Benchmark estimates
The estimates of Eq. (2) are presented in Table 2. We progressively allow for lower degrees of variability across specifications in our identification data by gradually augmenting the number of fixed effects. Column (1) includes a reduced set of origin-year and destination dummies which capture time-varying factors at origin and time-invariant factors at destination, including unobserved heterogeneity in cultural traits between migrants and non-migrants. This specification is very close to the predictions of the model proposed by Ortega and Peri (2013). Our parameter of interest, the coefficient of \(\mathrm{ln}({Xcult}_{\mathrm{ni},\mathrm{t}-1})\), suggests a significantly positive relationship between proximity of country i towards country n’s culture and bilateral emigration from origin i to destination n. All of the gravity controls are significant and have the expected sign. Income per capita at destination is confirmed as an important driver of migration flows, while the network effect is positive and its magnitude is in line with previous studies (see Beine et al. 2011; Beine & Parsons, 2015; Bertoli & Fernandez-Huertas Moraga, 2015). This result corroborates with the consensus in the literature on diasporas as the most important dyadic determinants of migration flows. Controlling for heterogeneity at destination-year level leaves our results substantially unaffected. The inclusion of destination-year fixed effects in Column (2), meanwhile, does not alter the coefficients of any of the dyadic explanatory variables.
Table 2 Benchmark results: impact of cultural exports on the emigration rate These two specifications, however, do not tell us whether the effect of cultural proximity on the migration choice is only driven by historical and pre-existing cultural similarities. In other words, we cannot detect whether the evolution of cultural relationships over time plays a role in affecting migration decisions, as the time invariant component of cultural proximity, \({G}_{in}\), may act as confounding factor for the impact of\({A}_{in,t}\). To address this issue, we include dyadic fixed effects \({\mathrm{S}}_{\mathrm{in}}\) which control for all time invariant bilateral factors, such as geographic barriers and pre-existing cultural ties. The results reported in Column (3) suggest that the time-varying determinants of migration remain significant despite the loss of identification power due to the large number of fixed effects introduced. In particular, the network coefficient retains the positive sign, but it lowers considerably in terms of magnitude, with a semi-elasticity of approximately 0.09 and statistically significance at the 1% level. More importantly for our purposes, the evolution of bilateral cultural proximity over time emerges as a significant driver of international migration: a “positive shock” in cultural proximity represented by an increase in cultural exports by 10% leads to a 0.13% increase in the reverse bilateral migration rate after controlling for all the dyadic and time invariant factors affecting migration decisions. In other words, cultural attractiveness affects the migration choice over and above the pre-existing cultural similarities. The effect is not only statistically significant, but also quantitatively relevant. A simple back-of-the-envelope calculation shows that moving from the sample median to the 75th percentile of cultural exports, leads to around 165 more migrants per dyad. This roughly corresponds to an additional 29,343 international immigrants per destination, which is about 16% of the average number of migrants per destination in 2010. This sheds some light on the importance of accounting for the evolution of cross-country cultural relationships and their linkages with recent migration phenomena. For instance, the 41% increase in international migrants from 2000 to 2014 may at least partially be explained by a trend of cultural convergence associated with globalization.Footnote 19 Our results are consistent with such an interpretation. The last two Columns of Table 2 enrich the gravity specification by, respectively, adding non-cultural exports (Column 4) as additional control and decomposing exports in cultural products into the share of cultural products and total bilateral exports (Column 5). The findings suggest that only the time-variation of exports is positively associated with emigration flows, with the share of cultural exports having an impact above and beyond the correspondent aggregate flows.
The results hold when estimating the gravity equation with PPML (Table 3), which provides consistent estimates in the presence of heteroscedasticity and performs well when the dependent variable has a relatively large share of zeros (Santos Silva & Tenreyro, 2006, 2011; Bertoli & Moraga 2015). In our sample the share of zeros is rather small, it represents only 6% of the observations.Footnote 20 Despite some discrepancies in terms of magnitude with respect to the OLS counterparts, the PPML coefficients shown in Table 3 generally have the expected sign. More importantly, in line with our hypothesis, the impact of bilateral exports on migration seems to be predominantly driven by flows of cultural products. To further test the validity of our results, we estimate the gravity model with alternative econometric techniques such as Gamma PML and EK Tobit (Columns 5–6) which accounts for the zero migration flows. Although we cannot compare the performance of these estimators with high dimensional fixed effects, we find it reassuring that the estimates are in line with the results presented in Table 2.Footnote 21
Table 3 Robustness check: alternative estimators The results presented in Tables 2 and 3 are consistent with different sets of fixed effects and across econometric techniques. However, the reported estimates may still be biased because of reverse causality. To further address the potential endogeneity of trade in cultural goods we instrument \(ln\left({Xcult}_{ni,t}\right)\) with the average bilateral tariffs in the manufacturing sector applied by the importer and the correspondent imputed tariff revenues (Table 4). Our hypothesis is that governments set the level of tariffs to affect only trade flows, but not migration inflows i.e. we assume that both tariff-related instruments affect migration indirectly i.e. only through their direct effect on the endogenous variable. Figure 2 provides some empirical support to this conjecture, as the average bilateral tariffs in manufacturing appear to be very weakly correlated to average migration flows.
Table 4 Robustness check: 2SLS results The sample size for this IV exercise is reduced due to the tariffs dataset which does not provide information on all the country pairs included in our OLS sample.Footnote 22 Table 4 reports the IV results. As expected, both the average bilateral tariffs and the imputed tariff revenues have the expected sign and are strong predictor of exports of cultural products. The Kleibergen-Paap F statistic of the excluded instruments is way above the conventional level and indicates that the instruments are well identified. Then we use the Hansen J-statistic to test the exogeneity and we find a p-value equal to 0.46, which points to the validity of our set of IVs. The reduced form in Column (3) suggests a direct relationship between the instruments and the dependent variable. By combining the first stage with the reduced form results (Columns 2–3) we can cautiously conclude that the effect of both instruments on the dependent variable runs through the endogenous variable. Of course, bilateral tariffs in the manufacturing sector are also related with non-cultural trade flows. This relationship might weaken the validity of our set of instruments if non-cultural trade is related with emigration flows and not accounted for in our model (Eq. 2). To address this issue, we perform the IV analysis by including non-cultural exports as additional control in our specification (Columns 4–6). The statistics suggest that while the time-variation of non-cultural exports is correlated with cultural trade flows (Column 4), it does not significantly affect emigration from importing countries (Column 5). The latter finding is in line with the baseline estimates reported in Column 5 of Table 2. All in all, the IV results essentially confirm the positive relationship that emerges from our baseline estimates and add consistency to our predictions on the importance role of cultural changes in the emigration decision.Footnote 23
Further addressing the measurement error bias
Measurement error can bias the estimated impact of our parameters of interest. While the use of trade in cultural goods as proxy for CP has many advantages for the purpose of this analysis, there are potential concerns regarding its validity in reflecting national cultural contents.
For instance, American music labels might export records with non-American cultural content, so the imports of music from the US in some cases doesn’t necessarily affect the perceived attractiveness towards the US culture. By the same token, French exports of fashion products (included in the UNCTAD classification of "optional" cultural goods) may not only reflect French cultural content, but also a third country’s cultural content embedded in the fashion design that is performed before manufacturing takes place in France (see Fiorini et al. 2021). Further, custom data does not include digital transactions (i-tunes, Netflix) that accounted for a relevant share of transactions of several "core" cultural goods, such as DVDs, Music and Books. However, digital transactions have increased dramatically over the last 5–6 years, a period that falls outside our sample’s time coverage, so the latter source of measurement error is unlikely to largely influence our results.Footnote 24
To address the issues associated with measurement error in Table 5 we first compare the benchmark findings reported in Table 2 (Column 1) with the correspondent estimates obtained with the “core” UNESCO classification of cultural products (Column 4–5). The products identified by UNESCO as cultural goods are arguably characterized by a larger cultural content compared to the classification proposed by UNCTAD. They are therefore likely to better capture proximity in cultural tastes. However, as noted in "Appendix A3", UNESCO’s classification implies the use of a more limited time span and is less representative of the cultural products traded by the South. Given the shorter time coverage we are not including our full set of FEs since the more limited information in the UNESCO sample would create problems in terms of identification power. Hence, we compare the two classifications only with country-year fixed effects. The results indicate that using a different classification does not alter our benchmark estimates and our main conclusions remain unchanged. Lastly, in Column (2–3) we propose trade in newspapers and other printed matter as a more refined/accurate alternative measure of cultural proximity (see "Appendix A3" for more details on these product categories). The idea behind this is that newspapers are less subject to the global value chain bias described above, as their production is not dislocated to foreign countries. This therefore minimizes the potential concerns regarding the measurement error introduced by the gross nature of cultural trade. The results point to a positive relationship between cultural changes and emigration, which corroborate our baseline estimates.
Table 5 Robustness check: UNCTAD versus UNESCO classification Extensions
This section proposes two extensions to the analysis conducted so far. We test whether the role of the time varying component of cultural proximity changes (a) at different levels of pre-existing cultural similarities and (b) when we account for the long-lasting effect of cultural goods in favoring cross-cultural convergence.
Table 6 explores the variation of the role of cultural proximity on emigration for different levels of pre-determined cultural affinity and stages of economic development.
Table 6 Extension: countries’ development status and initial cultural similarities We first divide our sample according to the degree of cross-country cultural affinity based on linguistic and genetic distance (Columns 1–4) as well as the average volume of cultural exports (Columns 5–6). In order to preserve enough identification power and to attenuate the selection bias we split the sample into, respectively, almost identical subgroups using the median of fst_distance_dominant from Adsera and Pytlikova (2015), lp2 from Melitz and Toubal (2014) and the average value of cultural exports over our period of interest, respectively.Footnote 25 Taken together, the results suggest that time contingent shocks to cultural proximity only play a role when historical cultural similarities between country pairs are relatively weak. This finding suggests a non-linear effect for cultural proximity on migration over pre-existing cultural ties and a potential role for trade in cultural products in promoting cultural convergence.Footnote 26 In particular, the evidence is consistent with a relationship of substitutability between the time-contingent, asymmetric and time-invariant, symmetric dimensions of cultural proximity in triggering migration, with the former operating as a bridgehead between otherwise culturally distant countries. A plausible explanation is that the cultural content embodied in these types of products enhances bilateral cultural affinity through what Tabellini (2008) defines as the horizontal transmission of values. The consumption and diffusion of cultural goods in countries of origin can contribute to transferring exporter’s cultural traits, making the culture at destination better known, more attractive and more widely accepted.
In Columns (7–8) we test whether positive shocks in CP influence migration between country-pairs at different stages of development. To do so, we split the sample according to what is typically classified as North–North vs South-North migration and define as North countries all the member states of the OECD included in our sample. Interestingly, the estimates suggest that the effect of the time-variation of cultural proximity comes from South-North migration. In other words, a positive shock of VACP—other factors held constant—raises emigration towards countries characterized by larger income differentials. While this finding corroborates the results reported in Columns (1–6)—as we expect that the cultural distance between OECD countries and non-OECD countries may be larger than between OECD countries—it also suggests that the role of cultural proximity in reducing moving costs appears to be particularly important in developing countries, where budgetary and credit constraints are more binding. Finally, when looking at the impact of diaspora across sub-samples, it appears to be stronger for North–North migration. This result is in line with the literature on the role of networks in micro-founded gravity models, as the elasticity of the stock of emigrants generally increases when focusing on emigration towards OECD destinations (see Beine et al. 2015). In addition, we do not consider this evidence at odds with our hypothesis—given that a larger diaspora coefficient might be explained by the skill composition of networks (Felbermayr & Jung, 2009), for which we do not have data that fully cover our sample’s time-span.Footnote 27
While in this study we are employing cultural exports mainly as a proxy for “revealed cultural preferences”, we are not ruling out the cultural transmission channel of cultural trade (Maystre et al. 2014), i.e. that cultural content embodied in cultural goods can transmit and diffuse information on values, beliefs, habits and cultural traits of migrant destinations in importing countries. This process would in turn lead to a rise in emigration from importing countries through a progressive cultural alignment between origin and destination countries. In Table 7 we explore more closely this potential long-lasting role of trade in cultural goods in favoring cross-country cultural convergence and its indirect impact on the decision to migrate. More specifically, we test whether the intensity of long-lasting bilateral cultural relationships have a stronger effect on migration. We are well aware that the transmission of values which shapes the utility of would-be migrants takes time (see Cavalli Sforza, 2001).Footnote 28 For instance, the effect of traded movies on cross-country cultural pervasiveness is neither instantaneous or brief; rather, movies can be repeatedly watched and broadcast once purchased. Hence, our empirical strategy accounts for the recent history of bilateral trade relationships between \(n\) and \(i\) by simply considering the impact of the cumulative exports of cultural products from destination \(n\), so that:
$$CumXcult_{ni,t} = \mathop \sum \limits_{t - 1}^{t - s} Xcult_{ni,t}$$
(3)
Table 7 Extension: impact of ‘"Cumulative’’ cultural exports on the emigration rate This strategy allows at the same time to attenuate the distortion due to business cycle factors and measurement error associated with trade data. We initially set \(s\hspace{0.17em}\) = 5 while the third column reports the correspondent estimates with \(s\hspace{0.17em}\) = 9. Interestingly, as \(s\) goes up the impact of cultural exports tends to increase. In other words, when we add past bilateral cultural exported goods to \({Xcult}_{ni,t-1}\) the impact of our variable of interest on the decision to migrate at time t gets larger and larger. This finding is consistent with the hypothesis of a long-lasting effect of cultural products on bilateral cultural affinity.
Table 8 deals with the asymmetric dimension of time-contingent variations of cultural proximity in the context of international migration. To do so, we add imports of cultural goods to our gravity specification. We start with a parsimonious specification including only cultural imports in the model (Column 1), then progressively add other variables, namely imports of non-cultural goods (Column 2), exports of cultural (Column 3) and non-cultural goods (Column 4). Generally, the evidence indicates that the asymmetric component of VACP matters. Taken together, the results suggest that only the time-variation of cultural exports have a positive and significant effect on emigration from importing countries—i.e. the preference of the individuals in the importing country for the exporter’s culture appears to be the only direction of cultural proximity that influences emigration decisions from importing countries. In other words—according to our conceptual framework—while a rise of, say, cultural affinity for Mexicans towards US culture leads to higher emigration to the US, the time variation of US preferences for Mexico’s culture does not appear to affect Mexicans’ emigration decisions. Finally, when restricting the analysis to the within variation of country-pairs both directions of trade in non-cultural products do not influence emigration flows.
Table 8 Extension: asymmetric cultural proximity