• No results found

The effect of trade on the income inequality in a developed country

N/A
N/A
Protected

Academic year: 2021

Share "The effect of trade on the income inequality in a developed country"

Copied!
43
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

The effect of trade on the income

inequality in a developed country

(2)

The effect of trade on the income inequality in a

developed country

University of Amsterdam

Faculty of Economics and Business Bachelor thesis economics

Supervisor: Dhr. Rutger Theulings Date: May 2014

Author: Hannah Nijsingh Student number: 10003365

(3)

Abstract

The model used in this paper investigates whether increasing trade between countries will increase income inequality in a developed country. The country representing the developed country is the Netherlands. The model used the variable: trade, the Gini-coefficient and

unemployment to investigate whether there is a change of the average spending for the income groups of the tenth till twentieth percentage of the population of the Netherlands and the average spending of the eightieth till ninetieth percent in the Netherlands. The paper starts with

describing established trade models and theory’s. Then it continues with findings about

increasing income inequality from theoriesabout the United States and Canada. The regression analysis performed in this paper shows the following results; unemployment is not significant on this subject. The Gini-coefficient shows that the income inequality becomes bigger. The gap between the low and the high group grows. Increasing trade volume will lead to an incline in spending for both groups, so this is not in line with the prediction.

(4)

Content

1. Introduction p.6

2. Theoretical framework p.8

2.1. Trade p.8

2.1.1. Ricardian model p.8

2.1.2. Specific factors model p.9

2.1.3. The Heckscher-Ohlin model p.10

2.1.4. The standard trade model p.12

2.1.5. Summary models p.13

2.2. Stolper Samuelson theorem p.13

2.2.1. Economic model p.14

2.2.2. Conclusion p.16

3. Prior empirical research p.17

3.1. United states p.17

3.1.1. Introduction p.17

3.1.2. Paper income inequality and economic growth p.17 3.1.3. Paper income inequality and economic growth: conclusion p.20 3.1.4. Paper Income inequality, economic growth, and the

distribution of income gains: evidence from the U.S.

States p.20

3.2. Canada p.21

3.2.1. International trade and wage inequality in Canada p.21

3.3. Summary p.24

4. Data and Methodology p.26

4.1. Introduction p.26

(5)

4.2.2. Model design p.26

4.3. Data p.27

4.3.1. Average expenditure p.27

4.3.2. Trade p.28

4.3.3. Inequality measure p.28

4.3.3.1. Criteria for inequality measures p.28

4.3.3.2. Gini-coefficient p.28

4.3.4.Unemployment rate p.29

4.4. Left out variables p.29

5. Results p.31

5.1. Introduction p.31

5.2. Testing the model p.36

5.3. The ordinary least squares estimators p.36

5.4 Unemployment p.37 6. Conclusion p.38 6.1. Trade p.38 6.2. Gini-coefficient p.38 6.3. Unemployment p.39 6.4. Conclusion model p.39

6.5. Limitations and recommendations for further research p.39

Appendix p.41

(6)

6

1. Introduction

The entrance of emerging markets, such as China, India, Russia and Brazil, into the global market in the past decades has led to an enormous output growth. The economic integration of these low and middle income countries on the world market could lead to gains from trade. Trade changes the distribution of income in countries. By opening up to trade the distribution of income can change between workers and the owners of capital and land. Under the assumption of perfect markets, the fundamental insight among economist is that a country as a whole unambiguously gains. Although there are gains from trade, theory also predicts there are always ‘losers’ from trade. The overall gain is bigger than the loss, which means that the winners can compensate the losers, but this is hardly ever the case. Therefore there is not necessarily always a gain from trade on a individual level (Behrens & Murata, 2012). Because some people lose from trade, the fact that there is an unambiguously gain from trade has not taken away a rising fear against globalization. This fear has showed itself in anti-free trade campaigns. Why did these campaigns emerge if increasing trade promises gains?

Among others, it is because the fear of growing wage inequality which has its

foundation in the Stolper-Samuelson theorem. This theorem states that the export competing goods benefit from increasing trade, but the sector that produces the goods that are being imported loses from increasing trade. The relative price of the goods that are being exported will increase. This is because the fact that the goods that are being imported are cheaper than if these goods were produced in the country itself, therefore the relative price of the exported goods goes up (Stolper & Samuelson, 1941, p. 62). The factor that is being used to produce the exported goods will also increase. This means that the price of the factor used for producing the imported goods will go down. The emerging countries are not producing the same goods as the high skilled workers in developed countries. The developing countries are producing goods that are comparable to the goods that the low skilled workers produce in developed countries. That is one of the reasons why there is a fear of increasing wage inequality. This fear comes from the low skilled workers in developed countries. They fear that the low-skilled workers in developed countries will lose their jobs to the relative cheap producers in developing countries (Jäkel and Smolka, 2012, p. 732). Is this really the case? Does increased trade lead to a higher wage inequality in developed countries?

` That is what will be investigated in this paper. Because these questions are for developed countries, this paper will investigate the effect of increasing trade for the

(7)

7

countries against trade than in countries that are still developing. Summarizing, this paper investigates what the effect of trade is on the income inequality in a developed country.

In section two, relevant models will be explained which try to explain the effects from trade and what the advantages and disadvantages are from opening up to trade. Section two also explains the Stolper-Samuelson theorem. Section three gives an overview of prior

empirical research from the United States and Canada and includes a summary of the findings of these papers. Section four describes the empirical model, estimation method and the data used for this research. Section five will show the results obtained from the data analysis and chapter six contains the conclusions and recommendations for further research on this matter.

(8)

8

2. Theoretical framework

As stated in the introduction, there is an unambiguously gain from trade. The overall welfare will go up by increasing globalization. The Stolper-Samuelson theorem explains what happens to the relative prices of the factors used in the imported and exported goods. The models described in this section explains what the effects of trade are under certain assumptions. The models and the Stolper-Samuelson theorem provide a good basic understanding for the existence of trade and the mechanisms behind it.

2.1. Trade

Countries trade with each other because they can benefit from trade. The fact that countries benefit from trade is because countries are not identical. If every country would be the same, the tendency to trade would completely disappear. But the fact that countries do differ gives them the opportunity to achieve economies of scale, which can give them a comparative advantage. Because countries can benefit from trade over time and countries are trading with each other for a long time now, there are multiple models that describe the causes and effects of trade. The next sub-sections will describe these models, with their assumptions and limitations. These models will help understand the mechanisms behind the results coming from the empirical model. This will help with explaining the results from the regression analysis.

2.1.1. Ricardian Model

This model has been around since 1821, introduced by David Ricardo. In this model international trade exist due to the fact that countries differ in technology differences. The Ricardian model is a really simplified model of the reality. In its simplest form it takes into account two countries, Home and Foreign. The difference in technology used in both countries can be explained by a difference in the unit labor requirement per produced good. Unit labor requirement stands for the amount of hours workers have to work to produce product X in the Foreign or Home country. This model shows that there are always gains from trade if countries relatively differ in their requirements (Ricardo, 1821, pp. 132-140).

These gains come from specialization. The countries will specialize in the good that they have a comparative advantage at. By specializing in one good they can reach a higher consumption level. If they would produce the good themselves, when the other country has an comparative advantage in producing the good there would be a possibility to reach a higher

(9)

9

consumption level. The current situation is not a pareto efficient allocation. When they do trade, the countries benefit from the comparative advantage and they can consume more goods now, in comparison to the situation where the countries did produce all the goods for their own consumption, therefore a higher utility can be reached by trade. By participating in trade, the countries can reach a pareto efficient allocation, meaning that no better situation can be reached compared to the situation they are in, at that moment. This means that it is always better to trade.

The downfall of this model is that it is actually overly simplified and it does not include any insight on the income distribution in the countries Home and Foreign, but it is a good model to understand why there is trade and it shows that if there is a comparative advantage, this will lead to gains. The main thing that can be learned from the Ricardian model is that the comparative advantage matters and not the absolute advantages (Ricardo, 1821, pp. 132-145).

2.1.2. Specific factors model

The specific factors model is much alike with the Ricardian model because it also relies on two goods and the allocation of labor. The specific factors model was developed by Paul Samuelson and Ronald Jones. In this model labor is the mobile factor, however labor is not the only factor anymore. There are also two specific factors: ‘capital and terrain’. This model contains diminishing returns. The output of a good will increase when you add more labor to the production process, but if the input of the other factor stays the same, the return will diminish for the reason that the worker has relatively less of the other factor to work with (Samuelson, 1971, pp. 369-372).

The allocation of labor depends on the price of output and the wage rate, this is because the specific factors model makes the assumption that the factors are mobile between the two goods (Samuelson, 1971, p.373). Given this, the wages for both goods will be the same because otherwise workers prefer to work in the other sections, so this determines the output. Key point is that international trade shifts the relative price of the goods that are traded. The factor in the export sector will benefit from trade. The effect on wages will be ambiguous, trade benefits the factor that is specific to the export sector of each country but hurts the factor specific to the import-competing sectors (Samuelson, 1971, pp. 381-383).

(10)

10

From this we learn there are winners and losers when trade appears. The specific factors model is useful for this paper, because it includes and explains why factors benefit or why factors lose from trade. This can be a reason or a helpful tool to explain why income inequality may arise when there is trade and that is the reason why it is included in this paper.

2.1.3. The Heckscher-Ohlin Model

The model thanks its name to the developers Bertil Ohlin and Eli Heckscher

(Deardorff, 1982, p. 683). The simplest version of this model is the one with two countries, two goods and two factors of production. The factors of production are now mobile across the countries Home and Foreign in the long run. The assumptions are that both countries have the same demand functions and have the same technology. The difference between the countries is a different endowment, a different set of recourses. This means that one country has

relatively more labor than the other country. The country with more labor is therefore the labor abundant country which implies that the other country is capital abundant. The theorem of Heckscher and Ohlin states that the country that is abundant in one factor exports the good which uses this factor intensively in production (Deardorff, 1982, p. 685). This will lead to the following: ‘if a country opens up to trade, the factors in the abundant sector will gain from trade and the factor in the scarce sector will loose from trade’. This will lead to the

expectation that developed countries, which have relatively more capital, will export capital intensive goods. This will mean that developing countries, which have relatively more labor, will export labor intensive goods. Looking at the past, this is strangely enough not always the case. For example in the Unites States in the aftermath of World War II they exported less capital-intensive goods than they imported. This is known as the Leontief paradox (Leontief, 1953, p. 335). This example shows that this model also has its shortcomings. But because this model, just like the specific factors model, explains that there are winners and losers from trade it will be a useful tool for answering the main question of this paper.

Illustration 1 shows that when there is trade, there is an eventual gain. This is an example of the Heckscher-Ohlin model. There are two factors of production, which are labor and capital and there are two goods, which are X and Y. Assumptions of this model are that the preferences are homothetic, and there are no factor intensity reversals (Marcus et al., p.137). The Rybczynski theorem states: Assuming constant relative prices, an increase in the availability of a certain factor of production will lead to an increase in the production of the good that intensively uses this factor of production and will lead to a decrease in the

(11)

11

production of the other good (Rybczynski, 1955, pp. 336-338). If Home has relatively more labor than capital, there can be said that Home is abundant of labor. From this follows that when the countries open their borders and start trading, home will export the good that uses more labor in the production process. Product X is capital intensive and product Y is labor intensive, Home will therefore export good Y.

Illustration 1: Opening up to trade (see appendix)

Before trade, consumption and production will be at point 1. The slightly curved line is the production possibility frontier. If the country opens up to trade, Home will produce more of good Y, because it has a comparative advantage in this good and will export it to Foreign which makes the relative price of good Y rise. Home can now consume every point on the budget constraint, because it is not limited by the output anymore but can import goods from Foreign. Point 1 represents the relative world price of good Y. If we compare point 1 and 2 we see that in point 2 consumption is higher and Home gains from trade. The same results show if we do the analyses for Foreign. Because the countries are trading, Home will put more of its resources into the production of Y and will trade good Y for good X with foreign. This is in line with the theorem of Rybczynski.

(12)

12

2.1.4. The standard trade model

The assumptions made within the standard trade model are again that each country produces two goods. The production possibility frontier is a smooth curve. The production depends on the relative price of one good to the other. The market value of production can be determined by isovalue lines, these lines represent sets of production for which the value of output is constant. To achieve the highest possible output countries will produces at the point where the possibility frontier is tangent with an isovalue line. The isovalue line that is tangent to the production possibility frontier is the line where the preferences of the consumers are shown. They can consume every point on this isovalue line. The point that they choose shows the preferences of the consumers. So again the highest indifference curve is chosen, the one tangent to the isovalue line.

Illustration 2: Opening up to trade, indifference curve (see appendix)

In illustration 2 the economy produces at point A and consumes at point B. So this country will export good X because it produces more than it consumes and imports good Y because the demand is higher than the production.

This model works with terms of trade which represent the price of the export good divided by the price of the import good. In this case, when the price of X increases the terms

(13)

13

of trade of this country improve. It will reach a higher indifference curve, and a higher

indifference curve means that you are better off. The statement that comes out of this model is that a rise in the terms of trade increases a country’s welfare, while a decline in the terms of trade reduces its welfare. One thing has to be kept in mind with this statement: the welfare level cannot go below the welfare level before trade (Krugman et al. 2012, pp. 140-149).

2.1.5. Summary models

All the models have their strengths and have their weaknesses, but the main thing that can be learned from the models is that they can predicted in a way what effect trade has on a country. The Ricardian model tells us that there are only gains from trade. The other models reveal that there will be winners and losers, but on the overall there is a gain due to

comparative advantages. Of course the models are stylized and some assumptions will not hold in the real world but they help us predict what effects can come from trade and which factors benefit from increasing trade and which factors loose from increasing trade. In the real world there are trade restrictions and embargos which are not included in the models above, but as said before they help us with the understanding of trade patterns.

A strong point from the Ricardian is that it explains that gains from trade come from differences in technology in countries. This is a good explanation why there is trade in the first place, thus a reason why countries engage in trade even though it could lead to an increase in income inequality. The other three models show that there are winners and losers from trade, and this leads back to the main question of this paper. It could be an explanation for the fact that there is a fear of increasing globalization and the hypotheses that it will lead to increased income inequality.

The Specific factors model, the Heckscher-Ohlin model and the standard trade model share the fact that the factor intensively used in the product that is exported will gain from trade. This can help explain why some people/factors gain from trade and why some will lose from trade. Maybe this also applies to the wages and spending and therefore support the hypothesis.

2.2. Stolper Samuelson theorem

The theorem of Stolper and Samuelson can be described by taking the Heckscher-Ohlin setting. Two countries, two goods and two factors of production. The theorem states that the factor that is scarce will lose from trade and so the factor that is abundant will gain

(14)

14

from trade. This means that protection of the import competing industries will lead to a raise in the real rewards of the scarce factor and lowers the returns (Jäkel and Smolka, 2012, p. 735). Now the question is, does opening up to trade really lead to a loss for the import competing sector, as stated in the Stolper-Samuelson theorem?

2.2.1. Economic model

Jäkel and Smolka (2012) test the Stolper-Samuelson theorem with an utility

framework. So they test if protection helps the scarce factor, or put differently, does free trade hurt the scarce sector. They measure the utility of person i, in country c, if the country moves towards free trade. Their function is linear and follows the settings of the Heckscher-Ohlin model.

For the data they used an information set from the survey of the Pew Global Attitudes Project which enabled Jäkel and Smolka to use information of 40 000 individuals worldwide. These people come from all over the world so the dataset includes individuals from labor-abundant countries and from capital labor-abundant countries. They obtained their information by asking question in the form of: ‘what do you think of growing trade and business ties between your country of resident and other countries?’ Followed by a second question: ‘does the respondents sees this as a good thing or bad thing in different grade levels?’ They created scatter plots ‘figure 1’. By looking at the scatter plots it can be concluded that the gross domestic product (GDP) is positively correlated with the amount of education received. So the higher the education level, the better the gross domestic product. They can state this per country because they see each individual as a representative of their country of residence. They also find a relationship between preferences of protectionism and implanted trade policies. Individuals who are more reserved against trade tend to have governments which are more inclined towards protectionist policies. Rich countries compared to poor countries are more trade skeptical. Remarkable is that U.S. residents are the most skeptical against trade (Jäkel and Smolka, 2012, pp. 740-742).

(15)

15

(16)

16

2.2.2. Conclusion

The results are what is expected for the developed countries from the economic model. In the United states their openness towards trade increases by 12% if you look at the lowest skill level compared to the highest skill level. So for the richer, developed countries. This effect works inverse for the emerging economies. So their openness to trade declines with the higher skill level individuals reach. They also conclude that bringing in the Heckscher-Ohlin model to prove the Stolper-Samuelson theory does add something to the subject. They prove that an individual’s preference towards trade includes abandonment or scarcity of his

production factor in the domestic economy (Jäkel and Smolka, 2012, pp. 755-757). So the fear that is introduced in the introduction is shown in these scatter plots. In section 3 will be seen whether the attitude of the citizens of the United States is justified.

(17)

17

3. Prior empirical research

This chapter contains a discussion of prior empirical research on the relationship of increasing trade on the wages of developed countries. These findings will make a framework for the hypothesis of this thesis. In section 3.2. this paper looks at research done for the United States. In section 3.3. the same is done except then it will revise research done in Canada. 3.4. will summarize the results.

3.1. United States 3.1.1 Introduction

The United States experienced in the past few decades a strong economic growth. It also experienced a growing income inequality. There has been a lot of research done, to see if these two events are connected with each other. In the next paragraphs this paper will look at the findings of these studies on the subject and what the conclusions are.

3.1.2. Paper income inequality and economic growth

The growth the United States has been experiencing in the past few decades can be ascribed to the progress in internet and information technology which created a lot of new jobs. The government followed a balanced budget which kept the interest rate low , this led to better investment opportunities (Hsjing, 2005, p. 639). Panizza (2002) used panel data on the income distribution of 48 states in the United States to investigate the relationship between inequality and growth. That is why this paper will look at the results of Panizza. Panizza has chosen panel data because there was already research done on cross-country studies but these studies found a negative effect between growth and inequality and this is not what Panizza was expecting and what the hypothesis of this paper is. Panizza uses adjusted gross income data to calculate the Gini indices. The years used are: 1940,1950, 1960, 1969 and 1980. The population is divided in quintiles. For calculating the quintiles Panizza used this split

histogram method suggested by Cowell (1995).

For calculating the Gini index he used a linear approximation of the Lorenz curve. He states that if the sample used is large enough the error will be small. In this way he justifies himself in using this sample.

(18)

18

1.

where , is state i’s annual growth rate of income per capita, , is state i’s log of income per capita, , is a variable capturing income distribution (measured using the income share of the third quintile or the Gini index), Xi is a matrix of controls, and Ri, is a matrix of regional dummies controlling for the possibility of different growth patterns in different regions of the US (South, Mid-West, and West). All the explanatory variables are measured at the beginning of the growth period. This resulted in table 1:

Table 1: Cross-sectional regression: ten and twenty-year growth episodes (see appendix)

This table shows that cross sectional estimates in comparison with quartile three are positive. But if these cross sectional estimates are compared with the Gini-coefficient they turn out to be negative. The table also shows that these coefficients are in either case not significant. With pooled OLS Panizza generated the following table:

(19)

19

Table 2: Pooled OLS: Ten and twenty-year growth episodes (see appendix)

For the ten year results it shows a significant negative effect between inequality and growth but if it is measured over a twenty year period the data are not significant. After this the paper looked at the following estimation:

2. Where all variables are the same as in the first equation except for which is the annual growth rate of income per capita from period t to period n+1, , is a state-specific intercept, and is a period-specific intercept. The inclusion of the time dummies works around the problem of multicollineairity. These dummies are highly significant. For the time span of twenty years there is again no significant correlation between changes in inequality and changes is growth. The model without the dummies does yield a significant correlation between the third quartile and the Gini-coefficient. The differences between ten and twenty year observations are probably due to differences in the short run and long run.

Panizza finds it puzzling that using the different inequality measures, quartile three and the Gini coefficient, gives different results. He thinks this is due to the fact that the third quartile lies by the median voter and the Gini index focuses more on the comprehensive inequality measure. He finds that the economic impact on inequality is lower compared to cross sectional studies (Panizza, 2005).

(20)

20

3.1.3. Paper income inequality and economic growth: conclusion

This paper did not find a positive relationship between economic growth and income inequality. It also does not find a correlation between the Gini index and the third quartile. The paper does find some evidence of a negative effect between income inequality and economic growth but these findings are not robust and small differences in method can make a big difference in the results.

3.1.4. Paper Income inequality, economic growth, and the distribution of income gains: evidence from the U.S.

Hasanov and Izraeli (2011) look at inequality channel of growth policy and assess the distribution of income gains among different income groups. In their article they use a state cost of living index (COL) because they find that prices between states are different as well as the inflation rate. They also look at nonlinearities in an inequality-growth relationship

between states, instead of between countries because they find that in cross-country studies nonlinearities are not fully taken into account.

In their paper they study the effect of economic growth on the shares of income of the poor, middle-income and the rich for the period 1970 till 2005. They put their findings in the table 3. Where I stands for the mean family income and Q stands for the share of income in

that quartile.

They found three remarking observations. The mean family income did rise over the periods 1970-1980 and 1990-2000 but the share of income decreased for all periods. For the highest income group (Q4 and Q5) they see their income only fall in the last period but not as much as the other quarters, so this suggests that the rich have been gaining a larger proportion

Table 3: Rate of change in real mean family income (I) and share of income (Q) by quintile for the U.S., 1970-2005 (see appendix)

(21)

21

of income than the other income groups. In other words, everybody did gain, but the rich gained more than the underprivileged, therefore the income inequality grew.

Overall they find that an increase schooling rates has a positive effect on the economic growth which is significant. But this growth in schooling has a negative effect on the mean income and the share of the lowest 3 quintiles. This effect becomes weaker for the third quintile (Hasanov and Izraeli, 2011, p.532).

They used the col index and added three forms of nonlinearity to the model which where Gini squared, the Gin/income interaction, and a change in Gini terms. By separating these three forms in different estimators they were all significant. They have found a fall in the income share of the poor and a rise in the income share of the rich so this suggests that the income gap is becoming wider. Furthermore a growth in education has a positive significant effect on economic growth so this again shows an increase in the higher quartile incomes. This also widens the gap and thus an increasing income inequality (Hasanov and Izraeli, 2011, p.539). This confirms the hypothesis made in this thesis that increasing trade, in this case seen as economic growth, will lead to more income inequality.

3.2. Canada

3.2.1. International trade and wage inequality in Canada

In the paper of Breau and Rigby (2009) they look at the widening wage gap and the possible connection with economic growth. They tried to move beyond the limitations such as measurements of worker skills and the availability of data on trade. They have done this by using individual data on worker characteristics and detailed business data from surveys. For the theoretical foundation they used the Heckscher-Ohnlin model as described earlier in this paper. They take into account that the Heckscher-Ohnlin model has the limitation that countries do have differences in production and factor endowments. They have found in earlier papers Lawrence and Slaughter (1993) and Baldwin and Cain (2000) that the

increasing wage gap in Canada was due to a skill-biased technological change and trade had a modest effect on the inequality (Breau & Rigby, 2009, p. 59). In other studies where they looked more at the different industries on their own they did find evidence that trade widened inequality in Canada. Breau and Rigby claim that these pitfalls in the research already carried out are due to the rough measurements of skills. They argue that they should be measured better and that they had to take a closer look at individual characteristics of the workers. There

(22)

22

is a difference in the wage rate between genders and in former studies that have been

performed on this subject they took no account for this. They have taken data from two micro datasets, which included detailed information about individual workers. They mapped the employers in the field they were working in. They found that their sample matched for more than 70% with the whole population of Canada. For the measurement of the import they classified the countries they were importing from in low, middle and high income countries with data from the world bank’s development indicator database. To compute how large the share of import is from one of the three classes they used the following relation:

3.

represents the amount of imports to i, which is Canada, from j which is one of the 3 classes. is the total amount of imports in Canada. This can then be used to calculate the import penetration rate:

4.

Shipments stands for the total value of domestic outputs of commodities in Canada. This gave them the impact from import and with this variable they set up their estimator for the wages of individual workers (w) in plant p. They used the following model for this:

5.

Where stands for the age of the worker and is the squared term of this.

is a dummy variable for gender, is a dummy variable for visible

monitory status, is a dummy variable for immigration status and for part-time or full part-time worker. is the size of the synthetic manufacturing establishment to

which the worker is associated. is the capital to labor ratio,

stands for whether the establishment is foreign owned. s an indicator of the

(23)

23

penetration rate (from j-income countries, as defined above) and and are dummy variables controlling for industry and province fixed effects (Breau & Rigby, 2009, p. 70).

To account for the difference between workers with different skill levels they

separated the workers in 5 pools. Ranging from less than 11 years of schooling to more than 18 years of schooling. They used ordinary least square estimators and to correct for

heteroskedasticity the Huber-White sandwich estimator. They also controlled for correlation between the variables (Breau & Rigby, 2009, p.71). This is probably a good thing to do for the estimators generated in this paper.

Table 4 is generated with using equation 5. A to E represent the 5 classes of educational level. As expected from former studies from Moore and Pacey, (2001) wages increase when people get older but at a decreasing rate. Males, white people and

nonimmigrant’s also receive higher wages. Workers in manufacturing plants that are larger, so with a larger output get paid more than in smaller manufacturing plants and also the ones working in more capital intensive plants earn more.

(24)

24

The most interesting results from this paper are the ones with the trade penetration variables. Trade coming from the high income countries for each educational group are not significant. This is different for the low income countries. The variables are significant with a 5% level and negative which implies they have a negative effect on wages. If trade increases by 100% from the low income countries, for the lowest education level their wage will

decline by 1.1% for the second education levels the decline is somewhat smaller. For the other education level groups the results are not significant. For the workers in educational group 4 their income rises when trade goes up with low-income countries. Because this is not what they expected, they investigated what could be the cause of this. They think that it is because the industries which rely on the imported goods from low-income countries are some of the most unionized industries (Breau & Rigby, 2009, pp. 71-80). This is interesting for this paper because as learned from Jäkel and Smolka there is a fear from developed countries towards trade. The main question of this paper is wheter trade will increase the income inequality and these results found in the paper of Breau and Rigby suggest that trade from low-income countries will lower the wage rate.

Because the trade penetration variables from the higher income countries for all

educational levels are not significant it can be concluded that this has no effect on the possible wage inequality. So the wages are not affected by trade from developed countries. For trade coming from low-income countries the lowest two groups experience a decline in their wages if trade goes up. For the other income groups this is not significant. So their hypothesis was confirmed. If trade with low-income countries increases, the lowest wages will go down. So higher import competition from low-income countries will lead to greater wage inequality (Breau & Rigby, 2009, p. 80).

3.3. Summary

The paper of Panizza did not find real evidence for a widening income gap when trade goes up, even though he used panel data instead of cross-country data. However Hazanov and Izraeli did find a positive relationship between trade and income inequality. They used data where they separated the population in income groups and even though in some periods all the groups saw their income fall, the lower groups would lose more than the higher groups which still implies that the income gap became bigger. Just like Hazanov and Izraeli, Breau and Rigby looked at different income classes but in this case for Canada. They also found that when trade goes up the income inequality becomes bigger. Trade with developed countries

(25)

25

did not have an effect on the income gap but only with developing countries. That is why this thesis will look at a high income group and the lowest ten percent incomes to distinguish between different income groups. Looking at the prior research this will be in the benefit of this investigation.

(26)

26

4. Data and methodology

This section describes the model used. It includes which data and methods are used. Section 4.1. describes the model design. Section 4.3. will describe why these variables are used and where they are coming from.

4.1.1. Estimation method

To estimate the effect of wages in changes in the other variables, this paper uses the method of ordinary least squares (OLS). The ordinary least squared method is used because it is the dominant method used in practice and therefore more comparable with prior research investigated in earlier sections of this paper.

4.2.2. Model design

To calculate the estimators this paper will be using two models, one for the

expenditures from the lowest ten percent of the population of the Netherlands and one for the eightieth till ninetieth percent of expenditures. The estimators are shown in equations 6 and 7:

6.

7.

The OLS estimator is an appropriate estimator under three assumptions. Assumption one states that the conditional distribution of the error given the variables have a mean of zero. The error term, , represents the other factors that influence the average expenditure and these have to be unrelated to trade, unemployment and the Gini-coefficient. Assumption two is a statement about how the sample is drawn. Therefore are

independently and identically distributed. Assumption three states that large outliers are unlikely. This means that observations that are far outside the usual range of the data are unlikely to appear (Stock and Watson, 2012, pp. 164-167).

In section 4.3, it is explained where the data for the variables are found and why these measures for the variables are used. This paper does not use the split histogram, as used in the paper of Panizza, because here the tenth till twentieth percentage and the eightieth till

ninetieth percentage are used and not quartiles. There is also no difference between two different time spans because first of all, the ten year time span was not significant in the paper

(27)

27

from Panizza and second the time span used in this paper is not that large to start with. In order to make the same distinction, as in the paper of Hasanov and Izraeli, between the different income groups, this paper uses two sets of income groups. Hasanov and Izraeli use three income groups, but the third group did not really add value to their research.

4.3. Data

This paper will look at the time period from 1995 till 2011, so a time span of

seventeen years. It takes into account the data for the Netherlands because there has not been many research done for this country. Because of the time span it is time series data. The data set contains data on four variables for seventeen time periods.

4.3.1. Average expenditure

The dependent variable taken for the measure of income is measured by the amount spend by the tenth till twentieth percentage of the population and by the amount spend by the eightieth till ninetieth percent. There has been chosen for the ten percent frame from the eightieth till ninetieth percentage, because by choosing this frame the richest individuals are filtered out. This is a good thing because the top percentage is not representative for the population and is considered an outlier. The assumption is made that the highest ten percent do not have normal wages. This group will include all the major business owners and that is why there has been chosen for the percentage from eighty till ninety. There has been chosen for the ten percent frame from tenth till twentieth, because by choosing this frame the poorest individuals are filtered out. The biggest part of population from the lowest ten percent will probably receive a minimum wage and allowances from the government which can cause a distortion in the data. For that reason it is better to take the average spending from the tenth till twentieth percentage, because there is higher expectation that those data are more

uniformly distributed and are a better representation of the people that earn low wage. These data are taken from Eurostat, which extracted the data from the European Union on income and living conditions (EU-SILC), which is the reference source for comparative statistics on income distribution and social inclusion. The data for 2002, 2003 and 2004 are missing. To still have some data for the missing values the expenditure off 2001 is deducted of 2005 and divided by 4. So for every missing year this value is added so it still follows the trend, this is called interpolating. This variable is taken in the logarithmic form because average

(28)

28

growth. This will make the variable linear and more stable, and will probably benefit the predicting power of the model.

4.3.2. Trade

This thesis is trying to find out what happens to the high and low wages in times of increasing trade. Therefore it is necessary to add a measurement for trade. The variable will be named trade and can also be seen as the openness towards trade of a country. This variable is a construct of two parts, one is the trade volume, which exists of the sum of the import and export of a country and two is the Gross Domestical Product (GDP). The trading volume (import plus export) is divided by the GDP. This is done because by doing this the size of a country is filtered out. This is because the number is relative now it is divided by GDP, instead of absolute. These numbers are also taken from the database of the CBS. This variable is also taken in logarithmic form because again trade is expected to grow exponentially over time and therefore by taking the natural logarithmic form of trade we make the variable more stable.

4.3.3. Inequality measure

4.3.3.1. Criteria for inequality measures

To check if the inequality measures, used in this paper, are correct this paper made use of the following criteria. These are the most commonly used criteria used for measuring inequality. First of all an inequality measure should be positive. Because when there is no inequality the index should be zero. You cannot have a negative inequality. Secondly, an inequality measure should be symmetric. This means that when you switch the incomes of two individuals in your sample the results should still be the same. Thirdly, the number should not rely on scale. This means that if you accumulate all the incomes by the same number the results are still the same. This is important because of the different currencies. Finally there is the Pigou-Dalton criteria. This criteria states that the measure has to become bigger if there is an income transfer from a poorer to a richer person and vice a versa (CBS, 2007).

4.3.3.2. Gini-coefficient

For the measure of income inequality between households, this paper uses the Gini-coefficient (Brakel-Hofmans, 2007). This is a number between 0 and 1. The more inequality

(29)

29

there is, the higher the coefficient would be. So if there is no inequality the Gini-coefficient would be zero. If the Gini-Gini-coefficient is 1, all the income goes to one household (CBS, 2007). The Gini-coefficient is equal to half of the average relative difference between the inequality:

8.

Where ∆ is equal to . is the size of the sample, is the income of person i and is the weight of person i. is the weighted average income so the

summation of times (CBS, 2007).

Looking at the criteria stated in the section about criteria for inequality measures, the Gini coefficient fulfills the standards. It is always a positive number and the order of the sample does not matter. The Gini-coefficient also becomes bigger when a rich persons gains, when at the same moment a poor person gets poorer.

4.3.4. Unemployment rate

For the unemployment rate the percentage of unemployed people is taken and divided by the labor force. In the Netherlands all people who are included in the labor force are people who work 12 or more hours per week, or accepted a job in which they will work 12, or more hours per week or declare they want to work this amount of time. The measure includes people from 15 till 65 years. Hence, the unemployed people are the ones that have no job, or at least work less than 12 hours per week, but are actively looking for a job and are available. This data comes from CBS.

4.4. Left out variables

The model set of the article from Breau and Rigby (2009) is different than the model used in this paper. They distinguish between the origin of import. Thus they look at what kind of country the import comes from. They have divided the countries up into three groups; poor, medium and rich countries. For the model in this thesis this is not taken into account because it will look at the wages when overall trade goes up and not from a specific country group. This can further be explained, that learned from the article about Canada there can be concluded that trade from developed countries to developed countries does not have any effect on the income inequality.

(30)

30

In the papers about the United States from Panizza (2002) and Hasanov and Izraeli (2011) they look at the difference in trade between states and have added this as a variable. This variable is also not taken into account because the Netherlands is relatively small

compared to the United States and therefore does not have to correct for the trade between the twelve provinces the Netherlands has.

The article about the income inequality gap from Hasanov and Izraeli (2011) put a lot of emphasis on the characteristics of individuals and businesses. In this study this is not been used. By taking part of the population which is representative for the lowest expenditure and comparing this with a representative part of the population with the highest expenditure there is already a separation between classes and people. It is commonly known that the higher earners, so the ones with higher expenditure, have a higher education level than the ones that earn the least, overlooking a few exceptions. Also there is not such a specific data set present for the Netherlands as there is for Canada.

The article about Canada also put a lot of emphasis on schooling. They divide their sample in five classes of schooling. For this article schooling is not taken into account. This can be explained by the fact that for this article the division is made between high and low expenditure. This will account for part of the division also caused by schooling because normally higher educated people earn more than people who enjoyed less education. For this reason there has been chosen to leave schooling out of the regression analysis. There is also no separation between genders because this effect is negligible.

(31)

31

5. Results

5.1. Results OLS estimations

The analysis started with a normal regression on the variables average expenditure for the tenth till twentieth percentage and average expenditure for the group from eightieth till ninetieth percentage.

The results for the tenth till twentieth percentage are shown in table 5:

Table 5: Regression analysis for the tenth till twentieth percentage

The results for the eightieth till ninetieth percentage are shown in table 6:

Table 6: Regression analysis for the eightieth till ninetieth percentage

These where the first estimators. To test for unit root the augmented Dicky-Fuller test has been performed. There has been chosen to perform a test for unit root because the model contains features of processes that evolve through time. The number of observations had to be brought down to sixteen because the variable average spending and trade suffered from unit

_cons 10.89084 .7017398 15.52 0.000 9.374819 12.40685 Unemployment 1.024917 2.655144 0.39 0.706 -4.711171 6.761006 ginicoeffic~t -2.91524 2.628238 -1.11 0.287 -8.593203 2.762723 lntrade 1.428442 .3310069 4.32 0.001 .713345 2.143539 lAveragesp~90 Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .623313149 16 .038957072 Root MSE = .12802 Adj R-squared = 0.5793 Residual .213071824 13 .01639014 R-squared = 0.6582 Model .410241325 3 .136747108 Prob > F = 0.0024 F( 3, 13) = 8.34 Source SS df MS Number of obs = 17 . regress lAveragespend80till90 lntrade ginicoefficient Unemployment

_cons 10.26914 .7050283 14.57 0.000 8.746021 11.79226 Unemployment .1469183 2.667586 0.06 0.957 -5.616051 5.909887 ginicoeffic~t -3.791547 2.640554 -1.44 0.175 -9.496117 1.913024 lntrade 1.449761 .332558 4.36 0.001 .731313 2.168209 lAveragesp~20 Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .688157069 16 .043009817 Root MSE = .12862 Adj R-squared = 0.6153 Residual .215073494 13 .016544115 R-squared = 0.6875 Model .473083574 3 .157694525 Prob > F = 0.0014 F( 3, 13) = 9.53 Source SS df MS Number of obs = 17 . regress lAveragespend10till20 lntrade ginicoefficient Unemployment

_cons 10.89084 .7017398 15.52 0.000 9.374819 12.40685 Unemployment 1.024917 2.655144 0.39 0.706 -4.711171 6.761006 ginicoeffic~t -2.91524 2.628238 -1.11 0.287 -8.593203 2.762723 lntrade 1.428442 .3310069 4.32 0.001 .713345 2.143539 lAveragesp~90 Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .623313149 16 .038957072 Root MSE = .12802 Adj R-squared = 0.5793 Residual .213071824 13 .01639014 R-squared = 0.6582 Model .410241325 3 .136747108 Prob > F = 0.0024 F( 3, 13) = 8.34 Source SS df MS Number of obs = 17 . regress lAveragespend80till90 lntrade ginicoefficient Unemployment

_cons 10.26914 .7050283 14.57 0.000 8.746021 11.79226 Unemployment .1469183 2.667586 0.06 0.957 -5.616051 5.909887 ginicoeffic~t -3.791547 2.640554 -1.44 0.175 -9.496117 1.913024 lntrade 1.449761 .332558 4.36 0.001 .731313 2.168209 lAveragesp~20 Coef. Std. Err. t P>|t| [95% Conf. Interval] Total .688157069 16 .043009817 Root MSE = .12862 Adj R-squared = 0.6153 Residual .215073494 13 .016544115 R-squared = 0.6875 Model .473083574 3 .157694525 Prob > F = 0.0014 F( 3, 13) = 9.53 Source SS df MS Number of obs = 17 . regress lAveragespend10till20 lntrade ginicoefficient Unemployment

(32)

32

root. There was still unit root if a time lag of one period was taken into account. Table 7 shows the test results for the average spending for the tenth till twentieth percentage.

Table 7: Augmented Dickey-Fuller test for the tenth till twentieth percentage

Table 8 shows the test results for the eightieth till ninetieth percentage:

Table 8: Augmented Dickey-Fuller test for the eightieth till ninetieth percentage

Table 9 shows the test results for the augmented Dickey-Fuller test for trade.

Table 9: Augmented Dickey-Fuller test for trade

To correct for this the first difference of average spending and trade is taken. The other variables, the Gini-coefficient and unemployment, did not suffer from unit root when a time lag of one period was taken into account. Therefore these variables were not corrected to the first difference.

Table 10 shows the test results for the augmented Dickey-Fuller test for the Gini-coefficient:

MacKinnon approximate p-value for Z(t) = 0.8206

Z(t) -0.795 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Augmented Dickey-Fuller test for unit root Number of obs = 15 . dfuller lAveragespend10till20, lags (1)

MacKinnon approximate p-value for Z(t) = 0.7662

Z(t) -0.964 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Dickey-Fuller test for unit root Number of obs = 16 . dfuller lAveragespend10till20, lags (0)

MacKinnon approximate p-value for Z(t) = 0.9484

Z(t) -0.110 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Augmented Dickey-Fuller test for unit root Number of obs = 15 . dfuller lAveragespend80till90, lags(1)

MacKinnon approximate p-value for Z(t) = 0.9244

Z(t) -0.308 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Dickey-Fuller test for unit root Number of obs = 16 . dfuller lAveragespend80till90, lags(0)

MacKinnon approximate p-value for Z(t) = 0.8235

Z(t) -0.785 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Augmented Dickey-Fuller test for unit root Number of obs = 15 . dfuller lntrade, lags (1)

MacKinnon approximate p-value for Z(t) = 0.7255

Z(t) -1.074 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Dickey-Fuller test for unit root Number of obs = 16 . dfuller lntrade, lags (0)

(33)

33

Table 10: Augmented Dickey-Fuller test for Gini-coefficient

Table 11 shows the test results for the augmented Dickey-Fuller test for unemployment:

Table 11 Augmented Dickey-Fuller test for unemployment

After the augmented Dickey-Fuller test the data are adjusted. Because with OLS homoskedasticity is an assumption, which means that the error does not vary according to the values of the model. If there is no homoskedasticity, the model suffers from heteroskedasticity and this is not desirable. That is why this model is tested for heteroskedasticity with the

Breusch-Pagan/ Cook-Weisberg test for heteroskedasticity. The results of the test from Breusch-Pagan/ Cook-Weisberg showed that the model suffers from heteroskedasticity and therefore has to be adjusted. To account for this the squared error is regressed to a quadratic function on the predicted value. Figure 2 shows the test results for the Breusch-Pagan/Cook-Weisberg test.

Figure 2: Breusch-Pagan/Cook Weiberg test

Other tests that have been performed are the Ramsey RESET test to see if the models suffer from misspecification. As can be seen in figure 3 and 4 the results show that the model does not reject the null hypothesis, so it can be concluded that the model does not suffer from

MacKinnon approximate p-value for Z(t) = 0.0001

Z(t) -4.668 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Augmented Dickey-Fuller test for unit root Number of obs = 15 . dfuller ginicoefficient, lags (1)

MacKinnon approximate p-value for Z(t) = 0.0356

Z(t) -2.992 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Dickey-Fuller test for unit root Number of obs = 16 . dfuller ginicoefficient, lags (0)

.

MacKinnon approximate p-value for Z(t) = 0.0001

Z(t) -4.625 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Augmented Dickey-Fuller test for unit root Number of obs = 15 . dfuller Unemployment, lags (1)

MacKinnon approximate p-value for Z(t) = 0.1373

Z(t) -2.416 -3.750 -3.000 -2.630 Statistic Value Value Value Test 1% Critical 5% Critical 10% Critical Interpolated Dickey-Fuller Dickey-Fuller test for unit root Number of obs = 16 . dfuller Unemployment, lags (0)

_cons -.4340877 .1099716 -3.95 0.002 -.6736952 -.1944802 Unemployment -.2248034 .5977771 -0.38 0.713 -1.527248 1.077641 ginicoefficient 1.802228 .5031491 3.58 0.004 .7059603 2.898496 lntradeadjustedunitroot -.1426143 .0643716 -2.22 0.047 -.282868 -.0023605 lAveragespend80till90~t Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .02123 R-squared = 0.5806 Prob > F = 0.0001 F( 3, 12) = 20.16 Linear regression Number of obs = 16 > t

. regress lAveragespend80till90adjust lntradeadjustedunitroot ginicoefficient Unemployment, robus Prob > chi2 = 0.5382

chi2(1) = 0.38

Variables: fitted values of lAveragespend80till90adjust Ho: Constant variance

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . estat hettest

(34)

34

misspecification. This test also helps explain whether non-linear combinations of the independent variable help explain the dependent variable.

Figure 3: Ramsey reset test for the tenth till twentieth percentage

Figure 4: Ramsey reset test for the eightieth till ninetieth percentage

Also the model is tested with the Breusch-Godfrey test for autocorrelation. The test result showed that the model has no serial correlation as can be seen in table 11 and 12.

Table 12: Breusch-Godfrey test for the tenth till twentieth percentage

Table 13: Breusch-Godfrey test for the eightieth till ninetieth percentage

If we look at table 14 it can also be concluded that the correlations between the different variables are not at such a level that the variables correlate too much with each other.

Prob > F = 0.2728 F(3, 9) = 1.53 Ho: model has no omitted variables

Ramsey RESET test using powers of the fitted values of lAveragespend10till20adjust . estat ovtest

Prob > F = 0.2187 F(3, 9) = 1.79 Ho: model has no omitted variables

Ramsey RESET test using powers of the fitted values of lAveragespend80till90adjust . estat ovtest

H0: no serial correlation

1 1.428 1 0.2320 lags(p) chi2 df Prob > chi2 Breusch-Godfrey LM test for autocorrelation

. estat bgodfrey

H0: no serial correlation

1 0.004 1 0.9514 lags(p) chi2 df Prob > chi2 Breusch-Godfrey LM test for autocorrelation

(35)

35

Table 14: Correlations

The corrected model, with sixteen periods now instead of seventeen leads to the estimators showed in table 15 and 16:

Independent variable is the average income of the tenth percentage till the twentieth percentage of the population of the Netherlands:

Table 15: Regression analysis for the tenth till twentieth percentage

The independent variable is the average income of the eightieth percentage till the ninetieth percentage of the population of the Netherlands:

Table 16: Regression analysis for the eightieth till ninetieth percentage Unemployment -6.0e-06 -.000061 .000234 1.4e-06 .000134

ginicoeffi~t .000105 .000226 .000194 .000141 lntradeadj~t -.000504 -.000545 .005897

lAv~90adjust .000433 .00086 lAv~20adjust .000385

l~10ti~t l~80ti~t lntrad~t ginico~t Unempl~t (obs=16)

> oefficient Unemployment, covariance

. correlate lAveragespend10till20adjust lAveragespend80till90adjust lntradeadjustedunitroot ginic Unemployment -0.0265 -0.1795 0.2630 0.0101 1.0000 ginicoeffi~t 0.4505 0.6498 0.2123 1.0000 lntradeadj~t -0.3348 -0.2419 1.0000 lAv~90adjust 0.7521 1.0000 lAv~20adjust 1.0000 l~10ti~t l~80ti~t lntrad~t ginico~t Unempl~t (obs=16)

> oefficient Unemployment

. correlate lAveragespend10till20adjust lAveragespend80till90adjust lntradeadjustedunitroot ginic

_cons -.2109423 .0971404 -2.17 0.051 -.4225931 .0007084 Unemployment .1581091 .4689101 0.34 0.742 -.8635582 1.179776 ginicoefficient .9092856 .3859048 2.36 0.036 .0684713 1.7501 lntradeadjustedunitroot -.1216055 .0489396 -2.48 0.029 -.2282358 -.0149752 lAveragespend10till20~t Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .01691 R-squared = 0.4050 Prob > F = 0.0128 F( 3, 12) = 5.53 Linear regression Number of obs = 16 > t

. regress lAveragespend10till20adjust lntradeadjustedunitroot ginicoefficient Unemployment, robus Prob > chi2 = 0.7773

chi2(1) = 0.08

Variables: fitted values of lAveragespend10till20adjust Ho: Constant variance

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . estat hettest _cons -.4340877 .1099716 -3.95 0.002 -.6736952 -.1944802 Unemployment -.2248034 .5977771 -0.38 0.713 -1.527248 1.077641 ginicoefficient 1.802228 .5031491 3.58 0.004 .7059603 2.898496 lntradeadjustedunitroot -.1426143 .0643716 -2.22 0.047 -.282868 -.0023605 lAveragespend80till90~t Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust Root MSE = .02123 R-squared = 0.5806 Prob > F = 0.0001 F( 3, 12) = 20.16 Linear regression Number of obs = 16 > t

. regress lAveragespend80till90adjust lntradeadjustedunitroot ginicoefficient Unemployment, robus Prob > chi2 = 0.5382

chi2(1) = 0.38

Variables: fitted values of lAveragespend80till90adjust Ho: Constant variance

Breusch-Pagan / Cook-Weisberg test for heteroskedasticity . estat hettest

(36)

36

So to summarize the estimators:

Estimators for the average spending of the tenth till twentieth percentage

Standard error

Trade -0.122 0.049

Gini-coeficient 0.909 0.386

Unemployment 0.158 0.469

Estimators for the average spending of the eightieth till ninetieth percentage

Standard error

Trade -0.143 0.064

Gini-coeficient 1.802 0.503

Unemployment -0.225 0.598

Table 17: summarize estimators 5.2. Analysis results tenth till twentieth percentage

First we look at the regression analysis for the average spending of the tenth till

twentieth percent of the population of the Netherlands. The estimators show, that if we look at them with a significance level of five percent, the estimator of the variables trade and the Gini-coefficient are significant. The estimator for unemployment is not significant, not even when a significance level of ten percent is taken. This means that no conclusions can be made related to the effect of unemployment on the average spending of the population of the

Netherlands. The estimator of the Gini-coefficient is a positive number. This is not what was expected. This is because when the Gini-coefficient now becomes higher, the average

spending of the tenth till twentieth percent also will be higher. The hypotheses was that if inequality becomes bigger, Gini-coefficient is a measure of inequality, the income gap will become wider. But now the spending will go up with a higher Gini-coefficient. Trade is a negative number and this is in agreement with the hypothesis. The larger the trade volume becomes the less the lower spenders will spend. This shows that if trade goes up the spending of the lower incomes will go down.

5.3. Analysis eightieth till ninetieth percentage

Similar to the regression analysis for the average spending of the tenth till twentieth percentage, the estimators of the variables of trade and the Gini-coefficient are significant with a significance level of five percent. Again the estimator for unemployment is not

(37)

37

significant so no conclusion can be drawn on the effect of unemployment on the average spending of the eightieth and ninetieth percentage. The estimator of the Gini-coefficient is a positive number and this is what was expected for the higher incomes. With a higher coefficient there is more inequality. This is in agreement with the hypotheses and the Gini-coefficient for the eightieth till ninetieth percentage is significantly higher than the one for the tenth till twentieth so this shows that there is an effect on the inequality, because it widens the gap between the high and low spending groups. The estimator for trade is again negative. For the higher spending group this is not what was expected. We would have expected a positive number because this would have widened the gap between the low and high spenders. In this case the estimator of trade for the eightieth till ninetieth is even lower than for the low group.

5.4. Unemployment

The estimator for unemployment is in both cases not significant. The p-value for the tenth till twentieth percentage is 74.2% and the p-value for the eightieth till ninetieth

percentage is 71.3%. These numbers are really high and this means that the estimator for the variable unemployment does not have a predicting power on the hypotheses and because of this, we cannot take this estimator into account in this model. It is possible that one of the reasons that unemployment is not significant is because of the size of the dataset. In the OLS estimation we use data for sixteen periods and to predict a good model this is not a large number. Another reason can be that unemployment does not have an effect on the income inequality, the other studies also did not put a lot of emphasis on unemployment.

(38)

38

6. Conclusion

6.1. Trade

When looking at the estimator for trade for the average spending of the tenth till twentieth percentage for the population of the Netherlands, which is -0.122, this is in line with the conclusion that can be made out of the models explaining trade. The Specific factors model, the Heckscher-Ohlin model and the standard trade model all say that there are winners and losers from trade. The number indicates that the higher the trade volume will be, the lower the average spending will be. Trade volume has consequently a negative impact on the spending. Because the low spending group is considered as the group with the low income this number shows what was predicted in the models mentioned before. This was also found in the article from Breau and Rigby about the international trade and wage inequality in Canada (2009). The results from this paper showed that if trade increases from low income countries the wages of the people with the lowest education level will decline in Canada. Because people with the lowest educational level represent the groups with the lowest spending these results are in line with the results from the OLS regression.

On the other hand, the estimator for trade for the eightieth till ninetieth percentage is also a negative number and an even larger negative than for the lower group, -1.43. If we look at the theoretical framework this number was expected to be a positive one. The models predict that when trade goes up the higher income groups gain. The number -1.43 tells that when trade volume goes up, the average spending of the eightieth till ninetieth percentage will go down. The results and the theoretical framework are therefore not in agreement. This result also showed in the paper from Breau and Rigby but in a different way. In this paper the number is significant but not that what was expected and in the paper of Breau and Rigby the number for the higher income groups were not significant.

6.2. Gini-coefficient

The paper for the United States from Panizza did not find significant effects with respect to the Gini-coefficient (2002). The article of Hazanov and Izraeli did find that the Gini-Coefficient had a significant effect on the income inequality in the United States (2011). The numbers found in this paper are both positive numbers, something that was not expected for the lower group, but the bigger the number for the Gini-coefficient becomes for the higher group is significantly higher. Therefore we can still conclude that if the Gini-coefficient

Referenties

GERELATEERDE DOCUMENTEN

 Natalia Vladimirovna Chevtchik, the Netherlands, 2017 ISBN: 978-90-365-4384-2 DOI: 10.3990/1.9789036543842 Printed by Gildeprint, Enschede, the Netherlands, Cover design by

The assembly of this protein coat in a polyhedral lattice on the cytosolic face of the plasma membrane requires the interplay between clathrin, the major component of the coat, and

Using in-situ electron microscopy, we observe and quantify how gold and silver nanocrystals nucleate from a supersaturated aqueous gold and silver solution in

It was also detected that within the current authorisation processes there is the issue of lack of attention in the authorisation requirements to certain activities that have

Surface roughness brings stress concentration point near the contact surface, possibly augmenting material anisotropic effects and are therefore detrimental for rolling bearing

At the skin surface, a higher fluorescence intensity was observed after 1 h at the test regions treated with massage (38.43–64.81 AU) and acoustic pressure waves (mean 47.51–72.40

Inclusion criteria: (1) all article written in English lan- guage; (2) interventional studies including RCTs and ex- perimental studies, which assessed the effects of

Daarbij zijn elf hypotheses getoetst, waarna we kun- nen concluderen dat het interne sociale netwerk via drie factoren een significante positieve in- vloed heeft gehad op