Long-run Impact of
Globalization on
Unemployment
Evidence from a Longitudinal world data
Xuxu Chen (10227253)
B. A. Universiteit van Amsterdam
Final version: 01.07.2014
Abstract: This paper focuses on the long-run impact of globalization on unemployment rate using a panel data from 1980-2013 for 12 countries. The main empirical method uses the robust standard error
estimation and Feasible General Least Square (FGLS) model to allow for heteroskedasticity and serial correlation. It finds that, in the long run, 1% increase in openness to trade reduces unemployment rate by 9% at the aggregate level. A subgroup analysis is conducted to further explore the impact of the export and import on unemployment level. It shows that import has a positive effect on unemployment rate as previous literature indicates, but export also has a positive impact on unemployment rate. This might happen due to change of structure, technology and change of capital endowment.
Table of Content
1. Introduction ... 3
2. Literature Review ... 5
3. Methodology ... 7
3.1 The model ... 7
3.2 Heteroskedasticity and serial correlation ... 10
3.3 Advantages of panel data ... 11
3.4 Issues involved in utilizing panel data ... 12
4. Data Description ... 13
4.1 The unemployment rate ... 13
4.2 Openness measurement ... 14
5. Empirical result ... 15
5.1 The summary statistics ... 15
5.2 The benchmark results ... 16
5.2.1 All countries ... 16
5.2.2 Subgroup Analysis ... 18
6. Conclusion ... 21
Bibliography ... 23
1. Introduction
Since the significant change in trade policies of many least developed countries during 1970s, international trade of goods and services has increased rapidly. The globalization has arrived at a new era, thanks to the globally negotiations on reduction of trade barriers; reduction in communication and shipping costs; acceleration of outsourcing activities and global culture integration. In 2008, $16 trillion goods and services of the total $50 trillion world production were sold across national borders; United States triples its level of imports and exports as shares of GDP from 1960 to 2009 (Krugman, 2009, p.10).
While consumers are enjoying the benefits of trade liberalization, the impact of globalization on unemployment is somehow ambiguous and often changes in a country-by-country level or even industry-by-industry level within one country-by-country. Understanding how globalization would affect the unemployment rate becomes a serious concern among policy makers and economists.
In developed countries, people earn higher wages often worried that the negative impact of globalization will result in a job replacement by lower-wage workers in less-advanced nations. Ross Perot, a presidential candidate of United States colorful phrased the negative impact of the North American Free Trade Agreement (NAFTA) as the “giant sucking sound” during 1992 U.S. presidential campaign. Supporters of NAFTA think the enhanced production for exports has the potential to create jobs while oppositions clarify the intensified free trade reduces job positions in U.S. by plant closures (Capuano & Schmerer, 2013). Does international trade create or destroy jobs? Globalization is associated with numerical changes in the labor market, such as changes in the labor demand, in wage gap between skilled and unskilled workers, in productivity and in employment elasticity. Research shows that owners of the industrial specific resources and
workers with specific skills are more likely be harmed by international trade (Krugman,2009. p.4). For example, when a famer loses his job due to increasing import of agriculture products, he cannot easily secure another job immediately without further education. These concerns have brought up the real-world policy debate about the impact of the free trade due to the fact that the growth rate of less-skilled workers’ wage in some developed countries has been far less than their growth rate of GDP. Unlike most of the developed countries where imports greatly exceed exports, the German labor market relies more heavily on its exports after the rise of “the East”, partly because of the higher demand of German products from China. Capuano & Schmerer (2013) show that 30 percent of total employees work for exports manufacturing in Germany. In 2011, its exports exceeded 1 trillion Euros and the sales of manufacturing products to China increase from 9 to 54 billion Euros between 2000 and 2010. The rapid expansion of export increases the labor demand thus reduces the unemployment in Germany.
A better understanding of the impact of free trade on unemployment level is necessary for policy makers to design a more suitable policy environment for workers. Felbermayr et al. (2011) found globalization will worsen worker’s prospects on the labor market unambiguously in the short run, because of the increasing search frictions during the transaction period, means people who lost their job due to an increasing in imports need time to search and learn for another job. It indicates that the short-term unemployment created by globalization is structural. Lasting debate on the justice of trade liberalization indicates that this issue is still remaining unsolved. Many of the papers emphasize on the short-run effect while there is little research about the long-run impact of international trade on unemployment. To fully understand the cost and benefit of free trade, it is very important understand how globalization affect the country in the long run.
The paper contributes to address the long-run effect of globalization issue by providing an empirical basis for some of the arguments in the debates using countries’ aggregate data from 1980 to 2013 of 12 countries (Brazil, China, France, Germany, India, Indonesia, Japan, Malaysia, Philippine, South Africa, United Kingdom and United States). It aims to present an empirical idea of the causal relationship between unemployment rate and openness to trade in the long term. Also, it further empirically proves the negative relationship between unemployment rate and openness to trade use panel data across 12 countries, meanwhile looking at the impact of increasing imports and exports on unemployment level.
The rest of the paper is organized as followings. Section 2 provides a literature review of both short-run and long-run effect of globalization. Section 3 introduces the methodology used in this paper. Section 4 provides the detailed explanation of the panel data used in this paper.
Section 5 shows the empirical results and conclusion is provided in the last section.
2. Literature Review
The globalization is defined as “the integration of international commodity markets” (O’Rourje, 2001). Most papers show that a relation exists between openness to trade and unemployment; however, the sign of the relationship remains controversial.
Earlier studies indicate the sign of the relationship depends on several elements.
According to the study of Brecher (1974), free trade will not increase employment and welfare when home exports capital-intensive products, an increase in the foreign demand of home capital-intensive goods will actually further raise the unemployment rate. Davidson and Matusz (1988) show that under the assumptions of constant return to scale technologies and homothetic preferences, the sign of the relationship between free trade and unemployment depends on a comparison of the minimum wage level and capital-labor endowment across the countries. More
specifically, when a foreign country has a lower minimum wage and capital-labor endowment, it has a competitive advantage in production of labor-intensive goods, so home will tend to import labor-intensive goods from foreign and export capital-intensive goods.
The research conducted by Davidson and Matusz (1988) provides a source of great concern about the impact of growing trade with least developed countries. Due to the fact that most developed countries have higher minimum wage and capital-labor endowment compare to developing countries, developed countries tend to produce more capital-intensive and less labor-intensive goods. More recent researches show a huge concern about the unemployment rate in developed countries. According to Hine & Wright (1998), the UK employment in the
manufacturing sectors decline rapidly since 1970 due to the increasing trade with low-wage economies. Increasing import of labor-intensive manufactures such as clothing lowers the labor demand; comparative cost advantage of developing countries reduces the number of firms in domestic industry. Both effects contribute to the decline of employment and wages among unskilled, blue-collar workers in developed countries. Ghose (2000) improves the findings by examining the validity of the relevant empirical results using available statistical data. He concludes that both skilled and unskilled workers in developed countries suffer from increasing trade in manufacturing sectors due to import competition, while in some developing countries where its export is more labor-intensive than its import, increasing trade stimulates employment growth in all branches of manufacturing industries.
Bernard et al. (2007) find that in the short run, many empirical evidences show that the trade liberalization raises unemployment due to the increase of job turnover. Lamo et al (2006) indicates that workers are not mobile within sectors due to their skill-specification, so free trade will destroy more jobs than it will create. Egger and Kreickemeier (2009) use a model with fair
wages and firm heterogeneity and find trade liberalization increase unemployment, because the global competition obsolete less productive producers results in involuntary unemployment.
However, most of the papers study the short-run affects on unemployment while long-run impact remains unclear, except a few studies like Janiak (2007) who performs a numerical analysis using US economy data over 1974-1988. They find that globalization leads to more job destruction than creation. It shows that a one-percentage increase of import raises job destruction rate by 14.7 percent and none in job creation. A one-percentage increase of export raises job destruction rate by 6.5 points and raises job creation rate by 4.5 point.
As indicated by Davidson and Matusz (2004), the impact of globalization on
unemployment is primarily an empirical issue but there is very little empirical work. Felbermayr et al (2011) contribute to this question by a thorough quest for the long term causal relationship between the unemployment and openness to trade. Their empirical research shows
unemployment rate has a non-positive relationship with openness to trade indicating that opening up to trade will not increase unemployment rate in the long run.
3. Methodology
For the purpose of analyzing the long-run effect of free trade on unemployment, panel data regression model is adopted to analyze a given sample of 12 individual countries over 34 years in order to provide multiple observations of each country.
3.1 The model
A variable-intercept model is adopted to address the heterogeneity problem across different countries through 34 years using panel data. The basic assumption of this kind of models is that, apart from the observed variables, there are only three kinds of variables that will
lead to the problem of omitted variables bias (individual time-invariant variables1, period individual-invariant variables2 and individual time-varying variables3).
For control variables, I included: Gross Domestic Product (GDP), the real interest rate and the per capita income. The relationship between GDP and unemployment rate can be explained by Okun’s law4
, which indicates that GDP is negatively related to unemployment rate since the output depends on the total labor used to contribute in the production process. Change of real interest rate could potentially affect the capital flows of international investment, which might have an impact on the unemployment rate. For example, when a country’s real interest rate increase and others remain the same, investors will invest more in this country in order to get a higher return. The capital mobility will indirect affect unemployment rate through firms’ hiring decision.
We calculate per capita income as total GDP divide by total population; it expresses the standard of living of one country in comparison of others. Moreover, a higher standard of living suggests higher wage, so there is a positive relationship between per capita income and wage. I use per capita income because it measures the wealth of the population; it is widely known and can be calculated easily from readily available GDP and population estimates. If a foreign country has a much lower average wage compare to home country, the home companies might
1
The individual time-invariant variables are those variables that differ across cross-sectional units but similar through time.
2
The period individual-invariant variables are those variables that differ through time but similar across cross-sectional units.
3 The individual time-varying variables are those variables that differ across cross-sectional units and through time.
The model assumes that the time-varying variables are individually unimportant but collectively significant and uncorrelated with all other variables.
4 In economics, Okun’s law (named after Arthur Melvin Okun) is an empirically observed relationship relating
unemployment to losses in a country’s production, for more details please refer to: Okun, Arthur M. "The gap between actual and potential output." Proceedings of the American Statistical Association. 1962.
abandon domestic plants and open new ones in the foreign country in order to save money. That might increase the domestic unemployment rate.
A Hausman test is conducted to test select the model between the fixed effect model and the random effect model. The test result suggests applying a fixed-effect model, in which researchers make conditional inference on the effects within the sample, unlike the random-effect model in which investigators make unconditional or marginal inference respect to the population of all effects. So the difference between a fixed-effect test and a random-effect test is whether an investigator decides to make inference with respect to the sample effects or to the population characteristics. Such decision is mainly based on the context of the data, how the data was gathered and where does the data come from. Nonetheless there is no distinction in the nature of effects.
The following fixed-effect model is used to estimate the relationship between international trade and unemployment level.
Eq. (3.1.1)
i = 1,….,N, ( N=12);
t = 1,….,T. ( T=34).
Where the left-hand side of the equation is the natural logarithm of unemployment rate of country i at time t and the right-hand side are the independent variables, which are natural logarithm of import, export, gross domestic product, openness to trade and per capita income respectively, r refers to the real interest rate, is the error term. In this model, we assume there is no cross sectional correlation but the error term is allowed to be heteroskedasticity and serial correlated.
In the fixed-effect model, two dummy variables are introduced to allow for the effects of those omitted variables, which are the period individual-invariant variables that are different across the time period but constant for all cross-sectional units and the individual time-invariant variables that are different among all cross-sectional units and constant over time. A subgroup analysis is also conducted by dividing countries into developed and developing countries. The purpose of the subgroup analysis is to see the different labor market reactions with respect to the international trade.
3.2 Heteroskedasticity and serial correlation
I use the robust standard error estimation in fixed-effects panel models for all countries to allow for heteroskedasticity and serial correlation. Homoskedasticity means the variance of the error term is constant; it can arise due to the varying size of the cross-sectional units in the panel data. Serial correlation can also occur due the effect of omitted or transitory variables last more than one period or vary systematically over time.
In the subgroup analysis, the ordinal least square estimators will lead to the wrong inference. When the variances are unknown, a feasible general least square estimator can be implemented by replacing the unknown value by their estimators5. Both the heteroskedasticity and serial correlation problem can be corrected by the feasible general least square estimators.
5
In a model , if we assume the error term is homoskedasticity, the OL S estimator is given by
. Now, if the error term is heteroskedastic, we can divide regression equation
by . So we get . Now the error term is homoskedastic and the General Least Square estimate is:
.by . So we get . Now the error term is homoskedastic
and the General Least Square estimate is:
.estimate is:
.by . So we
get . Now the error term is homoskedastic and the General Least Square estimate is:
3.3 Advantages of panel data
Panel data analysis is becoming widely used in economic research, and according to Hsiao (2005), there are several major reasons that panel data fits better to my empirical research compare to conventional cross-sectional and time-series data set. Firstly, panel data increases the efficiency of the estimations by giving a larger number of degrees of freedom and reduces the collinearity problems among variables. Secondly, panel data can analyze some important economic questions which cross-sectional or time-series data sets cannot address. For example, cross-sectional data is not a proper method to infer about the dynamic changes of unemployment level, because cross-sectional data cannot address the long-run effect of trade on unemployment since it solely contains a dataset for a given time. Furthermore, even if a simple one-year causal relationship between trade and unemployment is expected, it is only possible if data of additional variables, such as labor policies and people’s attitudes towards working of each country in order to estimate the different characteristics of each labor market, whereas the data on such variables are ambiguous and hard to access. However, panel data controls for such effects with dummy variables, which allow researchers to form a proper structure to analyze the causal relationships.
A Single time-series dataset cannot be used to estimate dynamic effects either because of perfect multicollinearity problem of explanatory variables6, but panel data can reduce this problem by utilizing the interindividual differences in x values. Cross-sectional data is not used because it cannot address long-run dynamic problems.
In addition, time-series data is not used since it may cause the multicollinearity problem and cross-sectional data is not significant because it cannot address long-run dynamic problems.
6
For example, consider the estimation of a distributed-lag model: , t=1,…,T, where is an
exogenous variable and is a random disturbance error term. In general, is close to , and still near ; so fairly strict multicollinearities appear among , explanatory
Overall, panel data fits perfectly for my data and generates more accurate predictions because it reduces the magnitude of a key problem that always arises in econometric empirical research: the existence of omitted variables that are correlated with the independent variables. If such case occurs, only fixed effect model could be used. Panel data controls for the effects of unobserved or missing variables by using the information of the individuality of entities and intertemporal dynamics7. Hereby, I use panel data in this paper and correct for the entity and time effect for each country to increase the consistency and unbiasness of the estimators.
3.4 Issues involved in utilizing panel data
The panel dataset involves two dimensions: A cross-sectional dimension of 12 countries and a time series dimension across 34 years (1980-2013). The panel data is based on two
important assumptions. The first one is: the data are generated from controlled experiments, and the random variables are normally distributed. Under this assumption, panel data will have the theoretical ability to isolate the specific effect and standard statistical method can be applied, otherwise heterogeneity problem might occur. The second assumption is: samples are randomly drawn from population. If the samples are not randomly selected, it will introduce a correlation between independent variables and error term, which will lead to a selectivity bias.
7
For example, consider a simple regression model: , ( i=1,…,N., t=1,…,T.), with mean
0 and variance , the ordinary least square provide the unbiased estimators and . If we suppose the value of is unobserved and related to , then the ordinary least square estimators are biased. Now if we
adopt panel data where repeated observations are available, we may forget about the effect of z. For example, if we assume , . The difference of first individual observation:
. The deviation from the mean: + ( .
Where and the mean of error team equals 0. Now the estimator is unbiased and
consistent. If we have a single cross-sectional data or a single time-series data the estimator won’t be consistent unless there is a instrumental variable that are correlate to x and uncorrelated with z and error term.
4. Data Description
The data are from a panel dataset of World Economic Outlook (WEO)8 database
constructed by World Economic and Financial Surveys and the World Bank, updated in October 2013. It contains a total of 34 years dataset (1980-2013) among 12 countries, which are Brazil, China, France, Germany, India, Indonesia, Japan, Malaysia, Philippine, South Africa, United Kingdom and United States. This section discusses the data used in the empirical research and provides a first glance of the relationship between unemployment and different measurements of openness to trade.
4.1 The unemployment rate
The unemployment rate expresses the percentage of unemployed individuals within the total labor force. Under the perfect market situation, the unemployment rate measures the excess supply in the labor market. It can be defined in several ways: The national definition,The Organization for Economic Co-operation and Development (OECD) harmonized definition and the International Labor Organization (ILO) harmonized definition. Among these different institutions, the calculation of the unemployment rate can be differed. The national
unemployment data only included those laid-off workers that are registered in the national labor office. The OECD rate gives the number of unemployed people as a percentage of the total labor force. The ILO unemployment rate is more internationally comparative because it provides an indication of unemployed workers who are without work but are available for working and are actively seeking for a job. However the drawback of the ILO unemployment rate is that the perfect information is required. For instance it is impossible to have a full list of the channels for
8
The World Economic Outlook (WEO) database contains selected macroeconomic data series from the statistical appendix of the World Economic Outlook report, which presents the IMF staff’s analysis and projections of economic developments and the global level, in major country groups and in many individual countries. The WEO is released in April and September/ October each year.
seeking jobs, etc. Hence the controversy construction of the ILO rate makes it less reliable than OECD unemployment rate.
This paper adapts the OECD unemployment rate definition, which is the total
unemployed populations divide by the total labor force. However, the accuracy of the rate is solely depending on the data that are provided by the national statistical bureaus of each country. The data quality problem is smaller in the OECD countries than in non-OECD countries, in some developing countries, the data quality highly depends on the countries’ characteristics, which is problematic. The data quality problem is one of the potential weaknesses of this paper since the measurement error of independent variable might bias the estimated effect of openness to trade. In this paper, two separated analysis are conducted for OECD countries and developing countries respectively. Thus, the potential data quality issue will be less of a concern in the analysis of the OECD countries.
4.2 Openness measurement
Three variables are used in this paper to measure the level of openness to trade: Export volume, Import volume and Openness to trade. The first two varialbes measure the total trade volume of manufacturing goods between countries and are expressed in million U.S. dollars. The trade volume of service is not included due to unavailable of data. This could potentially cause a biased estimation since economic growth is a process of industrial upgrading together with capital accumulation and technological advances. More specifically, after a long time of capital accumulation, the trade structure in some countries will gradually change from labor intensive goods (manufacturing) towards capital intensive goods (service). This issue is mentioned consideration in our empirical analysis in the later section.
The value of openness to trade is calculated as total export volume plus total import volume divide by nominal GDP. The advantage to use this measure is that it reflects the extent of globalization with respect to the international trade meanwhile the data are available and easy to measure. However, the drawback of this measure is the openness of an economy in respect of its current trade policy cannot be observed. But this problem has little impact on the value of this research because the pure relationship between the international trade and unemployment is what the paper is looking for in order to help government to adopt the most favorable policy.
5. Empirical result
In this section, I show the summary statistics for the outcome variables. Then I present the benchmark results of the regression to see the causal relationship between openness to trade and unemployment for all countries. Finally A subgroup analysis is conducted to give some economic intuitions to explain the extraordinary effect between export volume and
unemployment rate.
5.1 The summary statistics
Table 1 shows the summary statistics for the outcome variables. From the table we can see that the developing countries have a higher average unemployment rate as well as a higher average openness to trade than developed countries, we cannot just simply conclude that a positive relationship exists between unemployment rate and globalization. There are many possible reasons can explain the lower average openness to trade in developed countries, one is the lager average GDP compare to the import and export volume, which makes the denominator of openness to trade in developed countries much bigger. One possible explanation for a higher average unemployment rate in developing countries is because the larger number labor force in some of these countries, such as China and India. Furthermore problems like information
asymmetry are more severe in developing countries since the labor markets are less mature than in developed countries.
From the table we can also find an interesting fact that the standard deviations of all the variables in developing countries are all very large, which implies a huge gap that exists among those counties. More specifically, some developing countries have gone through a long period of rapid economic growth, which makes them more advanced than the rests.
Table 1. Summary statistics
Variables Observations Num. Mean Standard deviation Min. Max.
developing developed developing developed developing developed developing developed developing developed
Unemployment 238 170 7.506 6.954 6.259 2.493 1.52 2.022 11.78 28.15
GDP 238 170 570,290 3.72E+06 1.15E+06 4.01E+06 12,475 441,025 8.94E+06 2.74E+07 Import 238 170 111,826 513,914 247,165 467,698 6,476 93,914 1.87E+06 2.34E+06 Export 238 170 125,571 463,716 294,953 349,304 6,461 92,430 2.19E+06 1.59E+06
Openness 238 170 0.595 0.343 0.537 0.158 0.102 0.0593 0.811 3.542
Real interest
rate % 238 170 7.945 4.715 11.96 2.398 -24.6 -2.783 10.37 78.79
Income 238 170 2,426 29,249 2,444 26,273 30.43 7,818 12,584 332,480
(Note: The GDP, import, export and income are measured in million U.S dollars.)
5.2 The benchmark results
5.2.1 All countries
Table 2 presents the results of the regressions include all the countries using random effect model, fixed effect model and robust standard error for fixed effect model respectively. The dependent variable is the natural logarithm of the OECD unemployment rate and all variables range from 1980 to 2013. Column 1 and column 2 list the general least square
and country time-invariance variables as random, while column 2 treats them as fixed and everything else equal. In order to decide between two specifications, a Hausman test is
performed and the fixed-effect model is recommended by the test result. In order to prevent the estimation bias results from the heteroskedasticity and the serial correlation, the robust standard error estimation base on the fixed-effect model is provided in column 3.
According to table 2, openness to trade is negatively related to unemployment rate. More specifically, a 1% increase in the openness to trade decreases unemployment level by 9%, and the result is statistically significant at 1%. Hence, globalization encourages job creation more than job destruction, which is corresponding to the findings of Felbermayr et al (2009). There is also a negative relationship between GDP and unemployment, an increase in the aggregate production level by 1% reduces the unemployment level by 9.4%, the result is statistically significant at 1% and stays in line with the Okun’s law. The per capita income and the real interest rate are shown to have non-negative impacts on the unemployment; however, the effects are not significantly differing from zero.
The import volume shows a positive relationship to the unemployment level: a 1% increase in import will lead to a 3.9% increase in unemployment, significant at 1%. This finding is uncontested and stays in line with most of the literatures. However, surprisingly we also find a positive relationship between export volume and unemployment, 1% increase in export increase unemployment by 5.1%, this result is also significant at 1%. Since I don’t have a valid
explanation for this, I perform a subgroup analysis in the next section.
(1) (2) (3)
VARIABLES Random Effect Fix Effect Robust SE
Ln(import) 3.334*** 3.926*** 3.926*** (0.731) (0.750) (1.023) Ln(GDP) -7.577*** -9.414*** -9.414*** (1.550) (1.626) (2.6) Ln(export) 4.318*** 5.091*** 5.091*** (0.824) (0.848) (1.177) Ln(openness) -7.524*** -9.029*** -9.029*** (1.545) (1.586) (2.13) Ln(income) 0.0143 0.332 0.332 (0.109) (0.206) (0.755) Real Interest rate 4.78E-03** 3.74E-03* 3.74E-03
(2.18E-03) (2.25E-03) (5.3E-03)
Constant 6.132*** 10.16*** 10.16***
(1.186) (1.597) (4.323)
Observations 408 408 408
R-square 0.0266 0.262 0.262
Number of country 12 12 12
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(Note: log unemployment rate is the dependent variable. GDP, import, export and income are measured in million U.S. dollars.)
5.2.2 Subgroup Analysis
To look deeper into the impact of international trade, a subgroup analysis is conducted by dividing counties into developed and developing nations. The feasible general least square estimators and the robust standard error estimation base on the fixed-effect model of both
subgroups are showed in table 3. In this section, we are going to look at the feasible general least square (FGLS) estimators because the results are more significant.
We can see from the table 3 that the sign of the impact of openness to trade on unemployment level is identical between developed and developing countries, but the
magnitudes are different. A 1% increase in openness to trade in all countries will decrease unemployment by 7.2% in developing countries and 3.2% in developed countries; the results are significant at 1%. Such results can be explained by the different production structure of each subgroup. For example, since most of the developed countries are capital abundant and most of the developing countries are labor abundant, an increase in the openness to trade will expand the production of capital-intensive goods in developed countries and labor-intensive goods in developing countries. Thus there will be more job creation in developing countries.
Similarly, a 1% increase of aggregate production level (GDP) will leads an 8.6% decrease in unemployment in developing countries whereas 4.9% increase in unemployment in developed countries. This can be also explained by the same economic intuition that, the total aggregate production level is increased by expanding the production of capital-intensive goods in developed countries, while developing countries expand the production of labor-intensive goods. Thus more jobs will be created in developing countries than in developed countries.
According to the table 3, a positive relationship exists between import volume and unemployment in both subgroups, which is corresponding to most of the literatures. However, both subgroups show a positive relationship between export volume and unemployment and the results are statistically significant at 1%. An increase in the export volume by 1% increases the unemployment rate by 1.9% in developed countries. This effect can be explained by Brecher (1974) who shows if a country mainly exports capital-intensive goods; an increase in the
international demand of its export will worsen its unemployment level by cutting the production of labor-intensive goods within the nation to release addition capital for the higher export demand, which conform to the situation in most of the developed countries.
The situation is more complicated in labor abundant developing countries. As can be seen from table 3, a 1% increase in the export volume increases the unemployment rate by 3.8% and statistically significant at 1%. One possible explanation is the industrial upgrading that occurs in some developing countries. After a long period of capital accumulation and technology
advancing, some developing countries are in the process of structure change, such as from agriculture, industry to service dimension. The Kuznets facts shows that the agriculture share in GDP has a secular decline, the industry share demonstrates a hump shape and the service share increase (Herrendorf et, al. 2011). Therefore, for some developing countries that are involved in such industrial dynamics, an increase in export is no longer confined to the labor-intensive goods, with the new technology; such countries start to provide more capital-intensive goods, such as India, which has one of the biggest software engineering industries in the world. Thus, according to Brecher (1974), unemployment rate will increase in a developing country if it increases its export by producing more capital-intensive goods.
The second possible explanation is the change in the endowment structure. As I mentioned before, some developing countries have gone through a long time of capital accumulation, which may increase their capital endowment. According to the Rybczynski theorem9: For any given endowment of capital and labor, when the capital endowment increases, the output will increase in the industry that is capital-intensive and will decrease in the industry that is labor-intensive. In such case, unemployment will increase as well as the export since more capital-intensive goods are produced.
9
Table 3: Subgroup Analysis for Developing Countries and Developed Countries (1) (2) VARIABLES Developing countries Developed countries
(FGLS) (Robust SE) (FGLS) (Robust SE)
Ln(import) 2.787*** 7.594 1.295** 1.292 (1.037) (4.167) (0.556) (1.425) Ln(GDP) -8.601*** -15.6 -4.938*** -7.252** (2.173) (8.084) (1.363) (2.382) Ln(export) 3.758*** 8.461 1.918*** 1.847 (1.122) (4.352) (0.624) (1.754) Ln(openness) -7.213*** -16.77 -3.193*** -3.366 (2.150) (8.493) (1.165) (3.077) Ln(income) 1.417 -1.064 1.713** 3.843** (1.041) (1.524) (0.74) (1.178) Real interest rate -1.97E-02** -1.14E-02 1.86E-02 -6.62E-04
(8.58E-03) (1.78E-02) (1.84E-03) (4.8E-03)
Constant 19.93*** 15.47 12.19*** 24.70***
(4.336) (7.26) (3.654) (4.963)
Observations 238 238 170 170
Number of country 7 7 5 5
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
(Note: ln unemployment rate is the dependent variable. GDP, import, export and income are measured in million U.S. dollars.)
6. Conclusion
This paper contributes to the exiting literature by presenting one of the few empirical estimates using panel data from 1980 to 2013, among 12 countries. The robust standard error estimation in fixed-effect model is used to analysis for all countries and the feasible general least square estimators are used to show the long-term impact of globalization for each subgroup,
represented by openness to trade, on unemployment at aggregate and sub-group level. All the results presented in this paper are statistically significant.
The paper shows that both at the aggregate and sub-group level, the openness to trade is
negatively related to the unemployment rate in the long-run, which means free trade creates jobs at the overall level. This finding stays in line with the recent research by Felbermayr et al (2011), which indicates that the opening to trade has a non-negative relationship with unemployment in the long run. Corresponding to the Okun’s law, I found that an increase in GDP will decrease the unemployment level in the long run, which can be explained by job creation resulting from a boom in the aggregate production level. Furthermore, import increases the unemployment; this is because of the job destruction due to the import competition.
The most interesting finding of the paper is the relationship between export and unemployment level. I found a positive impact of export on unemployment, which violates most of the
international economics literatures. In order to give a proper economic intuition about this issue, a sub-group analysis is conducted by dividing countries into developed and developing. The results show that in both groups, export increases the unemployment, which stays in line with the research by Brecher (1974). Regarding the issue in developing countries, I suggest two possible explanations base on the economic dynamic evidence. One is the change of structure and technology, the other is the change of capital endowment. Both will gradually shift the production dimension from labor-intensive to capital-intensive, and increase unemployment. Further research can improve on this topic by adding more countries, expanding the time period or include more control variables.
Bibliography
Bernard, A. B., Jensen, J. B., Redding, S. J., & Schott, P. K, (2007). Firms in international trade.
No. w13054. National Bureau of Economic Research.
Brecher, Richard A, (1974). "Minimum wage rates and the pure theory of international trade." The Quarterly Journal of Economics 88.1: 98-116.
Capuano, Stella, and Hans-Jörg Schmerer, (2013). "A structural estimation of the trade and unemployment nexus."
Davidson, Carl, L. Martin, and S. Matusz, (1988). Multiple free trade equilibria in a model of frictional unemployment. (No. 8716).
Davidson, C., & Matusz, S. J. (2004). International trade and labor markets: Theory, evidence,
and policy implications. WE Upjohn Institute.
Egger, H., and U. Kreickemeier (2006), “Firm Heterogeneity and the Labor Market Effects of T rade Liberalization”, mimeo, Zurich University
Felbermayr, G. J., Larch, M., & Lechthaler, W. (2009). Unemployment in an interdependent
world (No.2788). CESifo working paper.
Felbermayr, Gabriel, Julien Prat, and Hans-Jörg Schmerer, (2011) "Globalization and labor market outcomes: wage bargaining, search frictions, and firm heterogeneity." Journal of
Economic Theory 146.1: 39-73.
Felbermayr, Gabriel, Julien Prat, and Hans-Jörg Schmerer, (2011). "Trade and Unemployment: What do the data say?." European Economic Review 55.6: 741-758.
Ghose, Ajit K, (2000). "Trade liberalization, employment and global inequality."International
Labour Review 139.3: 281-305.
Herrendorf, Berthold, Richard Rogerson, and Akos Valentinyi, (2011). "Growth and Structural Transformationprepared for the Handbook of Economic Growth."
Hine, Robert, and Peter Wright, (1998). "Trade with low wage economies, employment and productivity in UK manufacturing." The Economic Journal 108.450: 1500-1510.
Janiak, Alexandre,(2006). Does Trade Liberalization Lead to Unemployment? Theory and Some
Evidence. mimeo, Universite Libre de Bruxelles.
Krugman, Paul R. International economics: Theory and policy, 8/E. Pearson Education India, (2009).
Lamo, A., Messina, J. and Wasmer, E. (2006) ‘Are Specific Skills an Obstacle to Labour Market Adjustment? Theory and an Application to the EU Enlargement,’ CEPR discussion paper 5503.
O'Rourke, Kevin H. Globalization and inequality: historical trends, (2011). No. w8339. National
Appendix
The main empirical method uses the robust standard error estimation and Feasible
General Least Square (FGLS) model to allow for heteroskedasticity and serial correlation. In this section, two test results are provided to show that heteroskedasticity and serial correlation indeed occur in the fixed-effect model.
Groupwise heteroskedasticity means that the variance of the error term varies across units; it is very likely to occur in panel-data model. I use the xttest3 Stata command to calculate a
modified Wald test for groupwise heteroskedasticity in fixed-effect regression model with the null hypothesis of homoskedasticity. The test results are showed as following:
The results show that the null-hypothesis is rejected, so the heteroskedasticity indeed occur in the residual.
Serial correlation biased the standard error in linear panel data model thus results less efficient results. I use xtserial Stata command (Wooldridge’s test) to indentify serial correlation in the panel-data model, with the null-hypothesis: there is no serial correlation in the panel-data model; test results are showed as following:
Prob>chi2 = 0.0000 chi2 (12) = 360.14
H0: sigma(i)^2 = sigma^2 for all i in fixed effect regression model
Note that the null-hypothesis is strongly rejected. Hereby we conclude the serial correlation in the idiosyncratic error term of the panel-data model.
Prob > F = 0.0000 F( 1, 11) = 134.045 H0: no first-order autocorrelation
