A Gravity Model of Migration between Enc and EU

Due to ageing population and low birth rates, the European Union (EU) will need to import foreign labour in the next decades. In this context, the EU neighbouring countries (ENC) are the main countries of origin and transit of legal and illegal migration towards Europe. Their economic, cultural and historical links also make them an important potential source of labour force. The objective of this paper is to analyse past and future trends in ENC-EU bilateral migration relationships. With this aim, two different empirical analyses are carried out. First, we specify and estimate a gravity model for nearly 200 countries between 1960 and 2010; and, second, we focus on within EU-27 migration flows before and after the enlargement of the EU. Our results show a clear increase in migratory pressures from ENC to the EU in the near future, but South-South migration will also become more relevant.


INTRODUCTION AND OBJECTIVES
The free movement of workers is one of the fundamental principles upon which the European Union was once founded and, somehow, it is also present as a future goal in the bilateral negotiations with most neighbouring countries. As recognised in the Europe 2020 strategy, the European Union (EU) has a clear demographic challenge for the next decades. The EU will need to import foreign labour in response to gloomy demographic forecasts, in the context of ageing populations, low birth-rates, and prospects of a collapsing social security system, but it is also necessary to remain competitive in a global scenario and this means that we have to attract and retain the more skilled migrants.
This also requires improving the current control over migration flows and this is one of the reasons why the European migration policy was integrated into the European Neighbourhood Policy (ENP) from the very beginning. The EU neighbouring countries are the main countries of origin and transit of legal and illegal migration towards Europe. Moreover, their geographical proximity, economic, cultural and historical links make them an important potential source of labour force. In fact, nearly all Action Plans, the main tool of the ENP, contained proposals for actions in areas such as border management and management of migration flows. The EU proposed actions in the field of migration, asylum, visa policies, trafficking and smuggling, illegal migration and police cooperation.
The objective of this paper is to analyse past and future trends in ENC-EU bilateral migration flows. With this aim, two different empirical analyses are carried out. First, we specify and estimate a gravity model for nearly 200 countries between 1960 and 2010 and, next, we use the model to obtain medium-run forecasts of bilateral migration flows from ENC to EU; and, second, and in order to check whether our forecasts are consistent or not with previous evidence, we focus on within EU-27 migration flows before and after the 2003 enlargement of the EU.
The rest of the paper is structured as follows: first, in the next section, main trends in population and migration flows from and to ENC and Russia are described; next, the datasets and gravity models used in the analysis are shown and, last, we conclude with some final remarks.

POPULATION AND MIGRATION TRENDS FROM AND TO ENC
In this section, we provide a brief description of past trends in population growth and migration flows from and to European Neighbourhood Countries (ENC) plus Russia. With this aim, we use statistical data from the World Bank Development Indicators. As it can be seen from table 1, the population of the European Neighbourhood Countries (ENC) plus Russia is nowadays above 400 million people. While in the sixties of last centuries, the population in the ENC-South (Algeria, Egypt, Israel, Jordan, Lebanon, Libya, Morocco, Syria and Tunisia) was around sixty million people, a similar figure to the population in ENC-East (Armenia, Azerbaijan, Belarus, Georgia, Moldova and Ukraine), nowadays it is substantially higher: 204 million people vs. 75 million. The Russian population has also experienced a very important growth moving from 250 million people in 1960 to 420 million people in 2010. Population growth has been clearly higher in Russia and the ENC-South than in the EU-27 that has increased its population from 400 million people in 1960 to 500 million people in 2010.
As shown in tables 2 and 3, and according to data from the World Bank Development Indicators, there is a very high heterogeneity regarding migration trends in ENC countries during the last 50 years. While some countries such Israel during the whole period or Russia during the last thirty years have been net receivers of migration flows, other countries such as Belarus, Egypt or Tunisia have clearly lost population due to migration during the considered period. An additional interesting feature of migration from ENC is that it is highly concentrated in some destination countries due to geographical proximity or strong political, economic or colonialist linkages (see table 4). For instance, most migrants from Algeria or Tunisia go to France and most migrants from ENC-East go to Russia. In fact, one interesting result is that European Union countries are not always the main destination of migrants from ENC: for instance, emigrants from Egypt choose as Saudi Arabia as first destination, those from Lebanon prefer to migrate to the United States or those from Syria go to Jordan, Kuwait or Saudi Arabia. Migration flows between ENC has been quite relevant in the more recent period. Nowadays, about 10% of total population in ENC-East has been born abroad while this figure is around 5% in ENC-South and Russia. In the EU-27, the stock of foreign born population is around 10%. Total 331,209 85,414,766 110,537,797 142,514,909 172,587,534 203,790,715 Total ENC 128,263,707 156,474,474 187,351,345 223,797,604 250,950,902 279,306,781  Total 273,594,924,388,362,465,467,137,624,510

DATA SOURCES
It is a difficult task to collect data on homogeneous international migration for a large number of countries (Fertig and Schmidt, 2000;Crespo-Cuaresma et al, 2013). There are problems of data availability and difficulties in getting comparable statistical information across countries.  1960, 1970, 1980, 1990, 2000 and 2010. Over one thousand census and population register records are combined to construct decennial matrices corresponding to the last five completed census rounds. Immigrants are identified using the foreign-born criteria. The only problem with this dataset is that it provides information on stocks rather than on flows. However, migration stocks data have already been used by several studies such as Peri (2009), Brücker andSiliverstovs (2006) or Grogger and Hanson (2011) among others. Moreover, as highlighted by Brücker and Siliverstovs (2006), the analysis of stocks can be interpreted as a representation of a long-term equilibrium and, as data on immigration stocks are based on national censuses, they are probably of higher quality than those that report annual immigrant flows, as censuses deal with unambiguous net permanent moves and reduce the undercounting of undocumented immigrants.
Besides immigration stocks, an additional number of traditional variables related to pull and push factors of migration have been considered in order to explain migration flows and stocks. Table 5 summarises the different push and pull factors identified in the literature. The different determinants of migration are related to demographic, geographic, social, cultural, economical and political characteristics of both origin and destination countries. As our objective is not to explore the influence of the different push and pull factors on migration but to predict future movements, we only focus on a subset of these factors. In particular, and following a similar approach to Kim and Cohen (2010), we investigate the role of demographic, geographic, historical variables and relative differences in GDP per capita. Data for these additional variables have been collected from the CEPII Geodist dyadic dataset (Head et al., 2010) and the CEPII gravity dataset (Head and Mayer, 2013). Geographical distance has been defined as the distance between the two capital cities of immigrants' origin and destination countries using the great circle formula for cities' latitude and longitude. The area in km squared of the origin and destination countries are also considered. Dummy variables indicating whether the two countries are contiguous, share a common language, have had a common colonizer after 1945, have ever had a colonial link, have had a colonial relationship after 1945 or are currently in a colonial relationship have been included. There are two common languages dummies, the first one based on the fact that two countries share a common official language, and the other one set to one if a language is spoken by at least 9% of the population in both countries.
GDP and population data from the CEPII's gravity dataset have been updated using data from the World Bank Development Indicators and the same definitions as in the original source. observations. However, when GDP differences between destination and origin countries are considered the sample further reduces down to 141,112 observations. As previously mentioned, while the main aim of our analysis is to analyse the potential role of ENP, it is also interesting to analyse the effect of recent EU enlargements on migration flows from the new members to the EU. In particular, we use data from the EUROSTAT project

EMPIRICAL ANALYSIS
There are many theoretical hypotheses and models concerning the determinants of migration. Gravity models were initially based on Newton's gravity law, but recent contributions have also provided the microfoundations in the context of migration analysis (Grogger and Hanson, 2011). These models have been widely used in the empirical analysis of migration due to their relatively good forecasting performance (Fertig and Schmid, 2000;Karemera et al, 2000 or Kim and Cohen, 2010; among others). In particular, migration stocks or flows between two countries are supposed to increase with their size and decay with the distance between the two countries. Usually, the most representative variable of the size of countries is population.
Therefore, it is expected that migration be a positive function of population size of the host and home country and a negative function of distance (which controls for migration costs). As Santos- Silva and Tenreyro (2006) and Martinez-Zarzoso (2013) highlight, the most common practice in empirical applications has been to transform the multiplicative gravity model by taking natural logarithms and to estimate the obtained loglinear model using Ordinary Least Squares. One problem with this approach is how to deal with the potential presence of zero bilateral migrant stocks. As argued by Llull (2013), based on the law of large numbers, theory predicts that all bilateral stocks will be positive, though some may be very small. In finite populations, however, zero migration stocks may occur, if bilateral migration probabilities are small. In fact, in our sample, and due to the high number of considered countries, the presence of zeros is relevant accounting for around 55% of total bilateral observations. In order to estimate the log-linearized version of the gravity model, we have replaced the 0 values by a very small value (1) and then transform the variable into logarithms.
Usually gravity models are enlarged with additional variables related to different pull and push factors briefly discussed in the previous section (see, among others, Volger and Rotte, 2000;Hatton and Williamson, 2002;Gallardo-Sejas et al., 2006;Mayda, 2010;or Ortega and Peri, 2013). We also include in our specification year fixed effects, to control for common time shocks, and origin and destination country fixed effects to account for time-invariant unobserved heterogeneity. The importance of adding country fixed effects in the gravity model specification is noted by Bertoli and Fernandez-Huertas Moraga (2013), who argue that specifications without fixed effects may suffer biases due to the Multilateral Resistance to Migration.
Taking all this into account, our model specification is as follows: log = · log + · log + · log + · log + · log + · contiguity + ' · ( )* +, --+ . · ( +* +, ℎ+ + 0 · ( * +1 + 2 · ( )( * + · ( *45 + · log 5 6 ( 6 ( 7 + -8 9 --( + ; where log(M ijt ) denotes the logarithm of the stock of immigrants from country i (origin) in country  (2013) and Llull (2013). The coefficients associated to the year dummies also provide some interesting results. In particular, after controlling for the effect of demographic, geographical and social/historical characteristics, migration stocks have significantly increased when compared to the 1960s, similar results to those found by Massey (1999) and Kim and Cohen (2010). However, the economic crisis has deeply affected international migrations (Tilly, 2011): the value of the coefficient associated to the 2010 dummy is positive and significant but its value is similar to the one estimated for the 1980 dummy.
In model (2) (1), the stock of migrants is positively associated with relative differences in GDP per capita. This result shows that better economic opportunities positively affect migration.
In order to have a better description of migration patterns from and to ENC countries, in model (3) of table 6 origin and destination country fixed effects are replaced by dummies representing different groups of countries. In particular, origin and destination countries are grouped into five categories: EU, ENC-East, ENC-South, Russia and the rest of the world that will be used as the reference category. The results show that the EU has received and sent more immigrants in the considered period than the rest of the world even after controlling for demographic, geographical, cultural/historical and economical variables. ENC-East, ENC-South and Russia have also sent more immigrants than the rest of the world, but they have received significantly less. In Table 7, the same specification of the model is re-estimated but now looking at specific destination. While model (1)   has sent more immigrants than the rest of the world at the beginning of the period, but there is a clear downward trend. The opposite has happened when we looked at the EU as a migration destination: the EU has become much more attractive than it was at the beginning of the period.
ENC-East, ENC-South and Russia have sent more immigrants than the rest of the world, but the trend is negative. However, as destination countries, the trend for ENC-East and Russia is positive and not different from the rest of the world for ENC-South. When we look at models (2) to (5) in table 8 where different destinations are considered, no significant differences are observed when compared to the same models in table 7, so the previous results are stable across time and can be interpreted as evidence of the stability of the model in order to obtain bilateral migration forecasts.
In table 9 we present the results of a forecasting exercise using model (2)  From this table, we can see that migration from ENC countries to the EU will increase in more than 675,000 migrants (9%) with higher increases from ENC-South and Russia. It is worth mentioning that there is a high heterogeneity in the forecast, but also that the share of emigrants from ENC to the EU will fall from 23.6% in 2010 to 21.7% in 2018, a figure that reinforces the increase in South-South migration in the next years.

FINAL REMARKS
The objective of this paper was to analyse past and future trends in ENC-EU bilateral migration flows. With this aim, we have provided some empirical evidence on population and migration trends in ENC and, next, two different empirical analyses are carried out. First, we have specified and estimated a gravity model covering around 200 countries and used the model to obtain medium-run forecasts of bilateral migration flows from ENC to EU; and, second, and in order to check whether our forecasts are consistent or not with previous evidence, we have focused on within EU-27 migration flows before and after the 2003 enlargement of the EU.
The descriptive analysis of population and migration trends in ENC countries has shown some interesting results. First, the population of the ENC has increased in 170 million people between 1960 and 2010 while the EU-27 has increased its population only in 100 million. Second, there is a very high heterogeneity regarding migration trends in ENC countries during the last 50 years. While some countries such Israel during the whole period or Russia during the last thirty years have been net receivers of migration flows, other countries such as Belarus, Egypt or Tunisia have clearly lost population due to migration. Third, migration from ENC countries is highly concentrated in some destination countries due to geographical proximity or strong political, economic or colonialist linkages.
Our analysis of the long-run determinants of bilateral migration stocks has permitted us to conclude that demographic, geographical, social/historical and economic factors are relevant both to explain and to forecast migration patterns. Our results have shown that once these different pull and push factors are controlled, migration flows from ENC countries to the rest of the world are higher than they should be according to the model. When we concentrate on flows from ECN to the EU, this "surplus" in migration is even higher. This result shows the strong ties between these countries and the EU and how the ENC could clearly increase migratory pressure from these countries in the future. In fact, our medium-run forecasts show an increase in migration from ENC countries to the EU will increase in more than 675,000 migrants (9%) with higher increases from ENC-South and Russia. It is worth mentioning that there is a high heterogeneity in the forecast, but also that the share of emigrants from ENC to the EU will fall from 23.6% in 2010 to 21.7% in 2018, a figure that reinforces the increase in South-South migration in the next years. The analysis of the short-run impact of EU accession by the Czech Republic, Cyprus, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovenia and the Slovak Republic in 2003 on migration flows both as origin and as a destination have provided a benchmark that is also consistent with our forecast regarding ENC countries.
Regarding future directions for research, the availability of the compiled data set on bilateral migration stocks and several determinants can serve as a starting point to enlarge our benchmark specification with other variables that are potentially interesting in the context of the ENP. For instance, indicators on quality of governance or other institutional determinants could be included as additional explanatory variables and different scenarios regarding institutional convergence with the EU could be considered in order to assess the future evolution of migration from and to ENC.