Faculty of Business and Economics Amsterdam School of Economics
“Regional Migration in Chile: Analysis for the period 2002-2015”
Master thesis Amsterdam, 2017
Author: María Pía Daniela Iocco Barías University ID: 11374519 MSc in Economics, Public Policy track.
Thesis Supervisor: Prof. Dr. Erik Plug Academic Year: 2016-2017
2 Statement of Originality
This document is written by student María Pía Iocco who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3 Table ofContents
1. Introduction ... 6
2. Literature review ... 7
2.1. Migration and economic development ... 7
2.2. Determinants of migration ... 8
3. Overview of Migration in Chile ... 10
4. Data ... 11
4.1. Description ... 11
4.2. Data limitations ... 12
5. Methodology... 13
5.1. A Theoretical model of Migration ... 13
5.2. Determinants of migration ... 14
5.3. An exercise on correlation and causality ... 15
5.4. Exploiting exogenous economic shocks in household surveys ... 16
6. Results ... 18
6.1. Regional Estimation ... 18
6.2. Causality exercise ... 19
6.3. Shock to economic conditions ... 23
7. Discussion and conclusions ... 25
8. References ... 27 9. Appendix ... 30 9.1. Appendix 1 ... 30 9.2. Appendix 2 ... 31 9.3. Appendix 3 ... 32 9.4. Appendix 4 ... 33 9.5. Appendix 5 ... 35
4 Abstract
Have the determinants of regional migration remained the same in the last decade? Does migration influence economic conditions or vice versa? Does a specific shock to economic conditions lead to a change in migration? In this thesis, I investigate these three questions using data from the Chilean National Socioeconomic Survey for the years 2006, 2009, 2011, 2013 and 2015. In answering the first question, I analyse the migration patterns at a regional level in Chile for the last decades to understand what defines migration and if the effects have remained constant throughout the periods. For the second question, it is necessary to explore the relation between GDP and migration in order to understand in which direction the causality flows. For the third, I construct a probit model at the household level to exploit the effect of an exogenous shock to economic conditions on migration. Outcomes show that (a) people migrate now towards regions with worse economic performances, (b) changes in economic conditions drive migration especially lagged 4 or 5 years, and (c) the increase in the price of copper had a positive and significant impact on migration towards mining regions.
5 I thank the National Commission for Scientific and Technological Research of the Ministry of Education of Chile for the funding corresponding to the scholarship CONICYT Becas Chile 5306/2016.
6 1. Introduction
Historically, economists have given importance to the role of internal migration in the economic growth of a country or in its GDP convergence rate within the different units of a territory. The theory states that migrants tend to move to regions with higher salaries or more favorable economic conditions in order to improve themselves (Ravenstein, 1889). In this way labor accumulates in the regions of destination. In these high growing regions, the average capital is higher and the new influx of people implies that the speed of capital accumulation per person declines. Consequently, the per capita capital increases in the region of origin, bringing convergence along (see Sala-i-Martin,1996).
Because of this link with the convergence of economic growth among regions, migration is a relevant phenomenon to study. Williamson hypothesis (Williamson, 1965) links agglomeration and economic growth and points out that after a threshold, a dispersed economy boosts GDP. Migration is the channel to reduce the size of the agglomeration. Therefore, from a public policy perspective, governments can decide how to increase, or how to lead migration to regions or provinces where it is most needed.
The main aim of this thesis is to analyze the process of regional migration under three perspectives. The first approach is to estimate a model of regional migration and investigate if the patterns and factors influencing migration have remained the same or have changed during the last decade. The second step is to examine the relation between migration and economic growth to investigate which determines which because of the reverse causality between them. The third is to zoom in on a household scale and use an exogenous shock to economic conditions to determine whether migration was affected.
My contribution is based on an updated estimation using a new source of data, considering the lack of a census in the last fifteen years, that also covers an extended period of more than one decade. Additionally, the analysis of the causality between the migration and economic conditions contributes to the discussion about the direction of the flow. Furthermore, the zoom in on a household level allows analysis of a shock that affects only a geographic area and adds robustness to the study in relation to migration and development.
The main findings are that economic conditions show opposite results than found previously or are not significant anymore in explaining migration. In addition, economic growth
7 influences regional migration with a lag of about 3 to 5 years, and there is not enough evidence of migration in improving development. And at a household level, a positive economic shock increased the likelihood of migration towards these regions. These results suggest that migration in Chile is driven by economic conditions and not vice versa.
This thesis is organized as follows: section 2 presents a brief literature review. Section 3 provides a description of the migration phenomenon in Chile. In Section 4 a description of the data used is outlined. Subsequently, section 5 presents the theoretical and empirical methodology, followed by the results in section 6. Conclusions and discussion are shown in section 7.
2. Literature review
2.1. Migration and economic development
Migration is a process in which individuals look for a place that allows them to maximize their utility by way of their investment in human capital, taking direct and indirect costs into consideration. Ravenstein’s laws of migration (1989) introduced the idea that migrants move to improve their economics conditions.
At the same time, migration is seen as a tool to reach equilibrium by neoclassical economists. Migration allows smoothing of regional difference in incomes by the free movement of labor from low-wage to high-wage areas (Borts and Stein, 1964). Therefore, it is clear that the relationship between regional economic development and migration is a “chicken or egg” one (Richardson, 1978).
Greenwood (1981) utilizes data from the National Census of Population of the United States to estimate a simultaneous-equations model of urban growth and migration. He finds that higher rates of outmigration tended to significantly discourage employment growth. Conversely, higher rates of in-migration encouraged such growth, suggesting that migration affects local development. Zhang and Song (2003), using panel and cross-sectional data, conclude that in China the causal link flows from economic growth to migration and not vice versa. Xu and Li (2008) construct a recursive dynamic computable general equilibrium model to study the effect of inter-regional labor migration in China. The results show that migration has little
8 impact on the GDP per capita gap. However, the regional disparity in per capita consumption decreased with a higher number of inter-regional migrants.
2.2. Determinants of migration
Several studies have aimed to estimate the factors that determine this movement from one territory to another one inside a country. Two groups of variables are commonly used, the first group referring to the characteristics of the area or place of destination and origin of the migration, and the second one regarding the individual’s or family’s characteristics.
Among the regional variables, the main one is the economic performance of the regions. There are two interpretations about how to incorporate these factors (Molho, 1986). The first approach, the speculative, considers the situation when the labor market is not cleared and migration occurs in finding a job. This means that the number of vacancies and unemployment information are more relevant. On the other hand, the contracted migration approach considers when the migration happens after a person is being hired. This weighs in factors like the wage of the population at the destination region, expected income, and the evolution of employment.
A second characteristic is the distance between the origin and destination region. Greenwood (1995) presents a list of the costs associated with this distance, such as information, opportunity, search, and the direct costs of moving and finding a job in a new city. These increase with the remoteness of the regions.
The third one is population size. However, this variable presents a bias due to simultaneity since both population and migration are determined at the same time. To correct this bias, different methodologies have been used, specifically the Berkson Method (Fields, 1982; Gabriel Shack-Marquez and Wascher, 1993). This method consists of normalizing the probability of migrating from one region to another using the probability of staying in the region of origin. This way, population size will be independent of the migration variable. A positive effect has been found, indicating that people move towards more populated regions. The last regional factor is the amenities. Greenwood (1995) presents research that supports the approach that amenities do not have an impact because they were incorporated earlier into the decision before, while the view that regions with better infrastructure can lower wages to attract employers has received less support. However, there seems to be a consensus
9 that the differences in economic performances among regions are more relevant than amenities (Greenwood and Hunt, 1989).
Regarding the group of individuals’ or families’ characteristics, the first one is age. There is plenty of evidence that the likelihood to migrate reaches a peak between years 20 and 29, and steadily declines after (Rogers and Castro, 1981). The second variable is education. According to previous studies, the likelihood of migrating increases with the number of schooling years, mainly because educated workers have more access to information about job positions. Additionally, risk and uncertainty are lower for them since they are more prone to migrate under the contracted approach (Greenwood, 1975). The third is employment status. Unemployed individuals have a higher likelihood to migrate than employed individuals. Herzog, Schlottmann and Boehn (1993) present a summary of studies in which they indicate that the probability of migrating increases while being unemployed. Finally, the fourth variable is civil status. Previous research supports that married workers, and especially those married with children, are less prone to migrate due to an increase in the costs of moving (Greenwood, 1975 and 1985; Hughes and McCormick, 1989).
Discrete-continuous models for migration at a individual or household level are often estimated trough multinomial, logit, probit, tobit, two-stage (Heckman) and maximum likelihood techniques. Studies of this type include Perloff, Lynch and Gabbard (1998), Taylor (1987, 1992), Stark and Taylor (1989, 1991), and Barham and Boucher (1998). Explicitly or implicitly, such empirical researches are based on a random-utility theoretic model that assumes households or individuals make migration decisions that maximize wellbeing.
In Chile, Aroca and Hewings (2002) utilize a migration model using traditional consumer theory and estimate at a regional level using data from the National Census 1992 and 2002. They conclude that there is a strong force in the regional labor market, which concentrates the workforce around the largest populated regions and that differences in the economic performance of the regions influence the migration decision. Soto and Torche (2004), utilize the same model, and test and conclude that the prohibition to sell houses bought with a government subsidy has had a significant impact on the migration level.
10 3. Overview of Migration in Chile
From 1976 until 2006, Chile was formed by thirteen regions numbered from north to south, except for the capital, which is located in the centre and corresponds to the number thirteen or RM. In Appendix 1 there is a map of Chile with the former regional division. In 2007, regions one and ten split to create regions fourteen and fifteen, as shown in the same appendix. For consistency, I only work with the original division of thirteen regions to analyse the result for a longer period of time.
Table 1 shows an estimated net migration rate for a five-year period, defined as (inflow − outflow)/population, of the thirteen regions for different periods.
In the first period, 2002 to 2006, only two regions had a negative rate and therefore had more emigrants than immigrants. During the period 2007-2011, regions five and eight (the second and third largest ones) also had a negative rate. In the last period, 2011 to 2015, regions two and thirteen (the capital) had negative rates, both lower than -4%.
Table 1: Net migration flows by region of origin
Migration period 2002-2006 2005-2009 2007-2011 2009-2013 2011-2015
Region i Net rate Net rate Net rate Net rate Net rate
1 0.07% -0.02% 3.86% 2.36% 1.64% 2 -0.92% -0.93% 0.67% -2.51% -6.89% 3 2.01% 0.95% 2.30% 2.37% 1.97% 4 0.12% 0.07% 0.24% 2.01% 2.87% 5 0.78% 0.28% -0.40% 0.78% 1.10% 6 2.14% 0.53% 1.10% 1.80% 2.38% 7 0.77% -0.12% 1.94% 0.68% 2.21% 8 0.51% -0.32% -0.98% 0.27% 0.72% 9 0.75% -0.76% 0.03% 0.83% 2.02% 10 1.00% 0.78% 0.82% 0.41% 1.08% 11 1.43% -0.39% 5.81% 4.19% 2.88% 12 9.11% 6.97% 3.93% 2.03% 1.43% 13 -3.86% -0.41% -7.57% -4.81% -4.51%
11 4. Data
4.1. Description
To analyse migration, I use data from Chile for the original division (thirteen regions). The data are derived from the National Socioeconomic Characterization Survey (Encuesta Nacional de Caracterización Socioeconómica, or CASEN), which has been collected since 1990 every two or three years. This survey aims to periodically measure the socioeconomic situation of the households and the population that resides in private housing, by measures such as family and household composition, education, health, housing, work situation, and income. This survey is the main source utilized to measure and analyze poverty and inequality in Chile.
For the migration estimation, the data show the tabulated answer to the question about where the respondent lived five years ago. This region is compared to the current place of residence to determine whether she or he is a migrant. This question was first introduced in the year 2006, so I use that year as the starting point.
Table 2 shows a summary of the surveys and their data.
Table 2: Survey data
Year of the
survey Migration Period Observations Valid Observations* Migration observations Migration rate
2006 2002-2006 268,873 248,371 6,728 2.71%
2009 2005-2009 246,925 229,283 4,694 2.05%
2011 2007-2011 294,271 270,847 10,522 3.88%
2013 2009-2013 218,491 200,292 6,640 3.32%
2015 2011-2015 266,968 246,762 9,157 3.71%
*: Observations with both current and previous region
We can observe that the inter-regional migration rate for a 5-year period is between 2% and 4%. The five surveys contain between 218.000 and 294.000 observations, all having more
12 than 200.000 observations with both current and former region of residence. The rest have missing values, mistyped answers or were living abroad.
The economic variables: regional Gross Domestic Product, GDP growth, salaries and unemployment, are determined using data from the Chilean Central Bank (Banco Central de Chile). The demographic variables: population, urban share and employees, are determined using data from the National Institute of Statistics (Instituto Nacional de Estadísticas). A geographic variable, distance in kilometres by road, is collected from the Ministry of Public Infrastructure, Secretary of Roadways (Ministerio de Obras Públicas, Dirección de Vialidad).
4.2. Data limitations
The analysis is conducted from 2002 to 2015. However, it is not possible to know when the respondent exactly moved to the current region nor if he/she has moved to a different city earlier in the period. Therefore, the model cannot identify the yearly effects. In addition, the data consist of overlapping years among the surveys and the frequency is not regular, as can be seen in Table 2. This hinders the construction of a clean extended period. Due to this, I treat every period and survey separately and not as a regional panel data.
Furthermore, I cannot distinguish whether the composition of the household is the same as that of five years ago, or if any member moved separately from the other individuals. Nevertheless, I assume that the household migrated as a unity and single movements are rare, The data do allow for estimates of inter-regional medium-term migration, defined as migration from one region to a different one in a period of five years. Thus, coefficients from different periods can be compared to contrast the variation across time using the household as the unity of analysis.
13 5. Methodology
5.1. A Theoretical model of Migration
In this section, I follow Aroca and Hewings’ (2002) methodology, and I utilize a simple model to analyze the decision of the households to migrate at a regional level. The model is grounded on classical consumer theory. It is assumed that a worker’s moving decision can be represented by the following utility maximization problem:
Max {𝑋𝑗,𝑇𝑗}
𝑈𝑗(𝑋𝑗, 𝑇𝑗,𝑍𝑗) (1)
Subject to a budget constraint:
𝐼𝑗≥ 𝑃𝑥𝑋𝑗+ 𝑃𝑇𝑗𝑇𝑗
Where 𝐼𝑗 is the income of the worker in the region j, 𝑋𝑗 represents a composite good different than transportation that the migrant demands in location j, 𝑇𝑗 is equal to one if transportation is needed when moving from the origin to region j and zero otherwise, 𝑍𝑗 is a set of other characteristics of region j that are taken into consideration by the worker, 𝐼𝑗 is the income of the worker at region j, and 𝑃𝑥 and 𝑃𝑇𝑗 are the prices of goods and transportation respectively.
If the region of destination is denoted as i and the region of origin as j, the indirect utility function of a worker that is evaluating a decision to move from region i to region j can be expressed as;
𝑉𝑗= 𝑉𝑗(𝑃𝑥, 𝑃𝑇𝑗, 𝐼𝑗, 𝑍𝑗) + 𝑒𝑗 (2)
Where 𝑒𝑗 is a stochastic error that can have numerous sources (see Ben-Akiva and Lerman, 1985). Because of the lack of reliable regional data, it has been assumed that the prices of the goods are the same everywhere, and therefore this variable will not affect the worker’s migration decision (Soto and Torche, 2004).
Therefore, the worker compares the utility that he/she can get from every possible destination region (including the option of staying in the current one), and so selects the region that derives the maximum utility. This process can be cast as a random utility
14 maximization selection subject to a stochastic error which, if assumed to have a generalized extreme value distribution, results in a logit specification, with the probability of a worker migrating from region i to region j as (Aroca and Hewings, 2002):
𝑃𝑖𝑗= 𝑒𝑉𝑖
∑𝐾 𝑒𝑉𝑘 𝑘=1
(3)
Where K is the number of alternative regions to which the worker can move, including the origin region (in this case, 13 options).
Some additional derivations will be necessary to derive an estimable equation. Following Berkson’s method (Berkson, 1944), imposing the constraint that ∑𝑘𝑗=1𝑃𝑖𝑗 = 1, and normalizing by the probability of staying in the region of origin (𝑃𝑖𝑖), the expression can be adapted to the following form:
ln (𝑃𝑖𝑗 𝑃𝑖𝑖
) = 𝑉𝑗− 𝑉𝑖 = 𝛼0+ 𝛼1𝑃𝑇𝑗+ 𝛼2𝐼𝑗− 𝛼3𝐼𝑖+ 𝛼4(𝑍𝑗− 𝑍𝑖)
(4)
Where the 𝛼′s are coefficients or vector of coefficients associated with the variables that determine the indirect utility function, and are assumed to be linear variables.
5.2. Determinants of migration
In answering if the determinants of migration have remained the same in the last decade, I follow Aroca and Hewings’ (2002) methodology, and using the Berkon’s Method (1944), 𝑀𝑖𝑔𝑡𝑖𝑗 is defined as ln𝑃𝑖𝑗
𝑃𝑖𝑖 using the survey taken in the year y, where 𝑃𝑖𝑗 is the number of people who moved from region i to region j and 𝑃𝑖𝑖 is the number of people of region i who stayed. However, because of data limitations, I use as a proxy of the number of households instead of persons. A proxy used for ln (𝑃𝑖𝑗
𝑃𝑖𝑖) is ln ( 𝑀𝑖𝑗
𝑀𝑖𝑖), where 𝑀𝑖𝑗 is the number of households (instead
of people) who moved from region i to region j. Likewise, 𝑀𝑖𝑖 is the number of households who were living in region i and stayed. In Appendix 3 there is a matrix showing the values of 𝑀𝑖𝑗 and 𝑀𝑖𝑖 for the period 2011-2015.
15 For the characteristics of the region, I use region i’s urban share population, and population from regions of origin and destination, following the criteria used in Aroca and Hewings (2002). The distance variable is used as a proxy for moving costs (Greenwood, 1995) and it enters the equation in a quadratic form because it is expected that costs increase but at a decreasing rate (Aroca and Hewings, 2002).
There is no publicly available and reliable data for the median regional income variable that extends for the whole 2002-2015 period. As Aroca and Hewings (2002) do, I use two proxies for it: the first one is the median productivity of the region as a proxy of the regional salary, defined as the regional GDP over the number of workers in that region. The second one is a proxy of the regional variation of the salaries as the GDP’s growth rate of the region.
5.3. An exercise on correlation and causality
The aim of this section is to determine the direction of causality. Given the data limitations, I derive my first analysis by constructing a ten-year period using two non-overlapping surveys: the 2006 and 2011 data. The first covers the years 2002 up to 2006, and the second uses the years 2007 up to 2011. Thus, I have a period t and a period t – 1. As a proxy for economic conditions, I use the annual GDP growth rate of both regions of origin and destination regions. In the model, I regress a migration vector for the period 2007-2011 on previous migration and previous GDP growth of the regions of destination. Additionally, I regress the GDP growth of the regions of destination for the period 2007-2011 on previous GDP growth and previous migration. Likewise, I repeat the exercise for regions of origin. The equations are as follows:
𝑀𝑖𝑔2011𝑖𝑗 = 𝛼0+ 𝛼1𝑀𝑖𝑔2006 𝑖𝑗 + 𝛼2𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2006 𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2011𝑖𝑗 + 𝑢𝑖𝑗 (5) 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2011𝑗 = 𝛼0+ 𝛼1𝑀𝑖𝑔2006 𝑖𝑗 + 𝛼2𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2006 𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2011𝑖𝑗 + 𝑢𝑖𝑗 (6) 𝑀𝑖𝑔2011𝑖𝑗 = 𝛼0+ 𝛼1𝑀𝑖𝑔2006 𝑖𝑗 + 𝛼 2𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2006𝑖 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2011 𝑖𝑗 + 𝑢 𝑖𝑗 (7) 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2011𝑖 = 𝛼0+ 𝛼1𝑀𝑖𝑔2006 𝑖𝑗 + 𝛼2𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2006𝑖 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2011 𝑖𝑗 + 𝑢𝑖𝑗 (8)
16 Hence, the dependent variable 𝑀𝑖𝑔2011𝑖𝑗 is a vector of 156 observations (13 ×12) containing the logarithm of the number of households who moved from region i to region j over the number of households who stayed for the period 2007-2011. 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2011𝑖 is the GDP growth of region of origin for the period 2010-2011. Similarly, 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2011𝑗 is the GDP growth of region of destination for the period 2010-2011.
5.4. Exploiting exogenous economic shocks in household surveys
Lastly, I evaluate whether a shock to economic conditions has led to a variation in the probability of a family of migrating in the last decade. I was not able to find a shock large and lasting enough to have an impact at a regional level in that period, thus I use one at a household level. This shock should also affect either more than one region or not all the industries in a region in order to distinguish it from a regional only effect. In addition, it must last long enough to have an influence in the medium-term migration.
During the years 2006 and 2008 (both inclusive) the price of the copper, measured as pounds per ton, soared from an annual average of 164 in 2005 up to 307 in 2006, 322 in 2007 and 312 in 2008. Graph 1 depicts the market closing price for the period 1997-2017. We can observe that the price of copper increased dramatically for the period 2006-2008, before dropping in 2009.
Graph 1: Copper price 0 50 100 150 200 250 300 350 400 450 500 01 -1 99 7 01 -1 99 8 01 -1 99 9 01 -2 00 0 01 -2 00 1 01 -2 00 2 01 -2 00 3 01 -2 00 4 01 -2 00 5 01 -2 00 6 01 -2 00 7 01 -2 00 8 01 -2 00 9 01 -2 01 0 01 -2 01 1 01 -2 01 2 01 -2 01 3 01 -2 01 4 01 -2 01 5 01 -2 01 6 01 -2 01 7 PO UN DS P ER T O N MONTH-YEAR
Copper's Closing Price
17 According to Donoso (2014), this increase in the price is due to an increase in the demand driven by the Chinese emerging economy, and not due to a reduction in the production capacity of Chile.
To test this, I estimate a probit model where the dependent variable is a binary variable that takes the value of one when the household was living in a different region five years ago and zero otherwise (as in Taylor, 1987). I include a set of controls variables that can affect the migration decision at an individual and household level. This set is based on the document “Notas de Población” (Notes about Population), published by the Economic Commission for Latin America and the Caribbean (CEPAL) of the United Nations, and by the work of Taylor (1987) and Barham and Boucher (1995).
Regions one, two and three are considered as mining regions since their share of mining GPD over regional GDP surpasses, on average, the 40% for the last decade. No other region is dependent on the extraction of copper. Since the industry in which the respondent is currently working in appears in the survey, I can estimate whether this increase in the price of copper -as a proxy for a change in economic conditions- led to more workers to migrate to work in the mining industry in a mining region.
The migrations of these years are being captured by the 2009 CASEN survey. Therefore, I construct a dummy, 𝑐𝑜𝑝𝑝𝑒𝑟ℎ, that takes value one if the household is located in a mining region and the respondent works in the mining industry in the 2009 survey, and zero otherwise.
The equation can be written as follows:
𝑀𝑖𝑔 = 𝛼0+ 𝛼1𝑐𝑜𝑝𝑝𝑒𝑟ℎ+ 𝛼2𝑎𝑔𝑒ℎ+ 𝛼3𝑠𝑐ℎ𝑜𝑜𝑙ℎ+ 𝛼4𝑐𝑖𝑣𝑖𝑙𝑠𝑡𝑎𝑡𝑢𝑠ℎ+ 𝛼5𝑛𝑢𝑚𝑝𝑒𝑟ℎ+ 𝛼5𝑐𝑢𝑟𝑟𝑒𝑛𝑡𝑖𝑛𝑐𝑜𝑚𝑒ℎ+ 𝛼6𝑔𝑒𝑛𝑑𝑒𝑟ℎ+ 𝛼7𝑛𝑎𝑡𝑖𝑛𝑑𝑖𝑔ℎ+ 𝛼8ℎ𝑜𝑢𝑠𝑒𝑠𝑢𝑏ℎ+ 𝛼9ℎ𝑒𝑎𝑙𝑡ℎℎ+ 𝛼10𝑟𝑢𝑟𝑎𝑙ℎ+ 𝑢ℎ
(9)
Where 𝑎𝑔𝑒ℎ is the age of the head of the household h, 𝑠𝑐ℎ𝑜𝑜𝑙ℎ are the years of schooling of the head of the household h, 𝑐𝑖𝑣𝑖𝑙𝑠𝑡𝑎𝑡𝑢𝑠ℎ is the civil status of the head of the household where one is for married and zero otherwise, 𝑛𝑢𝑚𝑝𝑒𝑟ℎ is the number of people living in the same house, and 𝑐𝑢𝑟𝑟𝑒𝑛𝑡𝑖𝑛𝑐𝑜𝑚𝑒ℎ is the current income of the household.
Additionally, I include 𝑔𝑒𝑛𝑑𝑒𝑟ℎ which corresponds to the gender of the head of the household h. It takes the value of one if is male and zero if is female. 𝑛𝑎𝑡𝑖𝑛𝑑𝑖𝑔ℎ takes the value of 1 if the
18 family recognizes itself as belonging to a native indigenous population, ℎ𝑒𝑎𝑙𝑡ℎℎ is a self-reported health indicator where one is poor health and seven is the highest value, and ℎ𝑜𝑢𝑠𝑒𝑠𝑢𝑏ℎ is a dummy that takes the value one only if the property was bought with a governmental subsidy. This last one follows Soto and Torche (2004), who highlight that the restrictions of the government to rent or sell properties financed with any state subvention lead to a decrease in the likelihood of migrating.
6. Results
6.1. Regional Estimation
As described earlier, I estimate a model at a regional level in order to determine the impact of the economic conditions on the overall migration rates. Table 3 presents the point estimates of the model for the different periods. Since Soto and Torche (2004) find that 𝑆𝑖𝑗 is not significant, I present the results with and without it.
The coefficients for 𝑊𝑗, the median productivity of regions of destination, are positive and significant for the surveys 2006, 2011 and 2013. However, coefficients for 𝑊𝑗 are not distinguishable from zero for the last period, meaning that is not clear anymore that workers are moving to regions with higher economic performance. 𝑊𝑖, which is the median productivity of regions i, had a positive but not significant coefficient for the first four periods, but significant at the 1% level of confidence only for the last period. This result is opposite from what it was initially expected, since it indicates that workers are moving out from regions with higher mean productivity.
The variable 𝐷𝑊𝑗, the GDP growth of region j, shows a shifting pattern. It was positive for the first period, then negative and not significant for the second. It became positive and significant for the third term and negative and not distinguishable from zero thereafter. 𝐷𝑊𝑖, just like 𝑊𝑖, was positive and not significant at the beginning but it became significant, indicating that migration comes from regions with higher growth.
During the last period, people moved from regions with GDP growing faster to regions growing slower. One possible explanation is the cost of living and housing. Clark, Deurloo and Dieleman (1984), find that the number of square meters is the most consistent predictor of the propensity to move in Tilburg. Morrison and Clark (2011), analyse the migration process
19 in New Zealand and find that 55.82% of the respondents who moved in the last year did it because of housing size, satisfaction or cost. Since results show that workers migrate now to regions that grow less or have lower average GDP, this would be an interesting suggestion for further research.
As a summary, the influence of the economic performance of the regions on regional migration present a shifting pattern during the last decade and coefficients has not remained the same.
6.2. Causality exercise
In table 4a and 4b, I present the point estimates for the causality exercise described previously.
In the first table, we can see that for the dependent variable Migration 2007-2011, previous migration is significant, but also previous GDP growth of region j is positive and significant. For the dependent variable GDP growth j 2011 previous migration is not significant, but previous GDP growth of the region j is. These results suggest that migration for the period 2007-2011 is affected by previous GDP growth, but not vice versa.
The second table shows that migration 2007-2011, is affected by previous migration and by previous GDP growth of the region i. But when I regress GDP growth of the region i for the period 2007-2011 over previous migration and GDP growth, migration is not significant. These outcomes also suggest that migration does not cause GDP growth, but economic growth is determining the movement of workers.
Therefore, I construct a second exercise in which I regress Migration of the period 2011-2015, which is the latest one, over an increasing number of previous GDP growth rates of both regions i and j. The aim of this is to account for the autocorrelation of the GDP growth and to observe by how many lags is delayed the impact of this variable on the migration.
Adding previous year’s GDP growth of both destination region and origin region one by one, allows me to control for the correlation of GDP’s growth and analyse the change in the coefficients as I add new variables. As I stated before, due to data limitations, I cannot analyse the data as panel data but I can indirectly assess the correlation among variables.
20 For this period, the significant GDP growth is from region of origin in the year 2011. It is significant even after the inclusion of growth in years 2010, 2009 and 2008. However, GDP growth of the origin region in the year 2014 is significant until the inclusion of the GDP growth of the year 2010, which could indicate a seasonal correlation. GDP growth of the region j for the year 2012 is significant after the inclusion of the year 2010. We can conclude that after accounting for previous growth, GDP growth of the region of destination is significant but lagged 4 years. This means that changes in the GDP at the beginning of the period have a bigger influence on the migration decision than those closer to the middle or the end of the term.
Table 3: Model estimations Dependent variable: ln(Mij/Mii)
2006 2009 2011 2013 2015 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Population i -0.010 -0.014** -0.020*** -0.023*** 0.018*** 0.022*** 0.011*** 0.011*** -0.004 -0.003 (0.007) (0.006) (0.005) (0.005) (0.005) (0.005) (0.004) (0.004) (0.005) (0.005) Population j 0.014*** 0.016*** 0.017*** 0.019*** 0.007** 0.004 0.007** 0.008*** 0.016*** 0.015*** (0.004) (0.004) (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.004) (0.004) Urban pop i 8.072*** 8.658*** 5.357*** 5.654*** 1.454 1.142 2.084* 2.065* 2.903*** 2.863*** (1.491) (1.348) (1.586) (1.580) (1.131) (1.136) (1.214) (1.204) (1.039) (1.037) Distance -0.147*** -0.147*** -0.116*** -0.115*** -0.123*** -0.122*** -0.089*** -0.089*** -0.102*** -0.103*** (0.026) (0.026) (0.023) (0.023) (0.023) (0.024) (0.024) (0.024) (0.023) (0.022) Distance2 0.002** 0.002** 0.002** 0.002** 0.001** 0.001** 0.000 -0.000 0.001* 0.001* (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Wj 0.005** 0.004** 0.000 -0.001 0.010*** 0.010*** 0.010*** 0.010*** 0.000 0.001 (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003) Wi 0.001 0.000 0.001 0.001 0.002 0.002 0.003 0.004 0.010*** 0.010*** (0.002) (0.002) (0.003) (0.003) (0.003) (0.003) (0.004) (0.004) (0.003) (0.002) DWj 0.023* 0.021 -0.022* -0.020* 0.051*** 0.049*** -0.004 -0.004 -0.008 -0.007 (0.014) (0.013) (0.012) (0.012) (0.015) (0.015) (0.004) (0.004) (0.009) (0.008) DWi 0.002 0.004 0.004 0.004 -0.032** -0.028* -0.002 -0.002 0.029*** 0.028*** (0.020) (0.020) (0.015) (0.015) (0.015) (0.016) (0.005) (0.005) (0.008) (0.007) Unemployment i -9.816** -11.638*** -6.355 -7.456 3.696 8.547* -6.882 -7.776 2.313 2.864 (4.187) (3.835) (5.834) (5.824) (4.959) (4.883) (5.477) (5.286) (7.215) (7.385) Unemployment j -0.520 0.279 4.661 5.316 -0.527 -5.784 4.978 5.939 9.187 8.745 (2.437) (2.446) (4.332) (4.416) (4.985) (4.402) (4.867) (4.443) (7.073) (6.932) Sij 0.002 0.002 -0.005*** 0.001 -0.001 (0.001) (0.001) (0.001) (0.001) (0.002) Constant -11.693*** -11.961*** -9.670*** -9.830*** -8.072*** -7.858*** -8.184*** -8.179*** -9.948*** -9.933*** (1.011) (0.970) (0.786) (0.787) (0.876) (0.883) (0.924) (0.911) (1.124) (1.113) Observations 144 144 145 145 154 154 148 148 150 150 R-squared 0.585 0.579 0.480 0.471 0.570 0.546 0.503 0.501 0.562 0.561
22
Table 4a: Causality exercise
Dependent variable Mig 2011 GDP growth j 2011
Mig 2006 0.666*** 0.251
(0.098) (2.237)
GDP growth j 2006 0.017*** 0.713***
(0.003) (0.077)
Controls* Yes Yes
Observations 155 155
R-squared 0.768 0.664
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 *: Controls include change in urban population share of region j, change in population, distance, squared distance and S_ij.
Table 4b: Causality exercise
Dependent variable Mig 2011 GDP growth i 2011
Mig 2006 0.641*** -3.814
(0.107) (2.549)
GDP growth i 2006 0.007*** 0.768***
(0.003) (0.075)
Controls* Yes Yes
Observations 155 155
R-squared 0.783 0.672
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 *: Controls include change in urban population share of region i, change in population, distance, squared distance and S_ij.
23 On the other hand, I build the same exercise but reversing the relation of the variables. This allows me to see if any lagged migration has an impact on the GDP growth of either the region of origin and destination region. For that purpose, I subsequently add migration data, until the inclusion of the 2006 survey. In Appendix 5, I present the equations and outcomes of this exercise in tables 9a and 9b.
We observe in tables 9a and 9b that migration in the 2007-2011 period is significant to both the GDP Growth of regions i and j in 2015. In the table 9a the sign is positive, which is opposite to what I initially expected since migration reduces the GDP of the region of origin.
In table 9b, Migration 2013 is significant only after the inclusion of Migration 2011, and Migration 2011 is significant only after the inclusion of Migration 2009. Yet the coefficient of Migration 2011 is negative, it is significant only at the 10% level. These outcomes suggest a high correlation between the migration processes, indicating people have followed the same pattern and that there are not enough previous surveys to disentangle the effect. Although some coefficients are significant, the estimates for Migration 2013 have the opposite sign of what was intially expected, suggesting that migration does not improve GDP of the region of origin and that migration does not decrease the GDP of the region or origin.
This section answers the question whether migration explain economic conditions or vice versa by suggesting that, in Chile, migration is driven by economic conditions and there is no evidence of the other way around.
6.3. Shock to economic conditions
Knowing that better economic conditions determine migration, I use the price of copper as a proxy of these conditions, assuming that salaries increased and/or unemployment decreased due to this increase in the price of copper.
In table 5, we can observe in the first column that copper is positive and significant, i.e. the increase in the copper price led to a higher migration to mining industries in mining regions. When I add region effects, the coefficient decreases from 0.575 to 0.374 but remains significant at the 1% level. This is because of a leakage effect, since not only this specific industry in the regions 1, 2 and 3 benefits from the increase in the price but also the other industries do and that effect is captured by the region dummies.
24 In this table, we can see which household’s characteristics have an impact on the likelihood of migrating. Age has a negative and significant impact, as expected. Schooling years have a negative and significant impact without regional dummies, but positive and significant with them. Gender has no effect when regional dummies are included. Civil status presents a positive and significant effect meaning that married couples migrate more. The number of people captures partially the effects of children, and reduces the likelihood of migration.
Table 5: Copper price as a treatment Dependent variable: Mig dummy 2009
(1) (2) Copper 0.575*** 0.374** (0.156) (0.150) Age -0.022*** -0.005*** (0.001) (0.001) Schooling years -0.016*** 0.022*** (0.003) (0.003) Civil status 0.209*** 0.146*** (0.025) (0.024) Gender -0.053*** 0.003 (0.020) (0.020) Health index -0.144*** -0.001 (0.006) (0.009) Person number -0.116*** -0.026*** (0.007) (0.007) Native indigenous -0.087** -0.018 (0.036) (0.039) Housing subsidy -0.287*** -0.181*** (0.022) (0.022) Current income 0.007*** 0.005*** (0.001) (0.001) Rural -0.173*** -0.036 (0.022) (0.023)
Region Effects No Yes
Observations 138,988 138,988
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Civil status: 1 if married or in a civil union. Gender: 1 if the head of household is a man. Health index: from 1 to 7, 7 the highest. Native: 1 if belonging to a native population. Housing subsidy: 1 if the property was bought with a subsidy. Rural: 1 if currently living in a rural province.
25 7. Discussion and conclusions
In this thesis, I have analysed the factors that influence the decision of people to migrate. Since individuals maximize their utility and decide which region to migrate to, taking income into consideration, it is expected that they migrate to regions where these variables are high or improving. This logic is consistent with the results found from the first surveys, where coefficients for the economic performance of the regional destination were positive and significant. However, over time these coefficients become insignificant. At the same time, the coefficients for the economic conditions of the origin regions turned out to be positive and significant. This indicates that migration is flowing into regions with lower and/or decreasing salaries from regions with higher growth.
Workers are leaving the capital and the main mining regions, suggesting that other variables such as housing price, living expenses, or even amenities, may be driving the decision now.
Secondly, I have analysed the relationship between the economic conditions, GDP growth and migration. The results show that migration is caused by the improvement of the economic circumstances and not vice versa. This is an assumption of the regional migration model estimated in the first part that I used for the Chilean scenario and it was worth testing, due to the non-conclusive literature. This supports the idea that regional results are not just correlations, but a causal relationship.
Thirdly, using an exogenous increase in the price of copper between the years of 2006 and 2008, I estimated a probit model at a household level where I assessed whether this rise -as a proxy for economic conditions- had led to an increase of migration toward the mining regions. Results supported the previous hypothesis.
This research includes a study of migration at a household level which has not been analysed previously in Chile. This adds value to the international literature. It also contributes to the discussion about the causality of the effect and expands the analysis to a period of time that was not covered before. It presents an outcome at a regional level that was not expected and it opens up the discussion about why workers are not choosing areas with higher economic
26 conditions anymore. However, the lack of regional prices hinders further analysis of these findings.
From a policy perspective, even though the causality flows from development to migration, it does not mean that mobility is not relevant itself as it is a tool to bring regional convergence along. Migration, due to the reverse causality, can strengthen the progress of certain regions, especially the ones less developed.
For further research, there are two improvements that can be done: First, the construction of a variable of regional living expenses could be added to the analysis. As explained before, the results from the last surveys reflect that patterns are changing and one possible explanation could be that the increasing price of housing in high growth regions is impacting negatively the migration flow towards those regions. The second improvement is to create a more reliable proxy for expected income, such as median salary per region, which is also not available. If these two factors are included, a more accurate estimation of the determinants of migration could be carried out, as well as a more thorough understanding of this process and its link with development.
27 8. References
Aroca, P. & Hewings, G., 2002. Migration and Regional Labor Market Adjustment: Chile 1977-1982 and 1987-1992. The Annals of Regional Science, 36, pp.197–218.
Barham, B. & Boucher, S. 1998. Migration, Remittances, and Inequality: Estimating the Net Effects of Migration on Income Distribution. Journal of Development Economics, 55(2), pp. 307–31.
Ben-Akiva, M. & Lerman, S. 1985. Discrete Choice Analysis. The MIT Press, Massachusetts. Berkson, J. 1944. Application of the logistic function to bio-assay. Journal of the American
Statistical Association, 39(227), pp. 357-365.
Boots B. & Kanaroglou, P. 1988. Incorporating the effects of spatial structure in discrete choice models of migration. Journal of Regional Science, 28, pp. 495–507
Borts, G. H. & Stein., J.L.1964. Economic growth in a free market. New York: Columbia University
Press.
Clark, W., Deurloo, M. & Dieleman, F. 1985. Housing consumption and residential mobility. Annals of the Association of American Geographers 74 (1), pp. 29-43.
Donoso, M., 2014. Chilean copper market facing the world financial downturn. Ingeniare. Revista Chilena de Ingeniería, 22(1), pp. 99-115.
Fields, G., 1982. Place-to-Place Migration in Colombia. Economic Development and Cultural Change, 30 (3), pp. 539-558.
Gabriel, S., Shack-Marquez, J. & Wascher, W. 1993 Does migration arbitrage regional labor market differentials? Regional Science & Urban Economics, 23(2), pp. 211–233.
Greenwood, MJ. 1995. Internal migration in developed countries. In: Rosenzweig MR, Stark O(eds) Handbook of population and family economics (eds). North-Holland, Amsterdam. Greenwood, MJ. 1985. Human migration: theory, models, and empirical studies. Journal of Regional Science, 25(4), pp. 521-544.
28 Greenwood, MJ. 1981. Migration and economic growth in the United States: National, regional,
and metropolitan perspectives. New York: Academic Press.
Greenwood, MJ. 1975.Research on internal migration in the United States: a survey. Journal of Economic Literature, 13 (2), pp. 397-433.
Greenwood, MJ. & Hunt, G. 1989. Jobs versus amenities in the analysis of metropolitan migration. Journal of Urban Economics, 25(1), pp. 1-16.
Herzog, H.W., Schlottmann, A. & Boehm, T. 1993. Migration as spatial job-search: a survey of empirical findings. Regional Studies, 27(4), pp. 327-340.
Hughes, G. & McCormick, B. 1989. Does migration reduce differentials in regional unemployment rates? Migration and Labour Market Adjustment, J. Van Dijk et al. (eds.) Dordrecht, Kluwer Academic Publishers.
Kanaroglou, P.S. & Ferguson, M.R. 1996. Discrete spatial choice models for aggregate destinations. Journal of Regional Science, 36, pp. 271–290.
Molho, I. 1986. Theories of migration: A review. Scottish Journal of Political Economy, 33(4), pp. 396–419
Morrison, P. & Clark, W. 2011. Internal migration and employment: Macro flows and micro motives. Environmental and Planning A, 43(8), pp. 1948-1964.
Paredes, J. & Vega, V., 1996. Econometría Espacial aplicada al estudio de la migración en Chile. Thesis. Universidad Católica del Norte, Chile.
Perloff, J., Lynch, L. & Gabbard, S. 1998. Migration of Seasonal Agricultural Workers. American Journal of Agricultural Economics, 80(1), pp. 154–64.
Ravenstein, E. G. 1889. The laws of migration. Journal of the Royal Statistical Society, 52, pp.241–301.
Richardson, H W. 1978. Regional economics. Urbana: University of Illinois Press.
Rogers, A. & Castro, LJ. 1981. Model migration schedules. IIASA Research Report. IIASA, Laxenburg, Austria: RR-81-030
Sala-i-Martin, X. 1996. The Classical Approach to Convergence Analysis. The Economic Journal, 437(106), pp. 1019-1036
29 Sjaastad, L. A. 1962. The costs and returns of human migration. Journal of Political Economy,
70(Supplement), pp. 80–93.
Soto, R. & Torche, S., 2004. Spatial Inequality, Migration and Economic Growth in Chile. Cuadernos de Economía,41, pp.401-424.
Stark, O. & Taylor, J.E. 1991. Migration Incentives, Migration Types: The Role of Relative Deprivation. The Economic Journal, 101(408), pp. 1163–78.
Stark, O. & Taylor, J.E. 1989. Relative Deprivation and International Migration. Demography, 26, pp. 1–14.
Taylor, J.E. 1987. Undocumented Mexico-U.S. Migration and the Returns to Households in Rural Mexico. American Journal of Agricultural Economics, 69, pp. 626–38.
Taylor, J.E. 1992. Earnings and Mobility of Legal and Illegal Immigrant Workers in Agriculture. American Journal of Agricultural Economics, 74(4), pp. 889–896.
Williamson, J., 1965. Regional inequality and the process of national development. Economic Development and Cultural Change, 13 (4), pp. 3–45.
Xu, Z. & Li, S. 2008. The effect of inter-regional migration on economic growth and regional disparity. The Journal of Quantitative & Technical Economics, (2), pp. 38–42.
Zhang, K. & Song, S. 2003. Rural-urban migration and urbanization in China: Evidence from time-series and cross-section analyses. China Economic Review, 14(4), pp. 386-400.
30 9. Appendix
31 9.2. Appendix 2
Table 6: Year and source of Independent variables
Year of the
survey 2006 2009 2011 2013 2015 Source
Population 2006 2009 2011 2013 2015 Instituto Nacional de Estadísticas
Urban
population 2006 2009 2011 2013 2015 Instituto Nacional de Estadísticas
Unemploy
ment 2002-2003 2005-2006 2007-2008 2009-2010 2011-2012 Banco Central de Chile
Wage 2001 2004 2006 2008 2010 Banco Central de Chile
Dwage 2003-2005 2005-2008 2007-2010 2009-2012 2011-2014 Banco Central de Chile
Distance by
32 9.3. Appendix 3
Table 7: 𝑀𝑖𝑗 and 𝑀𝑖𝑖 for the period 2011-2015
region j 1 2 3 4 5 6 7 8 9 10 11 12 13 region i 15989 68 23 57 45 12 10 13 9 6 2 1 45 2 154 6881 63 127 54 18 1 26 6 22 0 2 43 3 28 19 6915 72 22 7 4 6 2 3 2 1 14 4 57 42 57 9159 36 30 2 8 7 5 11 5 45 5 66 40 29 54 18173 50 17 53 12 30 17 52 159 6 25 5 13 6 59 14714 54 26 18 9 10 4 77 7 28 4 13 0 25 52 13107 35 24 29 10 6 70 8 50 19 24 1 50 63 45 28383 89 80 37 26 127 9 32 21 18 8 20 32 19 82 15442 100 31 10 57 10 25 24 18 7 50 28 5 78 100 22201 74 56 45 11 0 0 0 0 0 5 1 10 21 29 4859 16 5 12 2 0 4 8 27 12 3 13 4 30 16 5226 20 13 209 88 86 155 338 267 226 340 269 259 84 69 32603
33 9.4. Appendix 4
The equations are as follows:
𝑀𝑖𝑔2015𝑖𝑗 = 𝛼0+ 𝛼1𝐺𝐷𝑃2015𝑖 + 𝛼2𝐺𝐷𝑃2015 𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (10) 𝑀𝑖𝑔2015𝑖𝑗 = 𝛼0+ 𝛼1𝐺𝐷𝑃2015𝑖 + 𝛼2𝐺𝐷𝑃2015 𝑗 + 𝛼3𝐺𝐷𝑃2014𝑖 + 𝛼4𝐺𝐷𝑃2014 𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (11) 𝑀𝑖𝑔2015𝑖𝑗 = 𝛼0+ 𝛼1𝐺𝐷𝑃2015𝑖 + 𝛼2𝐺𝐷𝑃2015 𝑗 + 𝛼3𝐺𝐷𝑃2014𝑖 + 𝛼4𝐺𝐷𝑃2014 𝑗 + 𝛼3𝐺𝐷𝑃2013𝑖 + 𝛼4𝐺𝐷𝑃2013 𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 2015 𝑖𝑗 + 𝑢 𝑖𝑗 (12) (…)
34
Table 8: Migration on previous GDP Growth yearly Dependent variable: Mig 2015
(1) (2) (3) (4) (5) (6) (7) (8) GDP growth i 2015 0.015* -0.001 0.002 0.005 -0.018 -0.024 -0.044*** -0.045** (0.008) (0.010) (0.010) (0.010) (0.014) (0.016) (0.016) (0.018) GDP growth j 2015 -0.001 -0.003 0.001 -0.006 -0.003 0.019 0.003 0.008 (0.007) (0.008) (0.008) (0.011) (0.012) (0.017) (0.024) (0.024) GDP growth i 2014 0.038*** 0.035** 0.032** 0.023* 0.017 -0.004 -0.000 (0.014) (0.014) (0.014) (0.012) (0.013) (0.018) (0.018) GDP growth j 2014 0.001 -0.004 0.003 0.004 0.017 0.001 0.004 (0.008) (0.009) (0.011) (0.010) (0.016) (0.018) (0.019) GDP growth i 2013 0.012 0.017* -0.014 -0.015 -0.024* -0.010 (0.008) (0.010) (0.015) (0.014) (0.014) (0.012) GDP growth j 2013 0.018* 0.006 0.011 0.015 0.007 0.000 (0.010) (0.014) (0.016) (0.015) (0.019) (0.026) GDP growth i 2012 0.003 0.005 0.005 0.004 0.003 (0.004) (0.004) (0.004) (0.004) (0.004) GDP growth j 2012 -0.007 -0.007 -0.008* -0.009* -0.009* (0.005) (0.004) (0.004) (0.005) (0.005) GDP growth i 2011 -0.036*** -0.039*** -0.056*** -0.047*** (0.012) (0.013) (0.013) (0.012) GDP growth j 2011 0.003 0.014 0.001 -0.000 (0.008) (0.011) (0.017) (0.021) GDP growth i 2010 -0.008 -0.022* -0.015 (0.008) (0.012) (0.012) GDP growth j 2010 0.020 0.010 0.012 (0.014) (0.016) (0.017) GDP growth i 2009 -0.052** -0.071*** (0.024) (0.024) GDP growth j 2009 -0.040 -0.019 (0.041) (0.037) GDP growth i 2008 -0.032** (0.016) GDP growth j 2008 0.016 (0.024) Constant -0.014*** -0.013*** -0.013*** -0.013*** 0.000 -0.005 0.010 0.009 (0.005) (0.005) (0.004) (0.004) (0.006) (0.008) (0.011) (0.011)
Controls* Yes Yes Yes Yes Yes Yes Yes Yes
Observations 155 155 155 155 155 155 155 155
R-squared 0.349 0.418 0.434 0.447 0.510 0.517 0.543 0.552
*: All regressions include urban population i, urban population j, population i, population j, S_ij, distance and squared distance as controls. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
35 9.5. Appendix 5 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2015𝑖 = 𝑀𝑖𝑔2015 𝑖𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (13) 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2015𝑖 = 𝑀𝑖𝑔2015𝑖𝑗 + 𝑀𝑖𝑔2013𝑖𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (14) 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2015𝑖 = 𝑀𝑖𝑔2015 𝑖𝑗 + 𝑀𝑖𝑔2013𝑖𝑗 + 𝑀𝑖𝑔2011𝑖𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (15) (…)
Table 9a: Region i's GDP Growth on previous Migration Dependent variable: Region i's GDP growth 2014-2015
(1) (2) (3) (4) (5) Mig 2015 1.324 1.306 1.930 2.044 2.426 (0.998) (1.620) (1.523) (1.643) (2.005) Mig 2013 0.031 -3.721 -3.787 -3.895 (2.085) (2.426) (2.457) (2.562) Mig 2011 3.880** 3.949** 4.156** (1.568) (1.603) (1.643) Mig 2009 -0.232 0.228 (1.181) (1.490) Mig 2006 -1.161 (2.422) Constant 0.159*** 0.159*** 0.167*** 0.167*** 0.163*** (0.037) (0.037) (0.037) (0.037) (0.039)
Controls* Yes Yes Yes Yes Yes
Observations 155 155 155 155 155
R-squared 0.158 0.158 0.191 0.192 0.193
*: All regressions include urban population i, urban population j, S_ij, distance, squared distance and populations as controls. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
36 𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2015𝑗 = 𝑀𝑖𝑔2015𝑖𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (16)
𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2015𝑗 = 𝑀𝑖𝑔2015𝑖𝑗 + 𝑀𝑖𝑔2013𝑖𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (17)
𝐺𝐷𝑃 𝑔𝑟𝑜𝑤𝑡ℎ2015𝑗 = 𝑀𝑖𝑔2015𝑖𝑗 + 𝑀𝑖𝑔2013𝑖𝑗 + 𝑀𝑖𝑔2011𝑖𝑗 + 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠2015𝑖𝑗 + 𝑢𝑖𝑗 (18)
(…)
Table 9b: Region j's GDP Growth on previous Migration Dependent variable: Region i's GDP growth 2014-2015
(1) (2) (3) (4) (5) Mig 2015 0.279 -0.685 -1.137 -1.743 -1.779 (0.859) (1.624) (1.710) (1.848) (1.868) Mig 2013 1.616 4.335* 4.682* 4.692* (1.962) (2.500) (2.538) (2.568) Mig 2011 -2.812 -3.175* -3.195* (1.755) (1.730) (1.867) Mig 2009 1.225 1.180 (1.012) (1.518) Mig 2006 0.111 (2.542) Constant 0.146*** 0.151*** 0.145*** 0.144*** 0.145*** (0.033) (0.036) (0.036) (0.036) (0.037)
Controls* Yes Yes Yes Yes Yes
Observations 155 155 155 155 155
R-squared 0.146 0.149 0.166 0.176 0.176
*: All regressions include urban population i, urban population j, S_ij, distance, squared distance and populations as controls. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
