Vol 20, No. 12;Dec 2013
A multiple regression model for migration in the EU, 20052012 Ascending flow from East to West Bálint Molnár* (Corresponding author) and Mária Lakatos** *Eötvös Loránd University, Faculty of Informatics, Information System Department , Pázmány Péter sétány 1/C, Budapest, 1117, Hungary , Tel: +36 1 372-2500 / 8042 , e-mail:
[email protected] **Budapest University of Technology and Economics, Faculty of Economic and Social Sciences, Department of Finance 1117, Budapest, Magyar Tudósok körútja 2. Tel: +36 1 4632342
[email protected]
Abstract: In 2004, when 10 new EU member nations entered the labor force, a potentially homogeneous market emerged: an irresistible magnet to EU and non-EU immigrants alike. Therefore, analysis of the impact of this phenomenon seems inevitable, specifically an analysis which enables social welfare systems to adjust their budgets in response to the rising influx of immigrants. In order to accomplish this, one must 1) analyze the direction of migrant flow; 2) to show the main factors for motivation to migrate. The authors has created a model that explicitly and implicitly incorporates the complex mutual dependencies. This model should also enable us to make predictions, specifically short-term predictions which are reasonably accurate. For this analysis, a multiple regression model with meaningful regressors were selected, which shows the hidden dependencies. We focused our study on two main factors, firstly the differences of social welfare systems within the EU and often referred as the motivation for social tourism and secondly the unemployment rate as regressor data. Once the model's forecasting accuracy has been established, the model can also be used for short term predictions that underpins the empirical observation of migration flows. Keywords: Migration, social tourism, multiple regression model 1.
Aim of the study
The migration process has received the increasing attention of politicians and economists, who have emphasized the advantages of newcomers over ageing societies. However, the point of view changed during the 2008 economic crisis: nowadays the main focus is different. The issue is how to measure the number of potential migrants and highlight their motivation: why do people decide to move to another country? They have an abundant variety of reasons, so during our research we analyzed the recent situation in the EU to discover whether only greater job opportunities and higher wages attract millions of newcomers from Europe's poorer side to its most developed side. We wondered whether this process would accelerate or not. If this tendency remains unchanged during the next few years, or speeds up, not only will the brain drain cast its shadow on these countries, but tax base erosion will follow, creating a burden on the central budget equilibrium, and making the fight against growing deficits in both origin and receiving countries very difficult (Ben-Gad, 2004). If the immigration flow did not achieve its basic aim (luring young people to the receiving country), then the societies of the host countries might not diversify and maintain their flexibility, and income taxes paid by immigrants would not help the host society to cover its future public expenses, i.e. mainly the age-dependent retirement pensions (Lee and Miller 2000). Furthermore, the remittances that immigrants sent back to their homelands would not finance
255
[email protected]
Vol 20, No. 12;Dec 2013
development in the home countries, raising income from the existing balance of payments.(Ratha and Shaw 2007) But if the number of immigrants from EU10 rises fast in the short term, EU15 governments might concede to their voters' demands and use administrative tools to restrict the inflow of legal and illegal immigration. (Recently a new direction occurred in analyzing migration in CEE countries: because of the income premiums for work experience at home, emigration from CEE countries was followed by return migration [Martin, Radu 2012]).
1.1. Introduction Europe and the Middle East count altogether 43 million people as part of an estimated worldwide 215.6 million immigrants. (World Bank 2011a). Loosely speaking, this means that 3 percent of the world's population lives in a country other than their birth country, while 10 percent of the total population of Europe and the Middle East is working abroad. Dramatic changes have taken place in migration in European countries; the most significant one is that Russia as one of the host countries is offering asylum to the former Soviet Union member states' migrants, who escape from civil strife or political upheaval, or simply seek better economic opportunities in their former homeland. Lagging behind the USA in pole position, Russia received 12.3 million immigrants till 2010. Both former soviet countries are within the top ten of emigrant origin countries as well. They suffer from a special type of population change, and basically differ from the other emigrantimmigrant flows in Europe. (MIRPAL, 2011) Europe and the Middle East are grouped together by the World Bank migration reports, and neither the World Bank (World Bank 2011b) nor the OECD (OECD 2011) offer a coherent database for evaluating migration inside the EU. The European Union forms a politically and economically separate unit, which is why we have mapped the regional differences in the directions of immigration and emigration flow within this closed circle, and showed their motives. (Jennissen, 2007) Using a traditional statistical approach, multiple regression modeling, we have proved that the number of emigrants from EU10 will grow in the middle term, mainly because of the countries development and social welfare gap. The migration movement would reflect immigration from third countries mainly to Spain, to Portugal, and the United Kingdom, but there are no statistics about the size of the flow of illegal workers. To measure the EU internal migration, we assume the level of third-party migration to the EU remains unchanged. Our study relates to the studies published starting in the early 60's, when social and political research focused on the phenomenon of globalization and its subsequent results. The main areas for analysis were the USA, Australia and Canada; they were the traditional receiving countries for Asian, Caribbean, and African refugees. (Massey, Arango, Hugo at al 1993). In one decade, the group of host countries extended to Spain, Portugal, Italy, and Greece, because the aforementioned emigrants tried to find a much closer destination country; therefore, the former origin countries became receivers. The opening of the borders in the 90's stimulated a new era of immigration, as citizens of the former socialist countries sought a better life in the EU. The financial crisis in 2008 briefly reduced the number of immigrants. Furthermore, some of the EU 15 countries: Great Britain, Germany, Spain, Italy, and Austria, responded with special restrictions, mainly to control illegal immigration. For instance, since the crisis, Spain has made burdensome the process of hiring migrant workers (Peixoto 2009). Another example is Great Britain, which tries to prevent immigrant workers from bringing their families into the country if they are earning less than a prescribed amount. Some other countries have taken steps to reduce labor migration from non EU countries, by cutting back labor migration quotas, or creating an administrative barrier for employers, i.e. an enterprise that wishes to hire employees must first prove that there are no appropriate native workers, even if the potential foreign employee comes from another, less developed country. Following these
256
[email protected]
Vol 20, No. 12;Dec 2013
measures, the arrival of immigrants mainly from Third World countries has reached not hundreds of thousands, but tens of thousands year by year. (World Bank 2011c) But according to the latest data, the emigration from EU10 countries has begun to rise again, but the roles have changed; European countries which had been worker-exporters to the most developed countries, have become the main destinations of immigrants. Inevitably, for putting the brake on migration, or at least controlling it, its causes must be investigated. Studies in literature focusing on migration use various methods to examine the same thing: why migration begins, and how it can be slowed down. (Todaro, and Maruszko, 1987). Neoclassical economists focus on the gap between wages and unemployment rate for the origin and receiver countries. They even take into account the cost of migration, earnings against migration cost, and risk. This group of experts seeks the motivation of the individuals as incomemaximizing units and views migration as a micro-level decision in the household taken by individuals or the whole family, as small units. Another group of experts, namely the dual labor market theory supporters, ignore the microlevel decision making process (Piore 1986). They focus on the problem of globalization and explain migration by the structural changes and job market demands of highly developed countries, which motivate workers from underdeveloped countries. (Massey at al 2011) Families on the same level form a community that will make the decision, albeit as separate units. These different points of view explain the motivation of migration flow individually, (Dumont at al, 2010) on the family level, nationally, and internationally, in addition to influences on national migration policies. Generally, however, the migration of workers is explained by differences in wage rates between countries. Furthermore, the elimination of wage differentials could stop the movement of labor force from underdeveloped countries to developed ones. That is because mainly the labor market demand induces the flows of immigrant workers, and the social welfare system plays an unsubstantial role in it. In spite of macroeconomic models, the individual decisions on migration issues remains on the microeconomic level. Individual decisions as whether to migrate or stay are based on cost–benefit calculations. In this approach, the social welfare system appears as a risk minimizing factor. (Stark 1991) and (Taylor 2007). Individuals who join groups which calculate the cost and benefit of the migration will act differently even in the same country. This more sophisticated approach clarifies the motivation of a family in migration; households are in a position to control the risks of their living standards, and will diversify the household resources to maximize the benefit and minimize the risk. Migration is just one of many potential methods. Households send family members abroad not only to improve their living standards, but also to increase income relative to other households, that is, to reduce their degree of deprivation compared with others. (Stark and Taylor 1991). Recent research has revealed a new model of migration within the EU as a whole. (Düvell, 2012) The European Union offers unique opportunities for its citizens to work abroad, but some studies (Ekberg 2011a) have revealed that immigration will cause an increasing burden on the public welfare system instead of helping the ageing society to cover its future costs. The fiscal impacts of immigration flow depend on its size, the age structure of foreign-born persons, and the level of their labor market integration. Ekberg says the future fiscal impact of immigration might be negative, and the immigrants’ contributions to the public sector will be reduced by the increasing costs of social expenses that are generated by the additional population coming along with immigrants. For the long term, two factors determine the positive or negative effect of net contribution to the public sector: the number and the age differences between natives and immigrants, and differences in their employment situation. According to the latest studies on migration, the unemployment rate of the immigrants is considerably higher than that of the natives within a destination country of migration, while their share from social security allowances is high also. Moreover, the immigrants with a higher unemployment rate will contribute less in taxes and social security than they receive from the public expenditures.
257
[email protected]
Vol 20, No. 12;Dec 2013
Several studies have tried to measure the fiscal contribution in the social sector. They revealed a reciprocal effect (Simon 1984) during the 80's when there was a positive effect on fiscal balance via taxes and social security, while in the 90's the situation changed dramatically. Blau observed (Blau 1984) a neutral impact which transferred a negative effect (Weintraub 1984) in the U.S. Their view was shared by Straubhaar and Weber (1994) in Switzerland also. Greater positive fiscal effects, namely considerably more public revenue from immigrants, would demand a higher level of immigration, which is thought to be unrealistic.
2. Direction and magnitude of migration within the eu on a multilateral basis In our analysis, which was compiled on the basis of empirical evidence, we wanted to predict the direction and magnitude of migration occurring only in the European Union in an economic context. (Jennissen, 2007) (Note 1) To achieve these targets, we complemented the bilateral approach with multilateral aspects and created an EU multilateral migration matrix that shows the individual's choice in emigration decisions based on special parameters. The EU community has a special framework for immigrants, so any of the models reflecting the situation should include these features. Financial conditions on the labor market vary, but other conditions are the same in the EU labor market, because the community's citizens enjoy the same rights and obligations as natives of the host country, such as individual decisions affecting their livelihood is influenced by salaries and social allowances. The motivation for migration - we predicted on the base of earlier analyses - was linked with relative differentiation between the income level of separate EU countries citizens to the EU average. In the micro-economic approach, the relative gap on the individual level is one of the main motivational factors we considered, but our hypothesis widened the number of explanatory and economic factors used for the matrix. The reliability of the forecast called for the use of geopolitically and economically limited units, since data show that the European Union migration is significantly different from the inflow from the Eastern part of Europe, or the traditional NorthSouth, East-West route characteristics. According to the definition in literature, those who belong to an immigrant group are people who were born in another country, or citizens of any other country than where they currently live (Note 2.). For the purpose of analyzing federal taxes, citizenship remains neutral. That is because the tax liability and the related social security charges are determined usually by place of residence (OECD 1977). Furthermore, employees’ salary is taxable from the very first day of employment. We accumulated 25 potential independent xijn (predictors, regressor variable) assuming a linear approximation of the relationship between xijn and yin (response) variables, where i is the year, j is predictor type, and n is one from 27 EU countries, and yi means disparity between immigration and emigration in the predicted year (net migration) for EU. A linear relationship was proved with scatter plot matrix, and we created several models with different standard macro-economic parameters, while using the Backward elimination method to reduce the number of variables. The models with the different predictors xijn have the following assumptions: The EU as a community forces its member states to harmonize their economic and financial policies, budget revenues and expenditures, and social redistributive policies, including social care systems. One of the pillars of the EU is non-discrimination, and within the member countries, individuals as well as business units enjoy the same rights. For migration decisions, however, this framework significantly reduces the risk of emigrants, and as opposed to micro-economic conclusions, the risk of all individuals is at nearly the same level. Consequently their decision will be affected primarily by the income gap between sending and receiving country citizens, however this factor is not confined only to wages, but also reflected in another figures, like in the money from social benefits as well. Taylor and Patrick underlined the yield from the social welfare system as a decision-making factor, but considered only its risk minimizing role. We accepted and used this result. We considered the social benefit as a direct influencing factor on the income level, instead of insisting on its indirect, risk-minimizing role. 258
[email protected]
Vol 20, No. 12;Dec 2013
Summarizing our perception, since the level of income depends in equal parts on wages and public transfer payments, the latter factor might be considered as a dependent’s income. Welfare expenditure is comparable on two bases, firstly the per capita base, where expenditure contains: social benefits, which consist of transfers, in cash or in kind, to households and individuals to relieve them of the burden of a defined set of risks or needs; administration costs, which represent the costs charged to the scheme for its management and administration; and other expenditures, which consists of miscellaneous spending by social protection schemes (payment of property income and other). Among those for whom access to the social welfare system is not equal, the cost per capita method can cause distortion. Since the labor market situation is still weak in Europe, with some exception, the unemployment rate among immigrants is higher than for natives, and the access to retirement is limited (eligibility for pension is earned with a fixed number of employment years, which usually is not fulfilled). Two optional ways exist for using data, firstly cost per capita figure – this approach not taking into account the different access levels, and secondly, restriction of the model to the family's direct beneficiary from the social welfare system. This approach is independent of the labor market status of the individual, e.g. the child care allowances. If the access to child care allowances is equal in this geo-political entity, it seems obvious that direct beneficiaries play a fundamental role in the micro-level decision making process, in addition to wages and the unemployment rate as indicators for the labor market. However, the politically affected exodus has already penetrated the EU, so the movement of the population reflects individuals' decision-making hierarchies based on lower risk. The group of predictors combine micro and macro level factors, and first of all, minimum wage. On the other hand, individuals' incomes might be compared with each other only after deducting taxes. Usually the tax wedge offers the basis of comparison for regular distribution of income among the different income groups of a population. The income distribution is not regular among immigrants; all studies have proved that immigrants typically work at lower wages, generally just about minimum wage level. A number of studies have been performed to prove (UN, 2010) that immigrants can find only seasonal work in agriculture or tourism, and have not yet been fully integrated into the labor market. The employment gap between natives and immigrants has been widening; non-native workers have often been concentrated in sectors that have experienced the most serious decreases during the financial crises (Migration Policy Institute 2010a). The high unemployment rate in the most vulnerable group keeps the wages under pressure. Average wages of immigrants fluctuate around minimum wage level–and not only in Europe. (Newland 2010) Competition in tax rates, or effective tax rates, should motivate individuals in their decisions. If the minimum wage is exempt of taxation, the impact of the tax rates is neutral, so we did not use deduction. Afterward we have included the social welfare system as regressor factors. Secondly, to avoid the statistical bilateral approach, our model includes all EU member countries, as their citizens are free to move, and enjoy the same legal rights all over the EU. To achieve a dynamic model, we incorporated the predicted yi-1,n factor (i=years) as independent regressor in the i year, using the predicted value as fact for the following year.(DeMaris 1946) The multiple regression model was based on differentiation of country data xijn from the EU average level, so that we could exclude the basic trend, and show the multiple connections among yin and xijn factors, (labor market, unemployment rate, minimum wage, and social allowances cost). The model indirectly includes such non-numerical features as administrative burdens, which will limit the labor force flow from one country to another. 2.1 Preparation of the data set Migration research studies focus on country level databases collected by the World Bank, the IMF, and the United Nations, using each other’s data. They contain some inconsistent data received from member countries, or data which was missing and was subsequently restored by estimation. We have used the World Bank latest Migration and Remittances Factbook, 2011, IMF Balance of 259
[email protected]
Vol 20, No. 12;Dec 2013
Payment, Migration and Remittances Unit calculation. To acquire the minimum wages figure, ILO offers the widest database, and social cost per capita and child care allowances data are published regularly by Eurostat, in the ESSPROS database. Our model consists of 27 lines, equivalent to the number of member countries in the EU, and columns with different macro and microeconomic data, (see below in Methodology) on 2010 databases, which are the latest ones. (Note 3.) The preparation included operations such as characterizing, cleaning, and transforming the data. Particular care had to be taken to determine whether sub-setting the data was needed to simplify the resulting models. A series of alternative models were explored, since all models work well in different situations 3.
Theses
In our analysis we were looking for the answer to whether a time series prediction model can be established, at least for short term, to quantify the main directions of migration for the next three years, based on indicators of a country’s overall economic situation such as GDP per capita and the annual rate of unemployment as macroeconomic data, as well as variables detectable and comparable at the individual level such as the minimum wage and automatic entitlement to family support. Within this topic, we were also looking for the answer to whether among social benefits, automatic child support in itself, and by that, minimum family income taken in a broader sense (i.e., family benefit, after two minor dependents in a one earner family with close to minimum wage income), would influence the willingness to migrate within the 27 EU member states, and to what extent. Third, the relationship between migration and unemployment was also analyzed, as we believe that black employment strongly impairs the accuracy of official migration statistics. 3.1 Methodology The observations were based on data from 2005-2009, and as mentioned above, the number of independent variables was determined to be more than ten in the first step, and Backward elimination model was used to narrow it down to the five presented below. Member countries, during the first phase of the research, were not separated as the 15 old EU member states and the 12 new ones that joined later, after 2004 (EU12), and also, immigration and emigration was not assumed to compensate each other in the long term. We were looking for the best fitting model, using a software called Rapid Miner: macro indicators such as indebtedness and annual inflation dropped out immediately from the first models. We assume that the difference between immigration and emigration, as the dependent variable was affected not only by the annual value of the independent variables (GDP/capita, unemployment, minimum wages, child care allowances for two children), hence a fifth variable was introduced: (Yj-1,n), that is, the difference between immigration and emigration figures during the previous year (net migration). The dependent variable shows a difference a priori, that is, by use of net migration we wanted to clean the model up from possible trend effects, on the other hand, this variable indirectly reflects such non-quantifiable relationships, that would be difficult to filter out by statistical methods and purify from other effects. The reception capacity is strongly influenced by such non-quantifiable factors as the number of available jobs to foreigners and administrative restriction on the right of residence, for which there is no measure for the time being, or such nonquantitative factor, as the relationship between locals and foreigners. All this, however, will show in the number of immigrants indirectly. For example, immigrants try to avoid a country where they are confronted with serious problems regularly, but it has the same effect when too many people arrived a year earlier and the host country restricts immigration in some way, or the jobs available for immigrants have run out, while the unemployment rate may even have decreased. It is therefore inevitable that previous year’s net migration be included in the model as an independent variable.As another assumption, we intended to study the impact of unemployment on Yijn
260
[email protected]
Vol 20, No. 12;Dec 2013
variable, since an increase in unemployment would cause a decrease in net migration provided every worker is officially reported. However, when the rate of unreported employment is high, this relation cannot be detected significantly, that is, unemployment will have no significant impact on the dependent variable. The predictive ability of the model was checked by, beyond the traditional multivariate regression model analysis, taking the data from 2005-2009 as the basis for the regression model, then 2010’s data series were forecasted based on that, which was then finally compared to the available actual data series. That is, we made an estimation of the migration inside the EU using the model based on the estimated and actual figures of the year 2010. Table 1 Descriptive Statistics Independent variables GDP_per_capita_2006_ GDP_per_capita_2007_ GDP_per_capita_2008_ GDP_per_capita_2009_ GDP_per_capita_2010_ Unemployment_2006_ Unemployment_2007_ Unemployment_2008_ Unemployment_2009_ Unemployment_2010_ Minimum_wages_2006_ Minimum_wages_2007_ Minimum_wages_2008_ Minimum_wages_2009_ Minimum_wages_2010_ Childbfor2_2006 Childbfor2_2007 Childbfor2_2008 Childbfor2_2009 Childbfor2_2010 Im-em_2006_ Im-em_2007_ Im-em_2008_ Im-em_2009_ Im-em_2010_
Expected value 27942.6667 32646.7407 35328.1111 31570.0370 31320.4444 7.5400 6.5519 5.9481 7.7800 9.8852 725.0788 751.8423 773.9654 782.6077 794.5462 151.0388 151.9020 146.0352 147.3941 152.3681 55.0852 70.8285 48.7444 35.2326 32.8652
Std error 3674.2804 4212.0037 4415.2272 3993.9775 4007.8799 0.5730 0.3785 0.3517 0.7157 0.8260 92.8889 94.1446 92.2002 92.1384 93.1067 20.4985 20.7336 22.2797 22.6335 22.6782 26.9014 31.8616 24.2406 16.0120
Std deviation 19092.1212 21886.2131 22942.1934 20753.3159 20825.5548 2.9772 1.9669 1.8274 3.7191 4.2921 482.6647 489.1897 479.0864 478.7650 483.7964 106.5132 107.7350 115.7685 117.6071 117.8391 139.7835 165.5580 125.9580 83.2008
17.7730
92.3512
Min
Max
N
4313 5498 6798 6403 6333 4 3.4 2.8 1.06 4.1 82 92 112 123 122 18.6 18.6 25 25 22 -36.1 -20.5 -82.8 27.9 77.9
90032 106902 118219 104354 105195 16.8 11.3 9.9 17.5 22.3 1503 1570 1570 1641 1682 440.72 440.72 440.72 440.72 440.72 698.55 731.2 459.54 362.35
27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27
380.08
27
Source: Authors’ calculation The descriptive statistics showed a slight right skewness and a moderate kurtosis for each variable. Overall, in the case of the Yijn variable, the majority of the data from the 27 member states is below average, i.e. the difference between immigration and emigration stays below the average, the kurtosis does not differ too greatly from the normal distribution. Graphical representation of the variable showed a normal distribution and, according to descriptive statistics, the standard deviation is approximately constant. Modell 1, developed after the first experiments, shows a linear relationship and contains data from the 27 countries from 2005 to 2009, by five regressor variables. The high multi-collinearity of the model was predicted by the high R2 data, as well. A 261
[email protected]
Vol 20, No. 12;Dec 2013
decision could not be made based on the Durbin–Watson test’s value of 1.06, but the partial correlation coefficients clearly showed that the variable values measured in each year (like values of unemployment from 2006 to 2009) did strongly correlate with each other and the minimum wage and GDP/capita variables could also not be considered to be fully independent. Despite its detected errors, on the basis of the first full model’s regression lines we forecasted the 2010 expected values, which showed almost complete match with the actual data.
Figure 1. The correctness of forecast for 2010 – comparison between the factual and predicted data by the regression function (normalized data)
Multi-collinearity was eliminated from the model in two steps: first we filtered out outliers by determining the Mahalanobis distance values with probabilities less than 0.05: according to the expectations, Luxembourg, primarily due to the GDP data and Malta, on the other hand, due to its size can be considered as outsiders. Based on the indicators of net migration Spain is considered outstanding as well, but the explanation for this lies elsewhere, it primarily sticks out due to the still high influx of migrants from former colonies, but the omission of Spain from further analysis would have caused a significant loss of information. As the next step, first, the 27 member states were divided into three clusters according to their GDP per capita. Here again, Luxembourg was in its own cluster based on GDP and the minimum wage, so we completely omitted it from the model, together with Malta. After that we reclassified Greece and Portugal into the group of lastly joined countries, so eventually the old member countries’ group had 11 members while the group of the least developed countries, most of which joined after 2004, had 14. In the second run the model that now had 25 elements (the 25 member states, excluding outliers) was cleared of multi-collinearity by principal component analysis, it was narrowed down to four components using the Warimax method. Table 2: Factor loadings principal components Factor 1*
Factor 2*
Factor 3*
Factor 4*
GDP/capita 2006 GDP/capita 2007 GDP/capita 2008 GDP/capita 2009 262
[email protected]
Vol 20, No. 12;Dec 2013
Unemployment 2006 Unemployment 2007 Unemployment 2008 Unemployment 2009 Minimum wages 2006 Minimum wages 2007 Minimum wages 2008 Minimum wages 2009 Children benefit 2006 Children benefit 2007 Children benefit 2008
-0-9211 -0.9449 -0.8365 0.9066 0.9095 0.924 0.9156 0.9006 0.8896 0.9030 0.9261
Im-em 2006 Im-em 2007 Im-em 2008 Im-em 2009 6.8838 Expl Var 0.22868 Prop. Total *Varimax normalized, not rotated **Marked loadings are > 0.7000
-0.8460 -0.9134 -0.9701 3.3876 0.1411
4.8108 0.2004
Table3 Regression results
Children benefit 2009
5.6830 0.2367
The model's reliability was slightly reduced, but the principal component analysis did not provide more information either about the individual components. That is why we tried to compile a model, where the regression equation was constructed by less data than in the first one. Using a stepwise regression method, we created a regression line for 27 countries, in which we sought those years where all five variables most significantly affect net migration measured in the year to be forecasted. It was the data of year 2006 from which Model 2 most exactly predicted the net migration data of year 2010, while in Model 3, used now for 25 countries (excluding outliers), it was the data series of year 2007 that showed correlation with that of year 2010.That is, the developments of net migration can be predicted with relative accuracy based on the historical data from three years before. This relationship was investigated further as follows. After leaving out the two countries that were regarded as outliers, the correlation was checked for 25 member countries in the year 2011 (Model 4), and then the run was repeated in two clusters. (The EU 11 group bringing together the more developed ones and the EU 14 group of those lagging behind). The delayed impact of the selected five variables was clearly confirmed in the whole group (EU25) and the group of the more developed countries by the data summarized in the table. However, in the group of the poorer countries it is clearly one factor, unemployment, that determined the difference between immigration and emigration. Moreover, in the group of the 263
[email protected]
Vol 20, No. 12;Dec 2013
more developed countries, it was precisely the annual unemployment data that had the weakest influence as regressor, it could have even been omitted from a statistical point of view. After this peculiarity has been found in each and every case, we needed to conclude that if there is no correlation between the data measuring legal employment and the fluctuation in immigration, then labor migration must be characterized by a significant level of illegal employment. In the group of developed countries, however, GDP/capita, indicating the richness of the country, was clearly the most important independent variable, whereas the social services of different countries, and within these the studied child benefit differences, fall in a much wider range. One possible explanation is that in spite of the same rights, the Eastern European, mostly illegally employed workers do not get the family allowance they deserve exactly because of their illegal occupation. As a direct consequence, despite our prior assumption, this kind of "income" does not strongly motivate the direction of migration. Table 4 Summary statistics Modell summary R
R
2
Adj R2
F
p
1
0,98
0,96
0,85
9,09
0,003
Std Error of est. 35,11
2
0,64
0,42
0,28
3,05
0,03
78,159
3
0,74
0,56
0,45
5,39
0,002
67,98
4
0,74
0,56
0,45
5,32
0,002
68,08
5
0,78
0,62
0,38
2,61
0,1
91,16
6
0,86
0,75
0,59
4,81
0,025
73,939
7
0,87
0,76
0,61
5,2
0,02
71,74
8
0,51
0,26
0,18
3,22
0,1
21,5
9
0,87
0,76
0,6
4,85
0,04
12,178
As it is shown, with the exception of the eighth model, there is moderate correlation between the regressors and the dependent variables, while the adjusted R2s are significantly different from the R2 indicators, which indicates the model to be applicable only in this closed range, that is, it can be considered robust only in this range. Therefore, we examined the likelihood of quadratic error and found the strength of the model to be 0.83, which is acceptable. Finally, we forecast the expected net migration data by country, from 2010 to 2012. Table 5 Net migration forcast 2010-2012 in thousand by countries Member State Austria
Net migration_2010_ 21,9
Belgium
Net migration_2012
33,3
-18,7
-2,7
72,6
Denmark
10,8
11,8
-75,3
Finland
13,7
9,8
63,6
72
70,5
57,8
151,6
150,4
80,5
Greece
-0,9
2,3
-34,8
Ireland
-34,2
-34,6
81,3
France Germany
264
Net migration_2011_
[email protected]
Vol 20, No. 12;Dec 2013
3,4
-117
3,8
-11,5
-11,4
62,2
63
72,6
50
46,3
140,2
251,7
255,8
-200,4
2
-1077,4
-5,5
-113,8
-12,8
-16,2
-892,1
-2,5
-11,8
-103,2
Hungary
-11,1
-9,8
227,7
Latvia
-10,7
-17,5
44,8
Lithuania
-77,9
-59,4
-111
4,7
137,4
-0,8
-1111,6
15,9
164
Netherlands Portugal Spain Sweden United Kingdom Bulgaria Cyprus Czech Republic Estonia
Poland Romania Slovakia Slovenia SUM
9,3 0,5
0,5
131,9
497,4
499,9
-2593,3
The data show an increasing migration, clearly assuming current trends to continue, a strong difference between the two halves of Europe being the primary cause on the one hand, and the reclassification of some countries such as Portugal and Greece amid the less developed, on the other. However, in this second group, the driving force behind emigration is unemployment, from a statistical point of view. Also conspicuous is the population loss throughout the EU, but the assessment of this requires a more detailed approach. On the one hand, Europe is obviously experiencing a decrease in population, but this may also result partly from the reverse flow of earlier immigrants from third countries. On the other hand, statistical data base used was insufficient for two countries, both essential in Eastern European migration, Romania and Bulgaria. They did not provide data in some years (2010) and at other times the data they provided has not been deemed to be reliable. Our forecast indicates a high emigration, which coincides with practical observations. However, if we filter out these two countries’ emigration, European net migration shows a decreasing trend. 4.
Conclusion
Regression analysis showed that the flow of medium-term emigration from Eastern Europe and simultaneously into Western Europe will accelerate in the coming years. Clearly outlined in the donor and recipient country groups, and from among the recently acceded EU-10 countries Romania, and Bulgaria, depletion is accelerating. Three out of ten countries from this group will partly lose their population, and immigration growth in Hungary, Slovenia, and Poland can hardly compensate. The other group forms from the EU-15, where multiple regression equation forecasts increases in emigration from Ireland, Netherland and Greece (countries which are losing their former appeal), while Germany and Italy, have also confirmed a unilateral position as hosts. Italy, on the other hand, seems to attract fewer immigrants, as does Great Britain. Special research might reveal the dual impact of financial crises and administrative restriction against illegal immigration
265
[email protected]
Vol 20, No. 12;Dec 2013
in these countries. Nevertheless, the model predicting the difference between immigration and emigration for three years has a significant limitation. Firstly, the database does not already contain full data of regressors for 2004-2010; specifically, Greece, Bulgaria, Romania, failed to provide their special immigration and emigration data for international databases. As already mentioned, the lack of data was bridged by a simple linear regression analysis on country base, but the estimation has obviously worsened the forecast accuracy, which appeared in some of the countries in the larger regression error as well. The model has an inconsistency regarding immigration from third countries. The available database until 2010 shows a large decline in the number of immigrants to Spain, Italy, and Portugal, where the inflow of illegal workers from South America and Africa traditionally was huge till the outbreak of crisis in 2008. Their number will grow once again, according to our projection, but the model is not clear whether the seasonal employment stream from Eastern Europe, (mainly from Romania, which would suffer from a high number of emigrants in the same period), or South American immigrants could cause the difference. Otherwise, the cause could be the imperfection of the statistical model, and this clearly requires further studies. References Barroso J.M.State of the Union 2012, www.ec.europa.eu/soteu2012/index_en.htm Liu, Ben-Chieh, 1975: Differential Net Migration and the Quality of Life, Review of Economics and Statistics, 1975. Volume Ben-Gad M. (2004). The economic effects of immigration – a dynamic analysis. Journal of Economic Dynamics and Control, 28, (3)1825-1845 Berry, Michael J. A, Gordon Linoff: Data mining techniques: for marketing, sales, and customer — 2nd ed. CE 2003: Recent demographic trends 2002. Strassburg: Council of Europe DeMaris, Alfred, 1946: Regression with social data : modeling continuous and limited response variables (Wiley series in probability and statistics) Düvell, Franck, 2012. Population, Space and Place, 18(4).415-427 Dumont, Jean – Christophe, Spielvogel, Gilles, Widmayer, Sarah 2010: International Migrants in Developed, Emerging and Developing Countries: An extended profile, OECD, Social, Employment and Migration Working Papers, No.114, www.oecd.org/els/workingpapers Ekberg J, at al, 2011. Will Future Immigration to Sweden make it easier to finance to welfare system? Springer Science + Business Media B.V. on line Glenn J. Myatt, Making Sense of Data, 2007:A Practical Guide to Exploratory Data Analysis and Data Mining, John Wiley & Sons, Inc., Hoboken, New Jersey ILO, ILOSTAT Database, (Laborsta), 2012. Social Security Inquiry 2012 ILO, ILOSTAT, Database (Laborsta),2012, Taxanomy Jennisen, Roel,2007: Causality Chains in the International Migration System Approach, Springer Science+Business Media B.V. (26),411-436 Newland, Kathlen, editor, 2010 Diasporas: new Partners in Global Development Policy, Washington, DC: Migration Policy Institute KovácsP-Petres T-Tóth L, 2008: A new measure of multicollinearity in linear regression models, International Statistical Review (ISR) 73.(3), 405-412. Lee, R.D. Miller, T. (2000). Immigration, social security and broader fiscal impacts. American Economic Review, 90. (2)350-354. Massey, Arango, Hugo, et al. 2011:Theories of International Migration: A review and Appraisal ,
266
[email protected]
Vol 20, No. 12;Dec 2013
Population and Development Review, (19) 3, 431-466, Mierswa, Ingo and Wurst, Michael and Klinkenberg, Ralf and Scholz, Martin and Euler, Timm , 2006: YALE: Rapid Prototyping for Complex Data Mining Tasks, in Proceedings of the 12th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-06), 2006.MIRPAL ( Migration and Remittances Peer Assisted Learning), 2011, Harnessing the Diaspora fpr Development in Europe and Central Asia, World Bank, Washington OECD Migration Databases ( http://www.oecd.org/document) 2012 Peixoto, Joao.2012 :Back to the South: Social and Political Aspects of Latin American Migration to Southern Europe, International Migration 50(3)58-82 Piore, MJ.1986:The Shifting Grounds for Immigration, Annals of the American Academy of Political and Social Science, 48 (5) :23-33. Ratha, Shaw, 2007: South –South Migration and Remittances, World Bank Working paper, No 102. Reiner,Martin, Radu, Dragos.2012 : Return Migration: The Experinece of Eastern Europe International Migration 50 (6)109-128. Stark,1991: The probabilitiy of return migration, migrant’s work effort, and migrants perfomance?1991. Journal of Development Economics, 35(2) 399-405. Taylor, Patrick, 2007: The drivers of immigration in contemporary society: Unequal distribution of resources and opportunities, Human Ecology: 35(6): 775-776. Todaro, Maruszko, 1987: Illegal Migration and United States Immigration reform – A Conceptual Framework Population and development Review,13, (1) 101-114. Ulrich, R. (1994): The impact of foreigners on public purse. In S. Spencer (Ed), Immigration as an economic asset. Staffordshire: Trentham Books United Nation, 2012.: World Economic Situation and Prospects, Global Economic Outlook(GEO) Database,www.un.org/en/development/desa/policy/proj-link/global-economic-outlook.shtml. Witten, Ian. H., Eibe, Frank, 2009: Data mining : practical machine learning tools and techniques, 2nd ed. , Morgan Kaufmann series in data management systems. World Bank : Migration and Remittances Factbook, Washington 2011 World Bank Migration Letters 16. Washington 2011 World Bank: Migration and development Brief 17, Washington 2011. Notes Note 1. Jennissen’s approach is to create a framework for studies on migration incorporating causalities. In that dimension countries form special and separate migration systems, not necessarily close to each other geographically. Migrants are influenced different contexts, social, political, ethical and of course economic one. We focus only on the last one. Note 2. Diaspora definitions varies, there is no single definition, but we use the following one: who lives and works in another country, where he or she was born. Note 3. Ratha and Shaw developed for World Bank studies a bilateral migration matrix on country base, and it was updated with immigrant stock data from various sources to estimate the stock worldwide in 2010. These data presented from the available information and may consist only estimates
267
[email protected]
