The differential effects of exchange rate fluctuations on local housing price growth: evidence from Australia

ABSTRACT We investigate whether real exchange rate fluctuations contribute to the fluctuations in housing prices differently across local housing markets through the differential compositions of immigrants. We first develop a new measure of the immigrant-weighted local real exchange rate, which varies across areas and over time within Australia. Using this measure and a fixed-effect instrumental variables approach, we find that the growth in local real exchange rate positively affects housing price growth. This positive relationship is mainly driven by urban metropolitan areas.


INTRODUCTION
Housing booms in global cities that are relatively open to immigration and foreign capital, such as Hong Kong (China) and Vancouver (Canada), have brought concerns about the role of foreign capital inflows on housing price growth in urban centres (Ari et al., 2020;Rabanal, 2018).Although there is some evidence suggesting that housing prices respond to exchange rate fluctuations at the national level (e.g., Adam et al., 2012;Aizenman & Jinjarak, 2009;Bahmani-Oskooee & Wu, 2018), there is scant evidence regarding the differential effects of real exchange rate (RER) fluctuations on housing price growth across regions and areas within the same country.
In this paper, we examine whether exchange rate fluctuations contribute to fluctuations in housing prices across local real estate markets in Australia through the differential compositions of immigrants across these markets.Australia is a particularly fitting setting for our study due to the diversity of immigrants in the population and its relative openness to foreign investments (Gradwell, 2017).We first introduce a novel measure of local real exchange rate (LRER) that is immigrant centric and varies across local government areas (LGAs) and over time within Australia.This LRER measure weights the bilateral RER between the Australian dollar and each foreign currency by the respective share of immigrants among the people who live in that LGA-year.According to this measure, when the Australian dollars depreciate relative to the Renminbi in real terms, for instance, the LRER increases more in LGAs with a larger share of Chineseborn immigrants relative to immigrants from other countries of origin.
The usefulness of our LRER measure in understanding the differential effects of exchange rate fluctuations across local real estate markets hinges on two important observations.First, immigrants and foreign investors have assets in their countries of birth or receive transfers from their overseas family members and, for this reason, their consumption or investment decisions may be sensitive to bilateral price differentials between origin and destination countries (Dustmann, 2003;Dustmann & Görlach, 2016;Stark et al., 1997).Indeed, past studies also find that the economic decisions of immigrants respond to exchange rate fluctuations (Abarcar, 2017;Monras, 2020;Nekoei, 2013;Nguyen & Duncan, 2017;Yang, 2006).Second, immigrants tend to purchase or rent in areas where other immigrants from the same country of origin live (Deng et al., 2021;Ley, 2017;Rogers et al., 2015;Saiz, 2007).This regularity is usually explained by the fact that existing communities from the same countries of origin help immigrants integrate into existing networks and increase their employment opportunities (Åslund, 2005;Munshi, 2003;Zavodny, 1999).Similarly, non-resident foreign buyers also exhibit this country-of-origin bias in their locations of real estate investment (Badarinza & Ramadorai, 2018).At the same time, in areas with large enclaves of immigrants from a certain country, real estate agents may specialize in providing services specifically tailored to these immigrants, increasing the amount and quality of information available to immigrants and foreign investors from that country (Rogers et al., 2015).These facts suggest that exchange rate fluctuations can have disproportionate effects across local real estate markets, depending on the composition of immigrants in an area.
We estimate the relationship between LRER fluctuation and housing price growth using a fixed-effect instrumental variables (FE-IV) approach and quinquennial census data from 1996 to 2016.In particular, we instrument the LRER measure with another LRER based on the settlements of immigrants in the LGA that preceded a major change in the focus of Australia's immigration policy in the mid-1990s.Additionally, we control for LGAlevel time trends and unobserved time-invariant characteristics that may confound the effect of exchange rates on housing prices.Our FE-IV approach addresses the potential concern of reverse causality deriving from the fact that trends in housing prices may determine the composition of the immigrant population in a local area.
We find that as LRER grows by 1%, property price growth increases by approximately 1%, on average.This positive and significant relationship exists for property prices but not for rental prices that are less sensitive to exchange rates.Because different areas in Australia have different compositions of immigrants by county of origin, the findings imply that exchange rate fluctuations have differential impacts on housing price growth across areas within Australia.The estimates are significant and similar regardless of whether we control for the effects of immigrant inflows, other macroeconomic influences and bilateral trade factors on housing price growth.
This paper contributes to the literature on the impacts of shocks across local housing markets.First, our findings add to the literature on the effects of immigration on local housing markets, which has documented that the effects on housing prices depend on the share of immigrant inflows and the response of native mobility (e.g., Accetturo et al., 2014;Gonzalez & Ortega, 2013;Saiz, 2003Saiz, , 2007;;Sá, 2015;Zhu et al., 2019).Specifically, we show that RER fluctuations can have differential impacts on housing price fluctuations across regions within a country through the differential compositions of immigrants.Second, we find that the exchange rate effects on housing prices are independent of the immigration effects that operate through the differential inflows of immigrants across regions documented in past studies.
Second, this paper helps advance the research that examines how foreign and global shocks may propagate differentially across regions within a country.For example, Badarinza and Ramadorai (2018) show that foreign investors influence housing prices in the UK by diversifying the financial risks in their home countries into areas with a high concentration of immigrants from their home countries.Similarly, Ari et al. (2020) find that this 'home bias abroad' influences house price growth and housing affordability in the United States by transmitting foreign political risks into local housing markets.Recently, Moallemi et al. (2022) find that the Australian local housing markets respond to the weighted index of economic activities in the origin countries of immigrants.We differ from these studies because we do not specifically focus on shocks in foreign countries but on price differentials between Australia and other countries.We add to this strand of literature by showing that fluctuations in RERs, whether they are driven by foreign, global or national shocks, can have differential impacts on housing price growth across regions.Particularly, because our LRER measure weights the bilateral RER between the Australian dollar and each foreign currency by the respective share of immigrants who live in that LGA-year, our findings imply that RER fluctuations related to a particular country can have differential impacts on housing price growth across areas, depending on the share of immigrants from that country in each area.
Finally, this paper introduces a new way to estimate the relationship between current account patterns and house prices using within-country time-varying data.Past studies in this literature, such as Aizenman and Jinjarak (2009), Adam et al. (2012) and Bahmani-Oskooee and Wu (2018), typically analyse the effects of current account balances on housing prices at the aggregate level by exploiting the variations in these variables across countries and/or over time.We contribute to this literature by developing a new measure of exchange rate fluctuations, which varies across local housing markets within a country.Our novel measure permits examining the regional impacts of foreign, global and national shocks to exchange rates and current account balances across regions and over time within a single country.Given this innovation, the method can be readily applied to understand the regional impacts of RER fluctuations on local markets.

THE AUSTRALIAN RESIDENTIAL PROPERTY MARKET AND ITS RELATIONSHIP WITH IMMIGRANTS AND FOREIGN INVESTORS
In this section, we provide a brief description of the Australian residential property market, paying particular attention to the role played by foreign investors and immigrants.
Foreigners are permitted to invest in the Australian real estate market, but their access to the market is subject to some restrictions.Specifically, non-resident foreign investors are allowed to purchase newly constructed dwellings subject to the approval of the Foreign Investment Review Board, but they are prohibited from purchasing existing properties.Temporary immigrants on a residential visa of more than 12 months can purchase only one property while they hold a valid visa, but they must sell it when their visa expires.In contrast, permanent immigrants have the same ownership rights as citizens and, therefore, have unrestricted access to housing markets.
As an effect of these regulations, while the proportion of housing stock owned by foreign investors in Australia is estimated to be only between 2.5% and 4.0% (Gradwell, 2017), foreign investors play a relevant role in the market of newly constructed properties.In the period 2015-16, for instance, foreign investors are estimated to have purchased between 30,000 and 50,000 newly constructed dwellings, about 80% of which are apartments, accounting for 15-25% of all newly constructed dwellings in Australia during that period.Across all types of properties, foreign investors' purchases represent between 7% and 13% of all sales.
Immigrants play an even more significant role in the Australian real estate market than do foreign investors.Overall, permanent immigrants own at least 11% of all residential properties in Australia. 1  In 2019, approximately 30% of the Australian population was foreign-born (Australian Bureau of Statistics (ABS), 2020). 2 Figure 1a shows that the share of immigrants in the population grows quite slowly between 1971 and 2000 before it starts accelerating in 2001.The greater increase in the share of immigrants since 2000 has been attributed to the change in the focus of the migration programme from family-to skill-based in 1996 (Larsen, 2013).Figure 1b shows that since 1996, permanent residential visas have become primarily skillbased.Comparing panels A and B, we can see that changes in the share of immigrants correspond to changes in the share of skill-based permanent residential visas granted.
Importantly for this study, a sizable number of immigrants own their residential properties within a few years upon their arrival in Australia.For instance, it is estimated that 50.6% of permanent immigrants who arrived in Australia between 2006 and 2010 owned a home in 2010, with 19% of them not owing any mortgage (Khoo et al., 2012).Furthermore, immigrants who arrived via a skill-based permanent residential visa are much more likely to own a home without any mortgage.For example, in 2006, immigrants who arrived via a skill-based permanent residential visa between 2001 and 2006 is 57% more likely to own their homes without a mortgage than immigrants who arrived via a family-based permanent residential visa during the same period (Khoo et al., 2012).These figures suggest that the availability of savings held in foreign currencies likely plays a significant role in immigrants' decisions to purchase a residential property in Australia.

DATA
We draw data from multiple sources.First, we use census community profiles of LGAs, which are collected at the quinquennial frequency by the Australian Bureau of Statistics (ABS).These profiles are based on 100% of the census count, and they contain, among other variables, information on the number of foreign-born individuals, the number of native-born individuals and weekly rent prices at the LGA level.Starting in 2001, ABS began disclosing the time-series community profiles that collect information over three consecutive census years.For instance, the 2001 time-series community profiles include LGA-level information for 1991LGA-level information for , 1996LGA-level information for and 2001. .We use these profiles to create time-consistent variables in each LGA from 1991 to 2016.In particular, we also use the ABS community profiles to determine the number of immigrants grouped by country of origin living in each LGA.This information is used in the construction of the instrumental variable.We focus the analysis on changes in housing prices and exchange rate fluctuations during the period 1996-2016 using data from a total of 497 LGAs that can be consistently followed across census years. 3In addition, we include data from 1991 because it is the earliest census data before a major change in Australia's immigration policy and for which the details of immigrant groups that we need are made publicly available by ABS.
Second, we use SIRCA's Corelogic property data.The SIRCA dataset contains comprehensive monthly data about real estate transaction data for houses and units at the LGA level, and from these data we construct mean property prices for each census year to match with the frequency of the ABS community profiles. 4In our analysis, we use the Australian classification of dwelling types as follows: a house is a standalone dwelling on its own block of land, while unitswhich include apartmentsare smaller dwellings grouped together on the same block that share common areas such as driveways and gardens; finally, the term property is used to indicate houses and units together.The SIRCA dataset contains information on the number of units and houses sold and their respective total values.Starting from this dataset, we obtain the mean house price in an LGA and census year by dividing the total value of houses sold in an LGA-year by the corresponding number of houses sold.We follow the same procedure to obtain mean unit and property prices.
Finally, we use the data on RERs from the Penn World Tables (Feenstra et al., 2015).Using these data, we construct the RER between Australia and each country of origin contained in the census data.
Table 1 shows the descriptive statistics of our final dataset.The average population of an LGA was 40,670, and almost one-third (11,930) of them were born overseas on average.The share of immigrant inflow is 2.7% on average, varying substantially across LGAs.The LGAlevel average house price is AU$296,007, ranging from AU$2000 to AU$4,337,920.The average unit price is AU$246,990, about 17% lower than the average house price.The LGA-level average price of properties that include both houses and units is AU$265,676.

EMPIRICAL STRATEGY
To estimate the average effect of LRER fluctuations on housing price growth at the local level, we use the following estimation equation: The differential effects of exchange rate fluctuations on local housing price growth 137 where the dependent variable, Dln( p it ), is the change in the log mean housing prices in LGA i between census year t and the previous census year t − 1.The key explanatory variable, Dln(LRER it ), is the change in the log of LRER in LGA i between two census years.We express this specification in first differences of logarithms for several reasons.First, the log difference specification allows us to interpret the estimated effect as a percentage change in growth rates.Second, differencing eliminates time-invariant factors specific to an LGA, such as local amenities, that may systematically affect housing prices.Third, this specification has been used in other related  recent studies making it easier to compare our results with the literature (e.g., Sá, 2015;Zhu et al., 2019).We include LGA fixed effects, r i , in equation ( 1) to capture time trends in housing prices that are specific to a given area.Census-year fixed effects, g t , control for factors specific to a given year, such as policy changes that are implemented at the federal level, which equally affect housing price growth across LGAs.Finally, to account for the fact that immigrant inflow can have a direct effect on housing price growth, which is independent of the effect of fluctuations in LRER, in equation ( 1) we control for the percentage changes in immigration flow specific to a given LGA, measured by the change in the number of immigrants living in LGA i between two census years, DM it , over the total population of LGA i in the previous census year, pop i,t−1 .
The coefficient of interest in this specification is b 1 , which captures the percentage change in housing price growth that corresponds to a 1% growth in the LRER.The extent to which b 1 in equation ( 1) can be interpreted as a causal estimate crucially depends on whether LRER fluctuations are exogenous to the evolution of housing price growth conditional on the other controls.In what follows, we first describe our measure of LRER, then we discuss our identification strategy to uncover the causal effect of LRER fluctuation on housing price growth.

Exchange rate fluctuations at the local level
We obtain the LRER measure that varies at the LGA and census year level as follows: we first weight the bilateral RER between Australia and origin country j in census year t, RER jt , with the number of individuals in LGA i in census year t, who were born in country j, M ijt , relative to the total population in LGA i in census year t, M it . 5We then sum these weighted bilateral RERs over all origin countries available in the census data and Australia to obtain the LGA-specific measure of the relevant exchange rate in a given year, LGA-level LRER as follows: where the RER is the nominal exchange rate between Australian dollars and the currency of country j multiplied the relative prices of a market basket of goods in the two countries (based on the gross domestic product (GDP) deflators of country j and Australia): When country j is Australia, RER jt is 1.By incorporating the share of Australian-born population in each LGA in equation ( 2), the LGA-level LRER reflects the proportion of LGA population that are foreign-born as well as the share of the foreign-born individuals in the population.
The larger the value of the LRER, the lower is the purchasing power of the Australian dollar relative to the (weighted) value of the set of foreign currencies relevant for the immigrants residing in LGA i in census year t.According to this LRER measure, an LGA is more exposed to fluctuations in bilateral RERs between the Australian dollar and the currency used in the origin countries of the larger groups of immigrants living in the LGA.Thus, the average purchasing power differentials between Australia and a set of countries of origin that the immigrants in one LGA face can differ significantly from the average purchasing power differentials between Australia and another set of countries of origin that the immigrants in a different LGA face.In sum, the immigrant weight in equation ( 2) indicates the relevance of the RER between the Australian dollar and the currency of the country of origin j in informing the price differentials between Australia and country of origin j in a particular LGA.
The relevance of this LRER measure in explaining the local housing price growth hinges on two key facts.First, fluctuations in the RERs affect the amount of savings and liquidity from their countries of origin, expressed in Australian dollars, available to immigrants and foreign investors for the housing market in Australia.For instance, immigrants might decide when to transfer their savings from their countries of origin to Australia, based on the movements in the exchange rates.Second, immigrants and foreign investors tend to purchase or rent properties in areas where other immigrants from the same countries of origin have settled (Badarinza & Ramadorai, 2018;Deng et al., 2021;Ley, 2017;Rogers et al., 2015;Saiz, 2007).They have stronger preferences for neighbourhoods that have a larger share of immigrants from the same countries.These two facts imply that the relevant purchasing power differentials between Australia and other countries in a particular LGA, captured by the LRER measure, account for the composition of the immigrant population from different countries of origin in the LGA.A shock to the exchange rate between one currency and the Australian dollar can thus have different effects on housing price growth in different LGAs.

Identification
The identification of the causal effect of LRER fluctuations on housing price growth hinges on the assumption that LRER fluctuations at the LGA level are uncorrelated with unobserved factors driving changes in housing price growth.
To facilitate the discussion of this assumption, we decompose the LRER measure into two components.The first component consists of the bilateral RERs between the Australian dollar and the currency used in country j in census year t (i.e., RER jt in equation 2), while the second component consists of the composition of immigrant groups by country of origin j who reside in LGA i in census year t (i.e., M ijt M it in equation 2).
As concerns the first component, since foreign capitals flowing into Australian LGAs are relatively small compared with the overall liquidity in the exchange market, changes in housing prices in an LGA are unlikely to affect The differential effects of exchange rate fluctuations on local housing price growth 139 the fluctuations of the RERs.For this reason, the first component in equation ( 2), RER jt , can be seen as exogenous to variations in housing prices in an LGA and year.
The second component, however, may be potentially problematic because the variation of housing prices in an LGA may affect the composition of immigrant groups who choose to reside in that LGA.In this regard, employing the first-difference model with LGA fixed effects in equation ( 1) allows us to reduce this concern to some degree by excluding LGA-specific trends and permanent LGA characteristics that may potentially confound the estimates.However, the existence of transitory factors that affect the housing demand or supply remains problematic to the extent that such factors may also contribute to changes in the composition of the immigrant groups in an LGA.For instance, extreme weather events such as wildfires or hurricanes, changes in school quality, or the opening (or closing) of construction sites may affect both housing prices and the sorting of immigrants across LGAs.
To address this potential problem of endogeneity, we introduce an instrument in which we replace the potentially endogenous component of equation ( 2) with the composition of immigrant groups by country of origin in 1991, the census year before the beginning of our sample period.We choose the year 1991 for several reasons.First, it is the earliest year for which detailed information about countries of origin in the data that we need is publicly available (see also section 4).Second, 1991 was before the focus of Australia's immigration policy was changed from family based to skill based in 1996 (see also sections 2 and 4).Third, 1991 precedes the base year (1996) of our sample period, which minimizes concerns related to endogeneity bias (which we elaborate further below).
For consistency with the first difference model used in equation ( 1), we use the first differences of ln z 1,it , Dz 1,it , as an instrument for the endogenous regressor Dln(LRER it ).This instrument takes the following form: To see the relevance of this instrument, consider each component of z 1,it in equation ( 3) and compare it with the right-hand side of equation ( 2).First, they share the same set of bilateral RERs between country j and Australia in census year t, RER jt , which is likely to be exogenous, as we explain above.Second, the instrument uses the past population share of each immigrant group M ij,1991 /M i,1991 , as opposed to the current population share of immigrant group M ij,t /M i,t , in an LGA.The relevance of past settlements in explaining subsequent settlement choices relies on the fact that immigrants tend to move towards regions that host large communities of immigrants from the same countries of origin.This regularity was first noticed by Bartel (1989).
The validity of the instrument requires that unobserved factors that determine transient variations in housing prices at the LGA level (i.e., e it in equation 1) are uncorrelated with the determinants of immigrants' settlement choices in 1991.In our setting, the time lag existing between past settlement decisions and variations in housing prices, coupled with the inclusion of controls for LGA-specific time trends and permanent characteristics in equation ( 1), make it unlikely for the residual component of housing prices, e it , to be correlated with settlements of immigrants in 1991.We further examine potential threats to the validity of this instrument in a robustness check section.
To illustrate, Figure 2a shows the movements in the RERs for the eight major immigrant countries of origin between 1991 and 2016.We use the national immigrant shares in 1991 to select these top eight countries of origin.There exist considerable differences in the movements of RERs for these eight countries.For example, between 1996 and 2001, the RERs increase significantly between Australia and the UK, but decrease significantly between Australia and Germany.Similarly, between 2006 and 2011, the RERs decrease significantly between Australia and the UK, but increase between Australia and China.Importantly, it is clear that there are considerable shocks in the LRER during the sample period.For instance, Figure 2b shows the LRER weighted by the composition of immigrants in 1991 at the national level fluctuates widely over time.
In the setting of our study, we face one additional challenge for identification as the flows of immigrants in an LGA, that is, the term DM it in equation (1), may respond to changes in housing prices.For example, the growth of housing price in an LGA may induce certain groups of immigrants to avoid buying or renting in that LGA.To address this endogeneity concern, we use the following instrument, Dz 2,it , for the immigrant flow variable in equation (1): This instrument allocates the national change in immigrants from the country of origin j in census year t across various LGAs according to the past distribution of immigrants from the country of origin j across LGAs.Note that this instrument has been used in many other studies to estimate the effects of migration on local markets (e.g., Altonji & Card, 1991;Datt et al., 2020;Labanca, 2020;Saiz, 2007;Sá, 2015).

RESULTS
In panel A of Table 2, we present the FE-IV estimates of the effects of LRER changes on housing price growth, distinguishing between property, house, unit and rental prices.First, results from all specifications indicate that the IV is strong.We also report the Anderson and Rubin (1949) F-statistics and the corresponding p-values to test whether the estimated effects on housing prices are robust to weak identification (Andrews et al., 2019).The secondstage estimates indicate that exchange rate fluctuations significantly and positively affect housing price growth.
The magnitude of the effects of LRER changes on property prices indicates that a 1% growth in the LRER leads to about a 1% growth in property prices.The 95% confidence intervals of the estimated effects on house price growth, unit price growth, and property price growth all include one.These findings indicate that the real estate markets for properties capitalize the price differentials between Australia and other countries.
Recall that the LRER fluctuation measure accounts for Australian dollars' purchasing power relative to the weighted value of a set of foreign currencies relevant for the majority of immigrants in an LGA.Suppose, for instance, that as an effect of a depreciation of the Australian dollar, a representative immigrant in an LGA decides to exchange money from their home currencies into Australian dollars with the intention to purchase a property in their LGA.Thus, if the Australian dollar depreciates by 1% relative to the basket of home currencies of immigrants in the same LGA, the average immigrant will obtain 1% more capital to purchase a property from the international transfer.Our estimates imply that property prices in that LGA will also grow by 1% on average, eliminating the gain due to the changes in the basket of exchange rates.
More importantly, the significant relationship between LRER fluctuation and property price growth implies that RER fluctuations have differential impacts on property prices across areas within Australia.Specifically, when the RER between one particular currency and the Australian dollar appreciates, for instance, areas with a high concentration of immigrants from the country of this currency will experience greater growth in property prices than areas with a low concentration of immigrants from this country.The effect of RER fluctuation on property price in an area depends on the composition of immigrants in that area.
Lastly, differently from what we found for property prices, the last column in panel A of Table 2 shows that the rental price growth rate increases by 0.33% following a 1% increase in the LRER growth, meaning that the effect of exchange rate fluctuation on rental price is weak.This result is consistent with the fact that renters are less likely to use international capital to pay rents and, therefore, to time their international transfer to ensure that exchange rate fluctuations are favourable to them.This type of behaviour makes the market for renting less exposed to purchasing power disparities between Australia and foreign countries.While our focus is on the   2 report the pooled ordinary least squares (OLS) estimates and fixed effects estimates, respectively, to show the extent of potential bias in estimates if endogeneity concerns are not fully addressed.The pooled OLS estimates show that housing price growth is positively associated with LRER fluctuations: A 1% increase in LRER fluctuations is associated with a 0.94% growth in house prices, 0.65% growth in unit prices and 0.19% growth in rental prices.In panel C of Table 2, we add LGA fixed effects to the pooled OLS specification.In this specification, LRER fluctuations remain strongly and positively related to housing price growth.Relative to the FE-IV estimates reported in panel A, the magnitude of the estimates presented in panels B and C is smaller, indicating a potential downward bias in pooled OLS and FE estimates.

ROBUSTNESS CHECKS AND ADDITIONAL ANALYSIS
6.1.Alternative way of constructing the LRER Our LRER is an arithmetic mean of bilateral RERs, which is easy to interpret in our context.However, as such measure can be distorted by potential outliers, a geometric (weighted) mean may be preferable (Organisation for Economic Co-operation and Development (OECD) et al., 2004, pp. 360-361).An alternative LRER measure can be constructed by taking the natural log of bilateral RERs first, and then, by averaging them over origin countries with their corresponding immigrant shares: We use the above geometric local real exchange rate (GLRER) and re-estimate the FE-IV specifications.
Note that the IV is also reconstructed similarly.Although the estimates in Table 3 are slightly higher than those in panel A of Table 2.The 95% confidence intervals contain the estimates reported in panel A of Table 2. Thus, the positive and significant effects of LRER on property, house, unit and rental prices are not too sensitive to potential outliers in bilateral RERs.

Past settlement patterns of immigrants
The validity of the instrument rests on the assumption that unobserved factors that determine transient variations in housing price growth at the LGA level are uncorrelated with the determinants of immigrants' settlement choices in 1991.Goldsmith-Pinkham et al. ( 2020) and Jaeger et al. (2018) point out potential threats to identification in the context of Bartik instruments (Bartik, 1991) when the weights used in the Bartik instruments are not exogenous.Our instrument for the changes in LRER is the first difference in the logarithm of the immigrant-weighted average RER between two periods.The first difference and logarithm transformations distinguish our instrument The differential effects of exchange rate fluctuations on local housing price growth 143 from a standard Bartik instrument.Nonetheless, there may still be concerns that the immigrant weights are not conditionally exogenous in the first difference equation that controls for LGA fixed effects and year effects.
To assess this potential threat to identification, we include in the regression the immigrant weight of the largest immigrant group of an LGA as an additional control variable.As the largest immigrant group is weighted most in the LRER measure and is thus most likely to drive its variation, this approach addresses, to some degree, concerns that unobserved factors correlated with the past immigration patterns of particular origin countries may be driving our findings.We perform this robustness test as follows.We fix the largest immigrant group according to the 1991 immigrant composition in an LGA.For example, if the largest group of immigrants in an LGA is from the UK in 1991, then this weight control variable takes the contemporaneous weights of immigrants from the UK in that LGA for all subsequent years.
Panel A of Table 4 show that the largest immigrant group's weight variable is significant in explaining housing price growth, while the estimated effects of LRER are robust to the inclusion of this variable.The results imply that the instrument is likely to be valid and that the estimated effects of LRER are robust to the potential threats of identification.
Alternatively, we include the interaction terms between census-year dummies and a dummy variable that indicates a high share of immigrant population in an LGA in 1991.The high-immigrant-share dummy  The differential effects of exchange rate fluctuations on local housing price growth 145 variable takes a value of 1 if an LGA has the share of immigrants above the average level of all LGAs in 1991.
Given that the variation in the LRER measure is strongly influenced by the shares of different immigrant groups in the population of an LGA in 1991, this interaction variable controls for the time varying unobserved influences specific to LGAs with different shares of immigrant populations.Panel B of Table 4 shows the results, and our key  findings are robust to this potential heterogeneity in the unobserved correlation patterns between housing price growth and immigrant shares of LGAs.
Given that the LRER measure can be sensitive to the share of immigrants in the population, we also drop some sets of LGAs that could potentially drive our findings given that those LGAs have a high share of immigrants (from a particular country) or have a large population, compared with other LGAs.To do so, we first drop the top 10 LGAs with the largest share of UK immigrants and estimate the FE-IV specification given that UK is the largest source of origin country (panel A of Table 5).Second, we drop the top 10 LGAs with the largest share of immigrants regardless of their countries of origin.We choose the top 10 LGAs based on the ranking either in 1991 or in 2016 (panels B and C of Table 5, respectively).The results from these robustness checks show similar coefficient estimates for our key regressor, Δln(LRER).Specifically, the 95% confidence intervals of the effects of LRER include the previous estimates reported in panel A of Table 2. Thus, our findings are not driven by a particular country of origin or certain LGAs with a large share of immigrants.
Lastly, we test if excluding the immigrant inflow variable yields similar results for the effects of LRER on housing price growth.Table 6 shows that the effects of LRER remain similar.

Additional control variables
While our empirical specification controls for new immigration flows, time and LGA fixed effects, there may still exist time-varying LGA characteristics that are correlated with both LRER and housing prices.To further control The differential effects of exchange rate fluctuations on local housing price growth 147 for LGA-level time-varying unobserved factors, we include two sets of additional covariates.First, we include a set of macroeconomic variables: the unemployment rate, the average individual income and the consumer price index (CPI).We calculate the unemployment rate using the quinquennial Australian censuses which have information on the numbers of people employed, unemployed and in the labour force.We obtain the LGA-level average individual weekly income from the Australian censuses as well.We use the state-level CPI data, measured at each state capital city, which are available from the ABS.Next, to disentangle the channel of transmission through which RER fluctuations affects local housing demand, we include foreign direct investment (FDI) and bilateral trade patterns as additional control variables.Because FDI and bilateral trade (both import and export) data are available at the national level, we distribute the national amounts to each LGA based on the distribution of immigrants.To minimize endogeneity in the distribution, we use the 1991 share of existing immigrants from each country in an LGA over the nationwide total immigrants from the same country to distribute the national amounts.
Table 7 presents the estimates for the specification with all the additional control variables, and the results are robust to including these variables.Our results indicate that the mechanism through which RER fluctuations affect housing prices are not sensitive to other local macroeconomic influences and are different from the direct FDI and trade channels. 6

Spatial heterogeneity
We also examine if the effect of RER fluctuations on housing price growth differs across rural and urban areas.Specifically, given the high share of immigrants in urban metropolitan areas, the effect may be driven by urban LGAs.To check this, we split the sample into two groups: urban LGAs and rural LGAs.The urban LGAs include a total of 142 LGAs in the surroundings of metropolitan cities: Adelaide, Brisbane, Cairns, Canberra, Darwin, Geelong, Gold Coast, Hobart, Newcastle, Melbourne, Perth, Sunshine Coast, Sydney, Townsville and Wollongong.The rest of LGAs are classified as rural LGAs.
The FE-IV results are shown in Table 8.The effects of RER fluctuations on property and house prices in urban areas are almost identical to those in the full sample, whilst the effects are insignificant for property, house and unit prices, but negative and significant for rental prices in rural areas.Thus, the significant positive relationship between LRER fluctuations and housing price growth is primarily driven by urban areas.This finding is not surprising because 75.6% of the total population and 85.0% of the total immigrants live in the urban areas, where 78.3% of the total housing transactions occur in 2016.

CONCLUSIONS
This paper investigates the differential effects of RER fluctuations on housing price growth across areas with the same country through the immigrant channel.We first develop a new immigrant-centric measure of LRER that varies across LGAs and over time within Australia.This LRER measure weights bilateral RERs between Australia and a set of foreign countries by the composition of immigrants by country of origin for each LGA.Thus, it captures the purchasing power differentials between Australia and a weighted sum of countries of origin that the representative immigrants in an LGA face.Because the LRER measure varies across LGAs and over time within Australia, we are able to estimate the causal effect of LRER fluctuations on housing price growth through a FE-IV estimation framework.
We find that as LRER grows by 1%, property price grows by approximately 1%.Our findings indicate that the Australian real estate market is strongly influenced by the purchasing power disparities between Australia and its immigrant sending economies.The significant positive relationship between LRER fluctuations and housing price growth implies that price differentials in the exchange rate market are absorbed in the local real estate markets.Moreover, this relationship is robust to controlling for the effect of immigration flows and other unobserved influences that may be correlated with the composition and share of immigrants in an LGA.Unlike most past studies examining the effect of immigration on house price, we do not find a significant positive effect of immigrant inflows on housing price growth.
Our results indicate that RER fluctuations affect property price growth differently across areas within the same country, depending on the compositions of immigrant groups in these areas.Housing prices in areas with a greater diversity of immigrants are thus more sensitive to exchange rate fluctuations.Housing policies and other policy interventions should take this feature into account.This study focuses on housing prices but does not examine other local market outcomes, such as job creation, income and price level.Future research focusing on these outcomes can potentially yield important insights into how price differentials in the exchange rate market are absorbed into other local markets.2016), over half of permanent migrants in Australia are homeowners (about 1 million people), and there are 8.8 million occupied private dwellings (https://profile.id.com.au/australia/population).Thus, permanent migrants own at least 11% of all occupied private dwellings in Australia.2. Slightly more than half of all immigrants came from 10 countries: England (2.9%),China (2.7%), India (2.6%), New Zealand (2.2%), Philippines (1.2%), Vietnam (1%), South Africa (0.8%), Italy (0.7%), Malaysia (0.7%) and Sri Lanka (0.6%).Roughly 21% of residential property owners in Australia are immigrants (Baxter & McDonald, 2004).3. Due to the fact that some LGA boundaries change between census years, we construct and use an LGA classification that allows us to follow LGAs consistently over time.The raw data have 625 LGAs available in 1991, 671 LGAs in the period 1996-2001 and 544 LGAs in the period 2006-16.4. We focus on census years because ABS community profiles are only available at quinquennial frequency.As the Penn World Table data are available at an annual frequency and our LRER measure can be constructed at an annual frequency, we also estimate reduced-form specifications using annual frequency data and find a significant positive relationship between the LRER fluctuation and housing price growth.The results are shown in Table A1 in Appendix A in the supplemental data online.5. We include in our analysis the 29 countries of origin that are consistently reported across census years.We omit those grouped under the other category in the construction of the weights.6.Table A2 in Appendix A in the supplemental data online finds no significant effects of LRER on local macro-variables, such as unemployment rate and average individual income.The results indicate that these macro-variables are not likely the channels through which LRER affects housing price growth.

Figure 1 .
Figure 1.Share of immigrants in the population (a) and share of skill-based permanent residential visas granted (b).Note: The permanent residential visas in the denominator for the series in (b) include skill visas, family visas and special visas.Sources: Data for the share of immigrants in the population were sourced from the Australian Bureau of Statistics (ABS) (2020).Data for the share of skill-based permanent residential visas granted were sourced from Philips and Simon-Davies (2017).

Figure 2 .
Figure 2. Trends in real exchange rates (RERs) between selected countries and Australian dollars (a) and immigrant-weighted local real exchange rates (LRERs) (b).Note and sources: The selected countries in (a) are the top eight immigrant countries of origin in census 1991 data.The RER data came from the Penn World Table.The immigrant-weighted LRERs in (b) are based on the 1991 immigrant weights of 29 different countries of origin.

FUNDING
Liang Choon Wang received funding support from the Department of Economics at Monash University [research support fund number 1758054].

NOTES 1 .
According to the Australian Bureau of Statistics (

Table 2 .
Effects of real effect exchange rate fluctuations on housing price growth.
Note: Each regression is weighted by the number of purchases or renters.See section 3 for a detailed description of the construction of the instrumental variables (IV).Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively.

Table 3 .
Robustness check: the alternative way of constructing the local real exchange rate (LRER) measure.
Note: The geometric local real exchange rate (GLRER) measure is constructed based on the equation in section 6.1, which is the natural logarithm of the geometric mean of real exchange rate weighted by immigrant shares.We then take the first difference to obtain ΔGLRER.Each regression is weighted by the number of purchases or renters.Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively.

Table 4 .
Robustness to potential threats to identification: fixed-effect instrumental variables (FE-IV) estimation.
Note: Each regression is weighted by the number of purchases or renters.Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.In panel A, the largest immigrant weight variable is constructed based on the largest group of immigrants from an origin country in an LGA in 1991.In panel B, the high immigrant share dummy takes a value of 1 if the immigrant share of an LGA is above average in 1991.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively.

Table 5 .
Robustness to dropping certainLGAs: fixed-effect instrumental variables (FE-IV) estimation.Each regression is weighted by the number of purchases or renters.Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.In panel A, we exclude 10 LGAs that have the largest share of UK immigrants.The UK is the country of origin that most immigrants came from during our entire sample period.In panels B and C, we exclude the top 10 LGAs that have the largest share of immigrants in 1991 and in 2016, respectively.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively. Note:

Table 6 .
Robustness check to excluding immigrant inflows.Each regression is weighted by the number of purchases or renters.Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively.

Table 7 .
Robustness to including additional control variables.Each regression is weighted by the number of purchases or renters.Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.The average individual income is measured in thousands of US dollars.The foreign direct investment (FDI) is measured in millions of Australian dollars, while import and export amounts are measured in millions of US dollars.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively. Note:
Note: Each regression is weighted by the number of purchases or renters.Robust standard errors clustered at the local government area (LGA) level are reported in parentheses.The urban areas include LGAs within the surrounding areas of the following cities: Adelaide, Brisbane, Cairns, Canberra, Darwin, Geelong, Gold Coast, Hobart, Newcastle, Melbourne, Perth, Sunshine Coast, Sydney, Townsville and Wollongong.The rural areas are the rest of LGAs.*, ** and ***Significance at the 10%, 5% and 1% levels, respectively.