Other Immigrants was negative. Both coefficients were significant. The 9 The U.S. Census Bureau s technical definition of Owner-occupied housing is: housing units where the owner or co-owner lives in the unit even if it is mortgaged or not fully paid for. 24

25 vast differences between the results of the owner-occupied and the renter-occupied housing markets are puzzling, given that rents should follow closely the value of housing units. To test whether immigration had somehow an effect on this relationship (between housing values and rents), regressions were carried where the dependent variable was the difference between owner-occupied housing price growth and renter-occupied housing price growth. In every single regression, even when controlling for several socioeconomic characteristics, including ones not included in the original models (vacancy rates, number of rooms per house, ethnic composition of the tract, units built in the previous 5 years), the coefficient for immigrants was significant. These results may be evidence that immigration somehow distorts the relation between housing value and rental prices. Clearly, this topic warrants additional research but given its complexity it will be left outside of the scope of this paper. VII. Conclusion In this study, I have tried to examine the effect of immigration on the housing market. The results obtained offer some useful insights on this topic. At a national level, the results indicate that without controlling for the socioeconomic characteristics of immigrants, the effect they have on the rental market is significant. It is negative for Hispanic Immigrants, and negligible for Other Immigrants. When controls are added, all groups of immigrants have insignificant coefficients. One explanation for this could be that immigrants may possess characteristics that make neighborhoods they move to less attractive. These characteristics, however, will vary depending on where the immigrants came from, and thus it is important to differentiate immigrants based on their national origin. In this sense, the five-msa case study 25

26 presents a better way to address this issue than the nationwide case, given the presence of data on the national origin of immigrants. The results from the case study on the 5-MSAs seems to suggest that the differences that exist among Hispanic immigrants in terms of how they affect the housing market arises not from what specific country they came from, but from other factors (such as education or income). There are several limitations in this study, of course. The use of the NCDB variable (of the number of people who speak Spanish and do not speak English well), while useful, is not an exact count of the actual number of Hispanic immigrants and might be subject to error. Educated Hispanic immigrants could have learned English before immigrating to the U.S., for example, restricting the sample to only less educated immigrants. The number of observations for the 5 MSA case was not large enough to allow for the inclusion of more controls. And allocating all non-hispanic immigrants to one group could have caused some problems with the results. However, despite these shortcomings, the insights obtained through this research are useful and hopefully will lead to further interest in this subject. Certainly, given the current levels of immigration and the importance of the housing market to the economy, more research is warranted to better understand how these two are linked. 26

27 Table 1: Nation Renter-Occupied Housing Prices (Median) Model 1 Model 2 Model 3 Model 4 Δ Population (9.62) (8.69) (4.81) (-0.71) Δ All Immigrants (-4.59) (-1.11) (0.82) Average Income, Poverty Rate, %Children Controls Additional controls (education, unemployment, older) NO NO YES YES NO NO NO YES MSA Dummies NO YES YES YES Number Observations 41,187 41,187 41,187 41,187 Table 2: Nation Renter-Occupied Housing Prices (Median) Model 1 Model 2 Model 3 Δ Population (6.41) (4.24) (0.20) Δ Hispanic Immigrants (-7.35) (-1.58) (3.81) Δ Other Immigrants (non-hispanic, 10+ in U.S.) (-0.84) (-0.33) (-0.95) Average Income, Poverty Rate, %Children Controls Additional controls (older, education, unemployment,) NO YES YES NO NO YES MSA Dummies YES YES YES Number Observations 41,187 41,187 41,187

28 Table 3: Five MSAs Renter-Occupied Housing Prices (Median) Model 1 Model 2 Model 3 Δ Mexican Immigrants (-3.04) (-1.50) (-1.68) Δ Central American Immigrants (-0.40) (0.25) (0.14) Δ South American Immigrants (0.59) (0.64) (0.84) Δ Other Immigrants (non-hispanic, 10+ in U.S.) (2.48) (-0.64) (-0.50) Average Income, Poverty Rate, %Children Controls Additional controls (education, share of pop renting, older) NO YES YES NO NO YES MSA Dummies YES YES YES Population growth control NO YES YES Number Observations