Apart from many geopolitical factors (wars, natural catastrophes, search for political asylum), economic factors play a very important role at determining international migration inflows and outflows. A very wide literature has highlighted many of them, such as wage differentials, employment conditions, migration costs, both pecuniary and cultural, the diversification of risk within family members, differences in relative prices between host and home country, the accumulation of human capital, the improvement of the health status, or the willingness to reach a savings target to overcome capital constraints in the home country (Massey et al. 1993; Dustmann and Weiss 2007). More recently, global value chains have also generated a new sort of temporary movements. Finally, the consolidation of international migration is usually related to network effects that support transnational movements generated by particular institutions in the host and the receiving country.
In order to analyze the Spanish case, we focus on the differential of employment conditions between origin and destination countries. Our strategy is similar to Grogger and Hanson’s (2011), who relate the log odds of residing in country h for a person from country s is thought to be determined by absolute differences in earnings between the two countries and by the cost of migrating that is idiosyncratic to that particular country of origin and destination pair. However, when considering migration to or from Spain, earnings do not appear to be a good proxy for economic opportunities, since high unemployment rates have been prevalent, ranging between 8 and 26 % in the period 2000–2015. Indeed, changes in unemployment appear to be more appropriate than changes in wages to measure how economic opportunities evolve over time, particularly when, as has happened during the current recession, wages have reacted slowly to the worsening of the economic situation due to significant real and nominal rigidities.Footnote 7 Since Harris and Todaro (1970), many other papers in the literature have included unemployment rate differentials as either push or pull factors to explain migration flows (see, for instance Zavodny 1997; Pedersen et al. 2008; Beine et al. 2013; Bertoli et al. 2013a).
In the spirit of Grogger and Hanson’s (2011) approach, migration flows are assumed to respond symmetrically to changes in relative economic opportunities, so that the effects on migration flows of a change in relative economic conditions in one particular country should disappear completely when the initial economic conditions are restored. In this regard, the recent experience in Spain shows that this might not be the case. The sharp drop in unemployment (from around 23 to 11 %) between 1995 and 2001 drove up the share of foreigners in the Spanish economy. Subsequently, unemployment remained roughly constant, but the foreign population continued to grow. In 2007, the Spanish unemployment rate headed up again, but the foreign population continued to increase, before declining slightly in 2012.
It seems, therefore, that analyzing changes in the stock of foreign nationals in Spain in the current situation needs a more flexible specification. For that reason, we estimate the effects of economic conditions on both entries of foreigners and exits of Spaniards separately (see Fig. 1). The sharp decline in unemployment between 1995 and 2001 prompted an increase in the number of inflows of foreigners in Spain. These inflows continued to rise until 2007, even though the unemployment rate was quite steady, due to decreasing migration costs and as result of the increase in the stock of migrants in Spain that precluded a generalized preference for Spain rather than other alternative destinations (Bertoli et al. 2013b). As from 2007, inflows decreased as unemployment rose. It is noteworthy that the impact of changes in unemployment on migration inflows is similar to that observed in the 1990s, but at a higher level of unemployment. In turn, in these last 4 years, outflows of both foreign and nationals have also increased as unemployment has risen at a similar pace.
One would be tempted to ask whether the current level of outflows of Spaniards has been enough to decrease the costs of emigration for this particular socioeconomic group as it happened for foreigners between 2001 and 2007. In particular, Fig. 2 illustrates that after a small decline in unemployment rates in 2014, emigrations of Spaniards born in Spain kept growing whereas those of foreigners already declined. In particular, the emigration rate of Spaniards born in Spain (per thousands) increased in 2014 from 1.2 to 1.35 whereas that of foreigners decreased from 83 to 65 (for further details, see Izquierdo et al. 2014).
This descriptive evidence suggests that even though bilateral inflows and outflows might respond to economic conditions as theory predicts, changes in costs of migration may blur the contemporaneous responses of the stock of migrants to economic conditions somewhat. To test quantitatively the importance of these mechanisms, we follow Bertoli et al (2013a) and relate the log odds of immigrating (I) (fraction of entries from one country of birth to one region divided by the corresponding population residing in that country of origin).Footnote 8 We also do the same for emigrating (E) (fraction of exits from one country of birth to one country of destination divided by the corresponding population residing in that region) to unemployment differentials (U) and to the costs of immigration/emigration between origin (h) and destination (s).Footnote 9 Using the superscript f to denote foreigners and e to denote Spaniards, our regression specifications are:
$$ \ln {I}_{hst}^f={\alpha}_0+{\alpha}_1\left({U}_{ht}-{U}_{st}\right)+{\lambda}_t{c}_t+{\lambda}_h{c}_h+{\lambda}_s{c}_s $$
(1)
$$ \ln {E}_{hst}^e={\gamma}_0+{\gamma}_1\left({U}_{ht}-{U}_{st}\right)+{\kappa}_t{c}_t+{\kappa}_h{c}_h+{\kappa}_s{c}_s $$
(2)
In subsequent specifications, we will allow different elasticities for the unemployment rate at origin and destination. We proxy the cost of emigration (c) in different ways (separate dummies for each origin country and for each destination region or dummies for time dummies coupled with a dyad country/region). We also include as covariates the time dummies (c
t
).
The literature has identified two problems with specifications such as (1)–(2): first, the potential bias introduced by the existence of zeros in emigration/immigration rates for a sizeable group of country pairs over time and second, multilateral resistance to migration, that is, the existence of positive correlation between unemployment rates of alternative origins.Footnote 10
Regarding the problem of inexistent flows, we consider bilateral movements between a given country and a Spanish region (Comunidad Autónoma) in one particular year during the period 1998–2012 for entries and 2008–2012 for exits. In our databases, considering both entries and exits of foreign and Spanish nationals, more than 15 % of the cells are nil. To avoid the problems associated to the use of Poisson’s methods to treat the problem of zeros, we eliminate the cells with smaller flows from the sample.Footnote 11
Regarding multilateral resistance to migration, autocorrelation of residuals in (1)–(2) cannot be ruled out. In consequence, the estimated coefficient of unemployment in the origin country might be upward biased. To solve this problem, we follow Bertoli et al. (2013a) who add as an auxiliary variable the cross-sectional (over countries) average of the dependent and independent variables, using monthly observations, to incorporate the changes in the willingness to migrate to alternative destinations (common correlated effect (CCE)). We will do the same averaging over countries and regions of destination. If there is correlation between unemployment rates of alternative origin countries, the estimated coefficient of the impact of unemployment rate in the origin country on foreign migration flows is biased upwards. This is the case because we do not observe all bilateral movements, and the flows between two different origin countries that present a certain correlation in unemployment affect the flows between that country and Spain. In principle, this problem should be less problematic for Spaniards since we have all the relevant information for this particular group and internal migration is not very much affected by unemployment differentials (Bentolila and Dolado 1991; Antolín and Bover 1997). In order to check the above hypothesis, we also estimate Eq. (1) on internal migrations showing that they do not increase with unemployment differentials across regions.Footnote 12 Multilateral resistance to migration is not relevant to identify the coefficient of the unemployment rate of the regions in Spain. Indeed, there is a high correlation of unemployment among Spanish regions, and it is likely that foreigners in a first stage choose Spain and decide to go to one particular region balancing many other reasons such as the size of the community of foreigners there.Footnote 13 When we apply this CCE methodology at the regional level in (1), the autocorrelation disappears for foreign entries.
Finally, to check the endogenous component of migration inflows, we will relate them to the decrease on the costs of emigration once there are networks of migrants abroad. In 1995, there were almost no foreigners in Spain, but the increase in foreign population took place very rapidly, so that network effects seem likely to have operated. Thus, we extend the specification of the immigration equation for Spain to include the lag of the stock of migrants of the same nationality (or who depart from a particular region in Spain) who reside in the corresponding region in Spain (or who reside in the corresponding potential destinations abroad) (S
hst−1).Footnote 14
$$ \ln {I}_{hst}^f={\alpha}_0+{\alpha}_1\left({U}_{ht}-{U}_{st}\right)+{\alpha}_2{S}_{hst-1}+{\lambda}_t{c}_t+{\lambda}_h{c}_h+{\lambda}_s{c}_s $$
(3)
$$ \ln {E}_{hst}^e={\gamma}_0+{\gamma}_1\left({U}_{ht}-{U}_{st}\right)+{\gamma}_2{S}_{hst-1}+{\kappa}_t{c}_t+{\kappa}_h{c}_h+{\kappa}_s{c}_s $$
(4)
When running the abovementioned specification, it seems that there is no enough variation in the stock of migrants left when we control for both time dummies and the pairs of country of origin and region of destination. This is the case because during the period of analysis, and once we control for the average location preference of any country of origin, all stocks increase over time. Therefore, we will run specification (3)–(4) only including separately time, country of origin, and region of destination dummies.
Since the stock and the flows of migrants might not be fully consistent because they come from different data sources, and given that we have a pretty large time series (15 years) for inflows of foreigners, we also run a dynamic panel model to analyze the endogenous component of those inflows, as follows:
$$ \ln {I}_{hst}^f={\alpha}_0+{\alpha}_1\left({U}_{ht}-{U}_{st}\right)+{\alpha}_3 ln{I}_{hst-1}^f+{\lambda}_t{c}_t+{\lambda}_h{c}_h+{\lambda}_s{c}_s $$
(5)
We use the results as a consistent device to simulate the decrease in unemployment rate necessary to reduce inflow rates for Spaniards born in Spain. In order to estimate (5), we have to take into account the typical problems of estimating a dynamic panel with fixed effects and that is the reason why we will follow Arellano and Bond (1991) instrumenting the autocorrelation term with past immigration flows. In this case, since the regression is estimated in differences (6), it does not matter whether we include separate country of origin and region of destination dummies or the pairs of country of origin/destinations.