Measuring International Migration in Azerbaijan

: International migration signiﬁcantly affects economic, social, cultural, and political factors of the country. Owing to this situation, it can be said that the reasons of international migration should be analyzed in order to control this problem. The purpose of this study is to determine the inﬂuencing factors of international migration in Azerbaijan. In this scope, annual data of 11 explanatory variables for the period of 1995–2015 was analyzed via Multivariate Adaptive Regression Splines (MARS) method. According to the results of this analysis, it was identiﬁed that people prefer to move other countries in case of high unemployment rates. In addition, the results of the study show that population growth and high mortality rate increases the migration level. While considering these results, it was recommended that Azerbaijan should focus on these aspects to control international migration problem.


Introduction
Michael Fisher [1] in his book "Migration: A World History" argues that we all are descendants of migrants and the process of migration began from the emergence of the homo sapience and still is ongoing and with great possibility will continue in the later periods of the history. According to him, climate changes, wars, starvation, technological developments, and some other historical process contributed intensity and extension of the size and geography of migration [1]. From this perspective it is possible to argue that the causes of the migration are different and they can be changed as a result of expected and unexpected economic, social, political, scientific, and apparently psychological process.
In addition, it is also possible to argue that while some of the migrations were peaceful, other were violent. For instance, wars, in particular two world wars, which were followed with high level of violence, triggered the level of migrant flow. formal and informal business institution of Azerbaijani labor migrants. In addition, some of them investigated social impacts, such as identity issues on younger generation of migrants [14].
Rumyantsov [15], in his turn, studied immigration and reintegration process in Azerbaijan. He divides immigrants in Azerbaijan into three groups: (1) natives of Azerbaijan (re-emigrants) and their family members; (2) ethnics from Georgia; (3) labor immigrants from different countries, who arrive to Azerbaijan to look for a job or to open their businesses. According to him, immigrants mainly accumulated in the capital city, Baku, and established transitional network in which they have integrated [15]. Rumyantsov [15], claims that majority of the immigrants are mainly ethnic Azerbaijanis who resided in the Post-Soviet countries and only few, whose number is estimated around 20,000-30,000, are from different nationalities. While ethnic Azerbaijanis immigrated for permanent purposes, others for temporary and mainly due to economic reasons.
In fact, actually, all above-mentioned works were able to reveal direct economic impact of migration to economy of Azerbaijan by using qualitative methods, but they failed empirically generalize whole picture of the process.
Indeed, few scholars implemented quantitative method to measure causes and consequences of migration to Azerbaijani society. For instance, Dermendzhieva [16], by using household data from CRRC (Caucasus Research Resource Center) Barometer, measured economic consequences of migration to Azerbaijan comparing it with the other two South Caucasus countries. Although, this research focuses on measuring of consequences of migration to economic situation in the three South Caucasus countries, it misses to determine factors influencing to dynamics of net migration in those states, including Azerbaijan. This paper aims, by using recent statistical data on migration process in Azerbaijan, to shed light on this important issue for this country. This article investigates the role of economic and social factors on international migration process in Post-Soviet Azerbaijan. To investigate the above-mentioned relationship, we employed the Multivariate Adaptive Regression Splines (MARS) method to Azerbaijani data ranging from 1995 to 2015. The study contributes to the current literature from different aspects. First, it is a first study devoted to the above-mentioned relationship in case of Azerbaijan. Second, it employs the MARS method, which can be considered as a new and original method for the concept. The study also provides some policy recommendations to minimize the negative effects of the migration. Besides, it should be also noted that the paper miss to measure political factors triggering migration process, which is aimed to be investigated in our future works.
The rest of the paper structured as follow: Section 2 reviews the related literature and conceptualizes the theoretical framework of the study. Section 3 discusses the Post-Soviet situation in Azerbaijan regarding the international migration. The employed methodology and data described in Section 4. Empirical results are given in Section 5. Finally, Section 6 discusses the findings of the study and provides policy recommendations.

Theorizing Determinants of International Migration: Literature Review and Theoretical Framework
Considering interdisciplinary character of migration, recently numbers of mobility theories have been developed in social sciences. Simmel [17], for instance, developed an agenda that links mobility and materials. In his description, tempo of modern cities occupies central role in the mobilization of the people. According to him, in such high ration of impulses of cities people need to integrate their relations and activities into a highly complex organism, which in the end stimulate the level of their mobility ability [17]. Development of communication technologies triggered some of the scientists to analyze social and technical factors as a hybrid system. According to this view, distance among people, objects, and technologies became closer as a result of extended physical movement [18]. Theorists of 'relationality' and circulation [19][20][21] more focusing on relation between structure and agency concludes that movement of people, capital, and things in the end will form or reform of the moved place [18]. Other scholar in the field studied emotional effects of movement of peoples not only themselves but also to their environment [21,22].
Actually, most of the works on mobility were researched qualitatively by using face to face meetings, interviews and focus groups, which are able to reveal problems in micro level and explain them in-depth, but lacks to generalize determinants of migration in macro level. In this regard, quantitative methods are the best to find relations between various factors in measuring of migration. Indeed, many quantitative studies found economic factors as the leading determinant in the fostering of international migration [23][24][25][26][27][28][29][30][31][32]. Jennissen [33], Balderas and Greenwood [27], Pânzaru [29] and Campaniello [31], analyzed the reasons of international migration in the European Union and identified that economic factors cause people to move to other countries. Moreover, Acupan and Agbola [34] and Deluna [35], determined similar conclusion for the Philippines. Mayda [24], evaluated the factors that affect international migration in the OECD countries and defined that economic factors play a significant role in this problem.
Rise of unemployment rate in the country of origin push people to move to other countries and in the end increases emigration indicators. As Naudé [47], states better access to jobs in the destination country is additional factor attracting emigration of people. On the other hand, it can be argued that decrease of unemployment rate has opposite effects on emigration process. In this regard, to examine importance of unemployment rate in Azerbaijan on migration, we posit the following hypothesis: Hypothesis 1. In Azerbaijan people more emigrate in the periods when there is high unemployment rate.
Inflation rate is another important economic factor that may have direct effect on the level of migration. Indeed, high rate of inflation negatively effects on net incomes of population and in condition of low level of salaries people may seek new job opportunities in abroad. Indeed, and higher inflation demonstrate the fragility in the economy. Due to this condition, they should increase international migration from this country [35,39]. In this case, it can be argued that during a high rate of inflation people more emigrate.

Hypothesis 2.
The level of emigration is higher in Azerbaijan when the rate of inflation is higher.
Economic growth and increase of foreign direct investment have positive effects on economy of the states. Both of these factors provide stability in the economy as a whole. Therefore, it can be expected that there should be indirect relationship between these variables and international migration from Azerbaijan [24,34,44,47]. In addition to the economic factors, it was determined that social factors also lead to international migration according to the results of some other studies [34,40,44,47,48]. Yang and Guo [48], conducted a study in order to analyze the factors that influence people to migrate from China. As a result of the survey analysis, it was identified that social factors play a key role in this problem. Furthermore, Carlos [41], Pedersen et al. [44] and Naudé [47], emphasized the similar conclusion by using regression analysis. Moreover, Brücker and Siliverstovs [40], evaluated international migration from other countries to Germany in their study. They used GMM analysis in order to reach this objective and concluded that there is a relationship between social factors and international migration.
Regarding social factors, 3 different variables were taken into the considerations, which are number of crimes, infant mortality rate, and number of population size. Since all these variables represent negative aspects, there should be positive relationship between these variables and international migration from the country [23,49,50].

Hypothesis 4.
When number of crime, infant mortality, and population increase rate is higher the level of emigration in Azerbaijan is higher.
The level of health care is another important variable measuring the role of social factors in international migration. In this regard, the number of the beds in Azerbaijani hospitals is used as a determinant factor for health care. Because this variable refers to the quality in health sector, it is expected that it should decrease emigration from Azerbaijan [42].

Hypothesis 5.
When quality of education sector is higher the level of emigration is lower.
With respect to educational factors, 3 different variables were considered. When the number of schools and teachers is high, it was expected that people would not migrate from the country [25,33]. However, since the rate of illiteracy represents a negative situation in the country, there should be positive relationship between this variable and international migration from the country [45,48].

Hypothesis 6.
The level of emigration is lower when the level of education is higher.

Migration in Azerbaijan
International migration is taken into the consideration significantly in Azerbaijan. The main reason behind this situation is that especially in the last 20 years, Azerbaijan experienced essential issues that cause migration. First of all, at the beginning of 1990s, the Soviet Union collapsed due to the economic problems and political instability [51]. As a result of that process, like in most of the Post-Soviet states, migration became one of the few choices for significant part of the population to search for better social-economic condition in Azerbaijan. In this regard, it is estimated that millions of Azerbaijani migrants moved to the various former Soviet Republics, particularly to Russia [9]. Indeed, establishment of labor migrant network of Azerbaijanis in the biggest cities of Russia fostered international migration process in Azerbaijan [13]. In this period, another important reason that caused the increase of the migration in Azerbaijan is the war with Armenia over Nagorno-Karabakh, which continued until the signing of ceasefire agreement between the parties in 1994 [52,53]. In fact, the war triggered the move of hundred thousands of Azerbaijanis from Armenia and Armenians from Azerbaijan. Besides, as a result of the war, almost one million of Azerbaijani became IDPs in their country [54]. Actually, the war worsened economic situation in the country. At a result not only internal migration, from rural to industrial urban areas, intensified, but also caused the increase of emigration from the country.
As it can be seen from Figure 1, for the period of 1995-2007, in Azerbaijan, the number of total immigrants was lower than the number of emigrants. In other words, Azerbaijanis more emigrate in this period. According to the Table 1, in the early 1990s, Azerbaijanis migrated particularly to Russia, Ukraine, and Israel. Azerbaijanis emigrated to Russia and Ukraine as labor migrants and were able to establish their ethnic, small, and medium enterprises [13]. Indeed, almost a great number of emigrants from Azerbaijan to Israel were ethnic mountainous Jews who are an indigenous group in this country.
Change of migration balance in favor of emigration in 2008 was caused due to global economic crises, while the change in 2013-2014, was mainly triggered by the drop of oil prices in the energy markets. Actually, Azerbaijan's economy being significantly depended on oil prices faced with economic recession and followed with the increase of inflation and the decline of the value of national currency "manat" at least twice. All these factors inspired emigration from the country.
While comparing immigration process for the same period, however the difference with emigration is almost more than two times, one can find almost similar destination for the movement (Table 1). This tendency can be explained by the return of Azerbaijanis to home due to increase of ethnic tension in their left countries.
In the mid-2000s, the level of emigration slowed down and began to demonstrate opposite tendency. Particularly, after 2007 Azerbaijan became an immigrant-receiving country. As it can be observed in the Table 2, Kazakhstan and Russian Federation play important role with respect to the immigration to Azerbaijan. Another important point for this situation is that there is an increase in the number of the immigrants from Iran and Turkey over the years. Actually, the increase of oil revenues in Azerbaijan caused the increase of immigration to Azerbaijan from neighboring countries. Indeed, recently Azerbaijani government enhanced state investment for infrastructure projects, which triggered involvement of foreign companies, particularly from Turkey. In fact, Turkish business organizations occupy one of the leading positions in the foreign direct investment to Azerbaijan. Indeed, there are 1685 Turkish companies that registered and operate in Azerbaijan.   Source: [55].    7057  6804  2202  617  476  336  462  647  248  146  142  82  42  Turkmenistan 1336  996  557  348  232  165  202  268  187  137  175  151  68  Kazakhstan  3190  2682  1528  742  553  341  339  405  306  183  150  107  44  Ukraine  4355  3130  1899  965  528  367  385  473  390  300  291  177  86  Kyrgyzstan  1562  1094  319  133  63  38  27  61  29  35  35  22  9 Source: [55].

Data
This study is going to utilize the annual data for the periods of 1995-2015 in order to define the determinants of the international migration in Azerbaijan, which was taken from the State Statistical Committee of the Republic of Azerbaijan [56] a and World Bank [57].
For understand the influencing factors of international migration in Azerbaijan, the authors of the study used net migration number of this country as a dependent variable. In this context, the difference between the people who came to Azerbaijan to live and the people who left the country represents net migration number [27,33,36]. In other words, if this value is negative, it means that people emigrate from Azerbaijan to other countries. In addition to this situation, for understanding the main indicators of this problem, 11 variables in 4 different categories were taken into the consideration after analyzing similar studies in the literature. The details of these variables were explained in Table 3.
Before estimating the econometric relationships, the stationarity properties of the variables are investigated. To test the stationarity of the variables the well-known Augmented Dickey Fuller (ADF) [58] and Phillips-Peron (PP) [59], tests are utilized, which is not described here and the interested reader can refer to the related literature. Unit root tests for the variables are performed by EViews 8.0 program. The econometric relationships are conducted by using the MARS 2.0 program.

Methodology: MARS Method
Multivariate Adaptive Regression Splines (MARS) method was developed by Jerome Friedman in 1991. It is a type of the regression method, which analyzes the relationship between dependent and independent variables. This method has some advantages in comparison with some other estimation methods. First, it allows using number of explanatory variables because of its "multivariate nature". Moreover, the word "adaptive" refers to the situation that it can choose the best model while evaluating many different alternatives. In addition, instead of making a simple regression line "smoothing splines" are used in the analysis process. Therefore, it can be argued that the relationship between the variables can be analyzed more effectively [62]. The equation of MARS method was given below.
In this equation, the term "Y" refers to the dependent variable. On the other hand, "X" shows the independent variables that may have an effect on the dependent variable. Therefore, it can be understood that there are k explanatory variables in the analysis. Additionally, "a n B n X t " refers to the basis function, so "a n B n " gives information about the coefficient of the basis of functions. In other words, it can be also seen that there will be k different basis functions with the combinations of t different explanatory variables. Furthermore, "ε" demonstrates error term of the equation.
MARS method has some advantages by comparison with other regression analysis. First, in this method, independent variables can take more than one coefficient in accordance to different conditions. In this regard, it can be possible to reach more successful results. In addition, in MARS method, there is no Multicollinearity problem, which refers to the high value of the relationship among explanatory variables. Thus, relatively higher number of independent variables can be included in the analysis process.
On the other side, with respect to the model developing process, MARS method has two different stages. First, all potential basis functions are developed by the system while considering the meaningful combinations of independent variables. This process continues until the reaching of the most complex model. The size of the most complex model depends on the data embedded to the MARS system by the users. Within this scope, on the system, the space of "the maximum amount of basic functions" should be completed with numbers that are between 0 and 250 [63].
The second stage of the modeling process in MARS method is reaching the best model. Within this context, some unnecessary basis functions are eliminated from the most complex method by the system. For this purpose, two different criteria play an important role, which are GCV (generalized cross validation) and R 2 values. Since GCV refers to error value and R 2 shows how well a model is explained, the bases of functions, which will be eliminated by the system, have high GCV and low R 2 values. In other words, the best model has the highest R 2 and lowest GCV values [62].
Since MARS is a relatively new method, it can be seen that there are few studies in the literature in which this method was used. Sephton [64], determined that MARS method is more successful than probit method [65] to predict the recession. Similar to this study, Bolder and Rubin [66], concluded that MARS method gives better results to define the best debt strategy in comparison with Kernel regression analysis. Moreover, Muzır [67], made a study about credit risk of Turkish banks and defined that MARS method measures this risk better than the logistic regression [68] analysis. Migues et al. [69], also reached the same conclusion with respect to the evaluation of customer attrition. Furthermore, Yüksel and Zengin [70] identified that MARS method is better than logit method [68] so as to determine the early warning signals of global mortgage crisis.
In addition to those studies, Tunay [63] and Oktar and Yüksel [71], used MARS method in order to evaluate the significant factors of the banking crisis in Turkey. Moreover, Sephton [64] and Tunay [72], made a study to analyze the recession by using this method. Yüksel [73], analyzed the determinants of non-performing loans and Tunay [74], evaluated the stability in Turkey with respect to the velocity of money circulation. Altun and Türkan [75], determined that employment rate is the most important indicator of the life satisfaction in OECD countries, employing MARS method.
The next section provides the empirical results based on the employing MARS method to the Azerbaijani data.

Empirical Results
In the first step of the analysis, a control was made to understand whether independent variables are stationary or not. In this context, Augmented Dickey Fuller (ADF) [58] and Phillips-Peron (PP) [59] unit root tests were performed to reach this objective. The details of these tests were given in Table 4. The probability values of these tests give information about the stationary properties of the variables. Considering the null hypothesis, which states the non-stationarity of the variable, if the p-values are less than 0.05, it can be concluded that the variable is stationary. As it can be seen from Table 4, the probability values of only 2 independent variables (inflation rate, infant mortality rate) are less than 0.05. Therefore, it was identified that these 2 variables are stationary whereas other 9 variables are not. (Note that, [76] also found GDP variable to be I(1) as a result of PP test). Owing to this condition, the first difference of these 9 variables and level forms of the stationary (2) variables are used in the analysis.
After performing unit root tests, an analysis was made by using MARS method to understand the relationship between dependent and independent variables. MARS system created 6 different models.
The details of them were demonstrated in Table 5. The model, which is at the bottom of Table 5, is called as starting model that has only one basis function and variable. After that, MARS system produced all different models according to the criteria embedded on the system. Therefore, the model, which is at the top of Table 5, is named as the most complex model. It has 6 basis functions and 4 different variables. Next, MARS system eliminated some basis functions from the most complex model. In this process, the functions, which decreased the explanatory power of the model, were excluded by the system. In other words, basis functions, which increase GCV (error value) and decrease R 2 values, were eliminated. As a result of this process, the best model can be obtained. As it can be seen from Table 6, the best model has 3 basis functions and 3 different variables. Additionally, it also has the lowest GCV and highest R 2 values. The details of the best model were explained in Table 6. As it can be seen from Table 6, p-values of all basis functions are less than 0.01. This means that the coefficients of these variables are statistically significant. On the other hand, the p-value of F test demonstrates that all coefficients of the model are jointly significant. In addition, adjusted R 2 value (0.976) is very high. This value means that independent variables can explain 97.6% of the change in the dependent variable. The basis functions in the model were detailed in Table 7. As it can be seen from Table 7, 3 different independent variables are significant in order to explain international migration in Azerbaijan. The variable of "population" is stated in both basis functions 1 and 2. On the other side, basis function is linked to basis function 4. Because both of the coefficients of basis function 2 and 4 are negative, it was defined that there is an inverse relationship between population growth and international migration. In other words, it was identified that when population increased, people preferred to move from Azerbaijan to other countries.
Another important explanatory variable that affects international migration in Azerbaijan is unemployment rate. It is stated in basis function 3 and the coefficient of this basis function (−0.588) is negative. This situation shows that there is a negative relationship between unemployment rate and international migration. That is to say, when unemployment rate is high in Azerbaijan, people opt for moving to other countries. While considering this result, it can also be said that these people move to other countries in order to have a job.
In addition to these variables, the variable of "infant mortality rate" also influences international migration in Azerbaijan. It takes place in basis function 4 and the coefficient of this basis function is −0.003. Therefore, it was determined that there is a negative relationship between infant mortality rate and international migration. In other words, it was identified that people do not prefer to live in Azerbaijan when there is a high infant mortality rates.

Discussion and Conclusions
The purpose of this study is to identify the determinants of international migration in Azerbaijan. In this context, annual data of 11 explanatory variables for the period of 1995-2015 was considered in this study. The Multivariate Adaptive Regression Spline (MARS) method was used in order to reach this objective.
First of all, we employed unit root tests to understand whether the variables are stationary or not. For this purpose, Augmented Dickey Fuller (ADF) [58] and Phillips Peron (PP) [59], unit root tests were performed. As a result of this analysis, it was defined that the probability values of only 2 independent variables (inflation rate, infant mortality rate) are less than 0.05. This means that these 2 variables are stationary while other 9 variables are not. Because of this issue, the first difference of these 9 variables and level values of other 2 variables are used in the analysis.
After unit root tests, we investigated the relationship between dependent and independent variables by using MARS method. MARS system produced 6 different models. Moreover, it was also identified that the best model has 4 basis functions and 3 different independent variables, which are statistically significant at 99% significance level. On the other side, adjusted R 2 value showed that independent variables can explain 97.6% of the change in the dependent variable.
According to the results of the analysis, it was determined that 3 different independent variables are significant in order to explain international migration in Azerbaijan. First of all, with respect to the economic factors, it was identified that there is a negative relationship between unemployment rate and international migration. This shows that when unemployment rate is high in Azerbaijan, people opt for moving to other countries to find a job. In other words, it can be said that better job opportunity is a significant factor for Azerbaijani people to move to another country. This result is very parallel to the thought in the literature that people more emigrate when there is unemployment problem in a city or a country. For instance, Ahad [39], Farooq et al. [42] and Sulaimanova and Bostan [46], also reached the similar conclusions in their studies.
On the other side, regarding social factors, it was defined that there is a negative relationship between population growth and international migration. This situation means that when population increases, people prefer to move from Azerbaijan to other countries. The main reason behind this situation is that in case of population increase, people start to face some difficulties to find a job. Owing to this condition, these people have some financial loss. It can be understood that this conclusion is very appropriate the general thought in the literature. For example, this issue was also underlined in many different studies [49,50].
Another important explanatory variable that influences international migration in Azerbaijan is infant mortality rates. It was determined that people do not prefer to live in Azerbaijan when it has high infant mortality rates. The main reason is that high infant mortality rate is an important social indicator that shows the poor quality of life in a country or city. Because of this situation, it is thought in the literature that it influences people to move this country. For instance, Farooq et al. [42], also emphasized the significance of this aspect on emigration rates.
The outcomes of the research show that Azerbaijani government should focus to solve economic and social problems in order to overcome problems that may rise from international migration. In particular, after the drop of oil prices in the global energy markets the revenues of the country decreased dramatically. As a result, incomes of the population decreased relevantly and according to independent observers there is high risk of the increase of unemployment rate in the country if substantive measures are not taken. Besides, there is a high risk of brain drain from the country. In the interview with the local specialists on migration issue from one of the Baku based NGO, it was stated that however there is not accurate statistic, the number of high quality emigrants increased since 2015 and the trend has been continuing during 2017.
On the other hand, political problems in Middle East may trigger inflow of immigrants from this region to Azerbaijan. Therefore, by giving importance to those problems Azerbaijani government may minimize their negative effects. To sum up, based on the estimations results we can conclude that there is enough evidence to support hypotheses 1 and 4, set in the Theoretical Framework Section. The results for hypotheses 2, 3, 5 and 6 are not straightforward, since the related variables, i.e., inflation rate, GDP per capita, foreign direct investment, health and education level representative variables are found to be insignificant. The impact of these factors is an interesting area for future research, employing different econometric methodologies, time span and theoretical framework. While analyzing an important topic for Azerbaijan, this study aims to make a significant contribution to the literature. Nevertheless, a new study that covers many different emerging countries for this framework will also be beneficial.