2. A Review on the Pro-Trade Effect of Migration

Positive correlation between migration and international trade has been widely documented for different countries, periods, and goods since 1994 (see Gould, Reference Gould1994). The literature has identified two possible channels via which migrants could stimulate trade flows: the network channel or the business and social network effect (Rauch and Trindade, Reference Rauch and Trindade2002), and the preference channel or the transplanted home bias effect (Gould, Reference Gould1994; White, Reference White2007).

The first channel is based on the idea that immigrants can reduce the fixed cost of trading because of the language, the specific knowledge of homeland institutions and norms. Moreover, immigrant networks may provide contract enforcement through sanctions and exclusions, which substitutes for weak institutional rules and reduces trade costs. Chaney (Reference Chaney2016) presents a very exhaustive literature review on ethnic networks and the patterns of international tradeFootnote ⁴.

The presence of migrant networks mitigates information frictions, in particular imperfect information, and facilitates trade. More specifically, Chaney (Reference Chaney2016) considers the difficulty in acquiring information about foreign products, the differences in taste between domestic and foreign consumers, troubles in communication among traders in the case of trade in highly differentiated and customized intermediates, considering them all as informal barriers. Ethnic networks are a proxy for the presence of social ties and they may affect the patterns of international trade through informal barriers and also contract enforcement and dispute resolution. Contract enforcement and dispute resolution relates to differences in legal systems between countries, ambiguity in the extent of the jurisdiction of the national court system, the inability of buyers and sellers to fully commit to pre-established contracts ex-ante, and the inability of the justice system to perfectly enforce existing contracts ex-post.

Concerning the preference channel, or the transplanted home-bias effect, as in Gould (Reference Gould1994) and White (Reference White2007), the idea is that immigrants are characterized by different habits in consumption when compared with natives. They may modify their original home-biased habits after settling in the host country, but it happens slowly over time. Supportive of preference stability, Atkin (Reference Atkin2013) investigated habit formation in tastes and concluded that preferences developed in childhood persist into adulthood. It means that preferences tend to move with migrants and thus consumers from the same ethnic group tend to share the same tastes. In an innovative work based on data of transnational migration and international trade from 40 countries, Zhang (Reference Zhang2020) introduces consumers’ heterogeneity according to the presence of migrants by ethnic groups and is able to explain the trade-bias departure using the traditional gravity model based on the representative consumer.

The existing literature suggests that the relevance of these channels would be different for different types of products and for different types of immigrants/source countries. In general, the empirical literature has shown that both mechanisms usually work together (Head and Ries, Reference Head and Ries1998; Girma and Yu, Reference Girma and Yu2002) and the network channel is found to be very relevant by some authors (Rauch, Reference Rauch1999; Herander and Saavedra, Reference Herander and Saavedra2005). However, when focusing on trade flows, imports can benefit not only from the network effect, but also from the preference effect, and the elasticity of import flows to migration is higher than the same elasticity of export flows, as found in Bratti et al. (Reference Bratti, Benedictis and Santoni2014). The preference effect tends to increase with the number of migrants, as it measures the market size of the host country for imports from the home country, but it may decrease over the time if the preferences of the migrants are influenced by those of the natives.

Disentangling the single contribution of each channel is difficult, especially with aggregate data. The magnitude of the preference channel is usually inferred by the difference between the estimated elasticities of imports and exports to migration. Our work contributes to this literature by taking advantage of the extraordinary historical experience of migration to Argentina in the First Globalization when transportation costs fell much more than communication costs (especially in comparison with the most recent globalization)Footnote ⁵. These latter costs are essential to acquire information on the destination market and in reducing demand uncertainty, i.e. one of the most relevant factors for exporting decisions (see e.g. Albornoz et al., Reference Albornoz, Calvo Pardo, Corcos and Ornelas2012). Then, immigrants in Argentina worked as a sort of insurance for importers of ethnic goods since the preference channel reduced uncertainty in the demand for origin-country goods. Instead, the (still) relatively high costs of communication between Argentine migrants and their contacts in the origin countries might have weakened the network effect on the Argentine exports.

Jacks and Tang (Reference Jacks and Tang2018) have already shown how the correlation between migration and trade has been holding since 1870 with the exception of the interwar period, but the Argentine case can offer an interesting case study. For instance, migrants built business networks in Buenos Aires (Abad and Sánchez-Alonso, Reference Abad and Sánchez-Alonso2018), but at the same time the country experienced a transition in consumption patterns with the introduction in the market of new products. Indeed, Ramon-Muñoz (Reference Ramon-Muñoz2009) highlights how migration from the origin countries is related to the relevant increase in Argentine imports of olive oil from Southern Europe, wine from Spain, and cloth from ItalyFootnote ⁶. Moreover, Fernández (Reference Fernández2004) has underlined the correlation between overall Argentine imports and immigration flows from both Spain and Italy before World War I. Regarding evidence from the origin countries, Zamagni (Reference Zamagni1997) has argued that the growth of Italian exports to Argentina and the United States prior to 1913 could be partly explained ‘by the massive presence of Italian immigrants in these countries’ (p. 125).

3. Historical Background

Argentina's provinces gained independence from Spain in 1816, but a series of civil conflicts occurred during the first half of the nineteenth century and the country finally united in 1862 with a constitution that contained modern values and strongly affirmed the rule of law.

Besides well-designed institutions, Argentina boomed economically thanks to both the rapidly increased endowment of economic resources – both land and labor – and the openness to trade. As stated in the introduction, Argentina outpaced many similar countries, such as Brazil and Mexico, in their share of world trade (see Figure 1).

Figure 1.

Share of world trade for Argentina, Brazil, Chile, and Mexico (left axis); USA for comparison (right axis), 1870 1913

Source: Fouquin and Hugot (Reference Fouquin and Hugot2016).

Land resources were acquired with the expansion of the frontier by means of military campaigns, differing from the US case of civilian settlers. During the Conquista del Desierto 135.000 square miles were taken from the indigenous dwellers and assigned to the Argentine settlementsFootnote ⁷.

The demographic dynamics was the other important factor that determined the rapid and deep transformation in Argentina in the second half of the nineteenth century, with migration playing a key roleFootnote ⁸. Argentine territory was highly underpopulated at the beginning of the nineteenth century with 600,000 inhabitants overall in 1810 and characterized by rural settlements. Still in 1870, 71.2% of the population lives in rural area. One century later in 1910, the population had increased tenfold and the urban population (51.1%) had overcome the rural population. Illiteracy was reduced from 78.24% of the population in 1869 to 37.87% in 1914 and reached percentages similar to the major developed economies.

Argentina showed an exceptional ability to attract immigrants, as shown in Table 1. In terms of stocks, between the mid-nineteenth century and 1930 approximately four million Europeans settled in Argentina, some two million in Brazil, and slightly fewer than 600,000 in Cuba and Uruguay (Sánchez-Albornoz, Reference Sánchez-Albornoz1974, p. 129).

Table 1.

Immigration to the New World by decade and by destination (10-year average and per thousand population units)

Source: Sánchez-Alonso (Reference Sánchez-Alonso2019).

There are three main reasons why among the South-American countries Argentina was so successful in attracting migrants.

First, the complementarity between land and labor induced a big increase in labor demand, created mainly by the need to populate the new lands following the expansion of the land frontier, and marginally also by the abolition of slavery (1853). As a result, the real wage increased and became one of the main drivers of immigration. In Figure 2, we show the migration inflows from Italy and Spain to Argentina and the increases in Argentine real wages when compared with Italian and Spanish wages, indicating a positive correlation, especially for the period 1870–1890 (between 0.33% and 0.36%). The inflows dropped during the years around the Baring crisis in 1890 – one of the first international sovereign-debt crisis that intensively hit ArgentinaFootnote ⁹ – but resumed in the early 1900s after the immigration policy restrictions in 1890s (Timmer and Williamson, Reference Timmer and Williamson1998).

Figure 2.

Real-wage premia of Argentine wages and immigration flows, 1870–1913 (5-year centered moving average)

Notes: Real-wage index. UK 1905 = 100. Sources: Williamson (Reference Williamson1999), Inklaar et al. (Reference Inklaar, de Jong, Bolt and van Zanden2018), Banco Central de la República Argentina (1915).

Second, the Argentine government implemented measures to promote immigration, which remained in place until the end of the centuryFootnote ¹⁰, and, third, network effects and the lower socio-economic costs of integration explain the selection of Argentina as a destination in the New World. Cultural affinity, language similarity, and the existing migrants were decisive in the selection of the final arrival countries. Compared to Italy, the average wage for unskilled workers in the 1870s was only 43% of the same wage in Argentina, but 22% of that in the United States. Notwithstanding the different wage gaps, in the 1870s and 1880s the destination of many Italian migrants was Argentina (Devoto, Reference Devoto2006). This phenomenon shows how the real wage differential can be the trigger to out-migration, but in the choice of final destination the presence of social networks and cultural affinity may play an even more relevant role, as argued by Moretti (Reference Moretti1999). Along the same lines, Sánchez-Alonso (Reference Sánchez-Alonso2019) argues that many skilled Italian migrants preferred Argentina over the US, given the similarity between the Italian and Spanish languages when compared to English. The lack of a language barrier would make it easier for them to find employment in occupations that rely on communication skills and offer higher wages.

As reported in Table 2, the foreign-born population kept on increasing and accounted for almost 44% on a yearly average in the first decade of the 1900. The table also highlights how migration tended to polarize into two main nationalities, i.e. Italians and Spaniards. The second important factor for the Argentine economic rise was provided by the increase in international trade during the First Globalization. Demand for agricultural products rose in the late nineteenth century mainly from North-Western Europe as a result of the fast population growth in conjunction with the rising average per-capita income. Given its increased endowment of land and natural resources, Argentina became of the major exporters in primary products in exchange for manufactured goods with advanced European countries (see Schedvin, Reference Schedvin1990 and Denoon, Reference Denoon1983)Footnote ¹¹.

Table 2.

Population and immigration in Argentina 1870–1913 (10-year average)

Source: Ferreres (Reference Ferreres2005).

Overall, between 1870 and 1913 its share in world exports rose from 0.8% to almost 4%. The main trading partners were Great Britain, France, Belgium, and Germany. Spain and Italy lagged behind, but still played an important role in trade. In terms of value, Argentina traded 4.45 million US$ with Italy in 1881 and the flow reached 59.27 million US$ in 1913. The trade flow with Spain soared from 3.82 million US$ in 1881 to 18.96 million US$ in 1913.

Several studies investigate the export-led characteristic of Argentine growth, e.g. Conde (Reference Conde and Zimmermann2005) and Taylor (Reference Taylor1994). More recently, Pinilla and Rayes (Reference Pinilla and Rayes2019) used a gravity model for Argentine exports between 1880 and 1929 and showed the relevance of both the supply and demand sides, as well as the positive impact of the reduction in transport costs. The authors show how the Argentine export-led model was a result of the strong integration and complementarity with European countries that were at a more advanced stage in their industrialization process (especially Great Britain).

Our main point in this study is to uncover the interplay between immigration flows and openess to international trade that contributed to Argentina's fortunes. As Figure 3 shows, the two phenomena occurred at the same time and in our empirical evidence we aim at disentangling the channel of immigration as a pro-trade factor.

Figure 3.

Share of foreign-born population and openness (exports + imports/GDP) in Argentina, 1870–1913

Source: Federico and Tena-Junguito (Reference Federico and Tena-Junguito2017) and Ferreres (Reference Ferreres2005).

4. Empirical Analysis

Our analysis aims at identifying the additional contribution of migration on the trade flows between Argentina and its major European partners that were also the major origin countries of Argentine immigration. The data source for bilateral nominal trade flows, nominal GDP, and gross national expenditure (GNE) is the TRADHIST dataset (Fouquin and Hugot, Reference Fouquin and Hugot2016).

Bilateral migration inflows of Argentina with eight European countries (Austro-Hungarian Empire, Belgium, France, Germany, Italy, Spain, Switzerland, and United Kingdom) from 1870 to 1913 are obtained from two historical archives, i.e. Banco Central de la República Argentina (1915) and Commissariato generale dell'emigrazione (1926). More detailed information is provided in Appendix A.

4.1 The Empirical Model

Our empirical strategy relies on the standard gravity model when there is one pivot country A as both exporter and importer, i.e. Argentina, with its partners i, that are: Austro-Hungarian Empire, Belgium, France, Germany, Italy, Spain, Switzerland, and United Kingdom. Following Head and Mayer (Reference Head, Mayer, Jones, Grossman, Kenen, Rogoff, Gopinath and Helpman2014), we combine definitions 1 and 2 of general and structural gravity with the pivot country A and obtain:

(1)$$X_{i, t}^A = \left\{{\matrix{ {\displaystyle{{Y_{i, t}} \over {\Omega_{i, t}}}\;\displaystyle{{X_{A, t}} \over {\Phi_{A, t}}}\;\Phi_{A, i, t}\;when\;A\;is\;importer} \cr {\displaystyle{{Y_{A, t}} \over {\Omega_{A, t}}}\;\displaystyle{{X_{i, t}} \over {\Phi_{A, t}}}\;\Phi_{i, A, t}\;when\;A\;is\;exporter} \cr } } \right.\;\;\;\;$$

where, for each time period t, $X_{i, t}^A $ is the trade flow between Argentina and its partners, either imports or exports, Y _h,t and X _h,t would represent respectively gross production and total expenditure in country h = A, i; Ω_h,t and Φ_h,t are multilateral resistance terms for country h = A, i.

The bilateral resistance terms Φ_A,i,t and Φ_i,A,t are assumed to be symmetric and such that:

$$\Phi _{A, \;i, \;t\;} = \Phi _{i, \;A, \;t\;} = ( {dist_i^A \;} ) ^\eta ( {MIG_{A, \;i, \;t}} ) ^\delta $$

where $dist_i^A $ is the sea distance of each country i from Argentina and $MIG_{A, \;i, \;t}$ is immigration to Argentina from country i at time t Footnote ¹².

By taking log transformation, we estimated model (1) with two different empirical approaches. In what follows, small-letter variables represent log-transformation of capital-letter variables, i.e. ln Z  ≡  z. With the first approach, we consider Argentina as an importer and an exporter in one single equation as follows

(2)$$x_{( {h, j} ) , t} = \alpha + \beta _{exp\;}y_{\,j, \;t} + \beta _{imp}x_{h, \;t} + \delta mig_{( {h, j} ) , t} + \eta dist_{( {h, j} ) } + \mu _{( {h, j} ) } + \mu _t + \epsilon _{( {h, j} ) , t\;}$$

where for h = A, i j = A, i h ≠ j and each time period t:

• • x _(h,j),t is imports of h from i;

• • y _(j,t) is the gross production of exporter j;

• • x _(h,t) is the total expenditure of importer h;

• • $mig_{( {h, j} ), t} = mig_{i, t}^A $, i.e. the only relevant migration flows for that period was from each country i to Argentina (see footnote 12);

• • dist _(h,j) is the time-invariant sea distance between h and j;

• • μ _(h,j) is a country-pair dummy to capture the time-invariant heterogeneity in the relationship between h and j not included in the sea distance dist _(h,j)Footnote ¹³;

• • μ _t is the year dummy to capture yearly common shocks.

With the second approach, we consider two different empirical models depending on whether Argentina is the importer or the exporter country:

(3)$$x_{i, t}^A = \left\{{\matrix{ {\alpha_{imp} + \beta_{exp\;}y_{i, \;t} + \delta_{imp}mig_{A, i, t} + \eta_{imp}dist_i^A + \mu_{i-exporter} + \mu_t + \epsilon_{i, t\;}when\;A\;is\;importer} \cr {\alpha_{exp} + \beta_{imp\;}x_{i, \;t} + \delta_{exp}mig_{A, i, t} + \eta_{exp}dist_i^A + \mu_{i-importer} + \mu_t + \epsilon_{i, t\;}when\;A\;is\;exporter} \cr } } \right.$$

The two approaches present pros and cons. By estimating (2), we save degrees of freedom, but we miss the possibility of disentangling the different effects of migration on imports and exports since there is a common δ estimate. When considering model (3), the number of degrees of freedom drops, but we estimate two distinct parameters for migration, δ _imp and δ _exp, as the effects of migration to Argentina on Argentine imports and imports.

4.2 The Identification Strategy

The OLS estimation is likely to produce biased and inconsistent estimates because migration inflows could be endogenous, violating the orthogonality condition with the error term, i.e. ${\epsilon_{i, t}}$ in Equations (2) and (3). Our estimation tackles two endogeneity issues.

First, OLS estimates might be biased because omitted variables could simultaneously affect trade and migration flows. Economic growth and technological progress have been proved to be positively associated to international trade and migration (see e.g. Felbermayr et al., Reference Felbermayr, Grossmann and Kohler2015). To control for the former, in the regression model we include a measure of countries’ absorption capacity or ability to sell (GNE or GDP). To control for the latter, we include year fixed effects, since we suppose that the technological progress that has decreased transportation costs during nineteenth century was common to all the countries considered in the analysis – all European CountriesFootnote ¹⁴.

Secondly, the Argentine upsurge in the world market might have increased labor demand, which, in turn, could have led to the arrival of foreign workers. If the increase of Argentine bilateral trade had eased migration flows along the same routes, OLS estimates would be inconsistent. In order to cut out the confounding effect of labor demand (the so-called pull factor), we propose an instrumental variable (IV) that could only account for movement of the Argentine labor supply due to migration, i.e. the general push factors of European out-migration.

Migration flows are determined by conditions at both destination and origin countries. In our analysis, origin countries are Argentina's trade partners and the only destination country is Argentina. The ideal instrumental variable should be unrelated to factors that pull migrants to Argentina, but be able to capture factors that push migrants out from Europe. An incomplete list of these factors includes out-migration public incentives, demographic trends, and past out-migration that characterized the so-called Atlantic Mass Migration or Migration to the New World (see Hatton et al., Reference Hatton and Ward2018). Therefore, we use migration flows from Europe to US as an instrumental variable. In formula:

$$Z_{( {i, \;t} ) }^A = mig_{( {i, \;t} ) }^{USA} $$

Data on migration flows by nationality into the United States from 1870 to 1913 are obtained from the US Bureau of the Census (1949). We sampled the migration flows from: Italy, other Southern countries – which include Spain, Portugal and Greece – Austria-Hungary, Germany, United Kingdom and other North-Western countries – which include France, Belgium, Netherlands, Switzerland and Luxembourg. In order to assure consistency between our trade data and the differently aggregated data in the US-based instrumental variable, we have merged trade flows from France, Belgium, and Switzerland.

Our IV approach is related to the identification strategy in Autor et al. (Reference Autor, Dorn and Hanson2013) where Chinese exports to all countries but the US were the instrumental variable for Chinese exports to the US. Similarly, we use out-migration of the selected European countries to the United States as our instrumental variable. Our maintained hypothesis is that the push factors for European out-migration in late nineteenth century are very similar to all destination countries in the New World. Moreover, we assume that the Argentine bilateral trade with each European origin country did not affect the European out-migration to United States (our exclusion restriction) and migration flows to the US did not increase Argentine imports and exports. More explicitly, in order to have a non-contaminated IV estimates, migrant demand shocks and trade flows should not be correlated across US and Argentina.

To substantiate this latter hypothesis, in Figure 4 we compare Argentine and US trade from 1870 to 1913. During this period, the two countries were characterized by similar long-run trends in entering world trade and the graph focuses on the short-run behavior by presenting the growth rates (first log-difference) and the HP-filtered cyclical components of US and Argentine total trade flows – imports plus exports with the whole world. The two-time series clearly show different short-run dynamics. More formally, correlation coefficients for the first log-differences is 0.1441, not significant at the 0.05 level. We interpret this result as suggesting that correlated migrant demand shocks in the US and in Argentina are unlikely to be occurred in the years of the analysis.

Figure 4.

The Different Dynamics of US and Argentine Trade Flows

Notes: In panel (a) first log-differences bilateral trade flows with the rest of the world are reported. In panel (b) HP-filtered cyclical components of bilateral trade flows with the rest of the world are reported.

Source: Authors’ elaboration on CEPII data.

4.3 Main Results

The estimation of the augmented gravity model in Equation (1) is performed with the two different approaches, presented in Equations (2) and (3). As mentioned above, only six of the original eight countries can be used for the IV estimation, therefore in the OLS estimation we consider the consistent sample with Austro-Hungarian Empire, Germany, Italy, Spain, BFS (an aggregate of Belgium, France and Switzerland) and United Kingdom.

In Table 3. we report OLS and IV estimates when using both the approach (Eq. (2)), i.e. flipping the role of countries depending on whether it is an importer or an exporter, and the approach (Eq. (3))Footnote ¹⁵. The IV estimation has been implemented with a standard two-stage approach (see Baltagi, Reference Baltagi1981). All the traditional gravity variables have the expected signs, are significant,Footnote ¹⁶ and the magnitude of the coefficients does not change from the OLS to the IV estimates. Measures of economic masses (capability of exporting and absorption) show a positive elasticity between Equation (2) and Equation (3), which – in comparison with the existing literature that shows an elasticity lower than 2 and refers to more recent period – may signal that during the First Globalization common event (as the decreasing trade costs) made trade more sensitive to measures of domestic economic activity. The measure of bilateral resistance is consistently negative as expected. Immigration flows had a positive effect on aggregated trade flows of Argentina in the years 1870–1913. This result emerges from both the OLS and the IV estimates, even though the two-stage estimation method produces a greater coefficient, pointing at a downward bias affecting the OLS regression. This might be the case with a measurement error, i.e. when a variable is measured with an error that is uncorrelated with its true value. Since, migration data are collected directly in historical archives, we cannot exclude that this was the case, making the IV results more reliable than the OLS ones. In particular, the IV significant estimates show an elasticity around 0.6% for overall trade.

Table 3.

OLS and IV estimation of the (log-linearized) gravity model for Argentina (1870–1914) – overall trade and distinctly for exports and imports

Notes: Authors’ elaborations. In the upper panel, we present the results of the OLS estimation of a panel model with random effects. In the bottom panel, we present the results of the IV estimation of a panel model with random effects. In model 1 (EXP + IMP) of both panels, we consider bilateral trade flows and we flip the role of countries depending on whether it is an importer or exporter. In this specification, we consider country-pair fixed effects. In model 2 (IMP) of both panels, we consider only Argentine imports and in model 3 (EXP) of both panels we consider only Argentine exports. In these latter specifications, we consider country fixed effects, since there is one common trade partner, i.e. Argentina. Time fixed effects are always included. The distance is the sea distance in maritime miles. The IV model is estimated following Baltagi (Reference Baltagi1981). Robust standard errors in parenthesis; *p < 0.10, **p < 0.05, ***p < 0.01.

Furthermore, we obtain different effects of migration on exports and imports when we consider them separately (see columns 3 and 4 in Table 3). Migration inflows have a positive and significant effect when implementing the IV estimation for imports, where a 1% increase in migration inflows is associated with an increase that is close to 0.9%. The estimates for exports are not significant and lower than 0.2.

A possible explanation for this result – a preference effect (more relevant for imports) stronger than a network/information effect (more relevant for exports) – may be related to the characteristics of our period of investigation, i.e. the First Globalization. As also mentioned in the Introduction and in Section 2, both communication and transportation costs lowered during all recent globalization periods, but at the end of the nineteenth century communication costs, normally more related to the network/information effect, decreased relatively less than the physical transportation costs, more important for the preference effect.

4.4 Robustness Checks

The literature underlines the importance of migration stocks besides migration flows. Reliable measures of migration stocks with the origin-country breakdown are only available for three years from the Census. Therefore, we resort to two indirect ways to measure yearly immigration stocks at the origin-country level. First, we cumulated the past five years of migration flows and used this variable instead of migration flows. This is an imperfect measure of stocks since we did not cumulate net migration flows and during that period there were return migration flows.

Table 4 reports the OLS and the IV estimation for model (3) with the cumulated flows when including all fixed effects. The migration variable has always a positive effect and it is confirmed that the import elasticity is significantly positive and higher than the export elasticity, in line with the previous estimation in Table 3.

Table 4.

OLS and IV estimation of the (log-linearized) gravity model (3) for Argentina (1870–1914) augmented with immigration cumulated flows as a proxy for stocks

Notes: Authors’ elaborations. In the first two columns, we present the results of the OLS estimation of a panel model with random effects. In the last two columns, we present the results of the IV estimation of a panel model with random effects. In column 1 and 3, we consider export flows. In column 2 and 4, we consider import flows. Immigration flows are cumulated over the past five years. In both cases, we consider country fixed effects and time fixed effects. The distance is the sea distance in maritime miles. The IV models are estimated following Baltagi (Reference Baltagi1981). Robust standard errors in parenthesis; *p < 0.10, **p < 0.05, ***p < 0.01

Secondly, we use lagged immigration flows up to five years to see whether past migration affects trade via the different channels (decreased information costs for exports and preference for imports). Table 5 reports the IV estimation results by including lagged immigration flows one at the time up to time (t-5)Footnote ¹⁷. The significant effect on imports is fully confirmed and it slowly decreases over time: after five years, the estimated elasticity is 0.65, i.e. only less that 0.2 than the 1-year lagged migration. The effect on exports is instead never significant at all lagsFootnote ¹⁸.

Table 5.

IV Estimation of the (log-linearized) gravity model for Argentina (1870–1914) augmented with immigration lags

Notes: Authors’ elaborations. In the table, we present the results of the IV estimation of a panel model with random effects. In the left panel, we consider export flows. In the right panel, we consider import flows. Immigration flows are lagged for a maximum of a five-year lag. In both cases, we consider country fixed effects and time fixed effects. The distance is the sea distance in maritime miles. OLS results are reported in the Appendix B. The IV models are estimated following Baltagi (Reference Baltagi1981). Robust standard errors in parenthesis; *p< 0.10, **p < 0.05, ***p < 0.01.

The stronger and consistent effect of migration for imports supports the interpretation that during this period migration worked more as a factor that increased the demand for products from the origin countries (the so-called olive oil effect) rather than as a facilitator for exports.

As a final robustness test, we have regressed trade flows (bilateral flows, imports, and exports) on future migration flows (Table 6). More precisely, the trade flow between country j and country h at time t is regressed over the migration flow from h to j at time t + 10. None of the models shows a significant correlation between current trade and future immigration, possibly reassuring us about the potential reverse causality of the model.

Table 6.

Placebo test: panel estimation of the (log-linearized) augmented gravity model for Argentina 1870–1903 with 10-year leaded immigration

Notes: Authors’ elaborations. Results of the OLS estimation of a panel model with random effects. In model 1 (EXP + IMP), we consider bilateral trade flows and we flip the role of countries depending on whether it is an importer or exporter. In this specification, we consider country-pair fixed effect. In model 2 (IMP), we consider only Argentine imports and in model 3 (EXP) we consider only Argentine exports. In these latter specifications, we consider country fixed effects, since there is one common trade partner, i.e. Argentina. Time fixed effects are always included. Immigration flows are leaded by 10 years. Therefore, the trade flow between country j and country h at time t is regressed over the immigration flow from h to j at time t +10. The distance is the sea distance in maritime miles. Robust standard errors in parentheses. *p < 0.10, **p < 0.05, ***p < 0.01.