Winners and Losers from trade: Does redistribution actually happen?

(1)

Winners and Losers from trade

:

Does redistribution actually happen?

Bor (J.S.) Kemkes 11036974

Kees Haasnoot

June 25, 2018

Abstract

Although free trade is beneficial to a country as a whole it can have distributional effects within a country. This can than lead to the creation of winners and losers. Compensation of losers of trade can be a solution for this problem. Countries with open economies tend to spend more on social security than more closed economies. This indicates that losers are at least partly compensated for their losses. Also this research indicates that the correlation between openness and government size is most likely due to this compensation channel, in contrast to previous research.

(2)

Statement of Originality

This document is written by Bor Kemkes who declares to take full responsibility for the content of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the content.

(3)

Acknowledgements

It is a pleasure to thank those who made this thesis possible. First and foremost I would like to thank my supervisor Kees Haasnoot for his great feedback and help. The feedback sessions in which we discussed the essentials of research form the foundation of this thesis. I also want to thank my friends and family whom supported me during my thesis and my entire bachelor.

(4)

1 Introduction

Free trade is beneficial to a country as a whole, so this means that the cake is increasing due to trade. This is the conclusion which is reached by all standard economic models. Although this is a widely excepted conclusion, the specific factor and the Heckscher-Ohlin model also emphasize that free trade can gener-ate ‘winners and losers’ within countries (Krugman et al., 2015, p.99 & p.128). This could explain why so many people are against free trade. To resolve this problem, some economist make a case to redistribute financial resources from the losers to the winners (f.e. Davidson, Matusz & Nelson, 2007). This seems reasonable, but there’s little research on the question whether this actually hap-pens.

The world today is highly globalized. This means that countries increasingly engage in trade. Recently there has been substantial skepticism about more liberalization of free trade. Public support about free trade agreements such as TTIP have plummeted. In two years, support for TTIP in the US was down from 53% to 18%. Similar trends are seen in public opinion rates on free trade in Germany (Reuters, 2016). How can this be explained since standard economic theory tells us that free trade should be beneficial to all countries?

This question is examined in this research. If redistribution is taking place or not is needed to be researched since this could explain the negative attitudes towards free trade. If losers don’t get compensated, their positions against free trade would be understandable. On the other hand if redistribution is indeed taking place than this mechanism cannot explain their negative opinions and an alternative explanation is needed. This research will thus focus on the following question.

What is the relationship between the level of exposure of the economy to the world economy and the level of redistribution?

In this research this issue is examined by looking at the relationship between the openness of an economy and the amount of redistribution that is taking place. When a economy gets more open, the negative effects of trade will get more pronounced as well. We would therefore expect more open countries to redistribute more to compensate the higher proportion of losers of trade. This should be visible when examining the size of social expenditures in relation to the openness of a economy. The goal of this research is to examine this relation in detail.

Research on this particular issue should be a priority because free trade agendas can only be realized with the sufficient level of political support. When the negative impact of trade on specific groups are not addressed, support for free trade will likely decrease. This would then be unfavourable to economic growth. Also this will shed new light on economic theory of trade. This research will show that more emphasis should be on the distributional effects of free trade. This is true for both the academic as well as the policy-making spheres.

(6)

2 Literature review

2.1 Winners and losers

As was stated in the introduction international free trade can generate both winners and losers. If we consider the period from 1988 to 2008 the overall picture is actually quite positive. Research by Milanovich (2013) shows us that almost all income percentiles have seen a rise in their real income. But he also finds there are big differences in how much they won. The big ‘winners’ from the recent period of globalization are found in the top 1% of the world income distribution and in the middle class of the emerging markets. The top 1% richest people saw their income rise with 60%. Although this is a staggering percentage actually the emerging market middle class did even better with an increase of income by 70-80%. The biggest losers or what Milanovich calls ‘non-winners’ whom income increase was close to zero where between the 75th and the 90th percentile (Milanovich, 2013, p202). Those are the upper middle class of the world. This group consist of rich people in former communist countries and Latin America as well as the middle class of the rich western countries (Milanovich, 2013). In the United States this trend seems to be even more extreme. Real income for the low skilled workers in the US has fallen by 18% from 1980 to 1994 (Kapstein, 2000). The question is whether this can be explained by trade.

One of the most prominent economic models of free trade provides this causal link. The Heckser-Ohlin-Stolper model predicts that countries tend to export the good that uses the factor that is relatively abundant in the country. If country A has a relative abundance in low skilled labour it will export the low skilled labour intensive good, and import the high skilled labour intensive good. For western countries this would imply importing low skilled labour intensive goods because they have relatively less low skilled labour then non-western countries. The Stolper-Samuelson theorem then predicts factor prize equalization. This means that when western countries engage in trade the wages of low skilled workers will decrease. The rationale behind this is the following; the western countries can import the low skilled labour goods more cheaply from abroad. Then the local producers have to lower their prices, leading to a decrease in wages for low skilled workers to compensate this (Kapstein, 2000). When wages are sticky this would also imply loss of jobs.

An extended version of the HOS-model by Haskel, Lawrence, Leamer and Slaughter (2012) provides us with a theoretical framework that fits the recent developments even better. This model is developed for the US in mind. The US is an interesting case because of the divergence of income equality from 1991 to 2010. The top 1% of the income distribution have seen their incomes rise sharply. Their incomes rose from about 500.000 in 1991 to 1.000.000 in 2007, just before the crisis and stayed high even during the crisis. The picture is a lot different for all other groups classified by their educational level. From the 2000s they all saw their real incomes fall. This includes highly educated workers with bachelor and advanced degrees (Haskel et al., 2012). According

(7)

to the standard HOS-model we would not expect that highly skilled workers would see a wage decrease in developed nations. Since developed nations are assumed to be relatively abundant in high skilled labour. We would expect to see an increase in wages for high skilled workers.

Figure 1: HOS model (Haskel et al., 2012)

The extended version can explain this trend better. In this model the basic assumption is that labour is heterogeneous. This means that high skilled labour is more productive in the capital intensive sector, but not in the labour intensive sector. Workers vary in the amount of ’talent’ (Haskel et al., 2012).

Consider figure (1) above. On the vertical axis we find capital K and on the horizontal axis labour L. There are three isoquants, A, B and C, in the capital intensive sector and there’s one isoquant for the labour intensive sector. This means that in the labour intensive sector all workers have the same productivity. In the capital intensive sector we divide the workers in three groups, A, B and C. High talent workers are more productive when using capital. The unit cost lines follows the standard expression 1 = wL + rK. Capital and labour are assumed to be mobile across sectors. Profits are assumed to be zero. Group A is the most talented group. They are more productive in the capital intensive sector. That’s why their wage wa is higher. Talent B workers are seen as the marginal talent workers. They are indifferent between working in the capital or in the labour intensive sector since there productivity is the same in both sectors and workers are paid their marginal product. Talent B worker earn wb. The last group, talent C workers are only employed in the labour intensive sector because they are not productive enough in the capital intensive sector. Although type B and C workers work in different sectors, they both earn the same wages, which are determined by the tangency of the unit cost lines and the isoquants of each group respectively wb= wc (Haskel et al., 2012).

(8)

The first result of this model is that because labour is heterogeneous in talent levels type A workers earn more than type B workers even though they work in the same sector. Even in autarky there would be inequality in income (Haskel et al., 2012).

Figure 2: Extended HOS model (Haskel et al., 2012)

What happens if we allow for trade? Consider figure (2) above. When a western capital abundant country opens up to trade the relative price of the capital intensive good will rise. This will then lead to an expansion of this sector and eventually will result in a rise in the price of capital, r. Since r rises, wages in the labour intensive sector should drop to remain profitable. Wages of type C workers thus drop. Type A workers will see an increase in their income. The rise in r is completely offset by a higher price of the capital intensive good, resulting in a higher wage. This process in known as Jones amplification. With this in mind we can also see why type B workers that work in the K-sector do not see an wage increase as well. Since type B workers are not talented enough the price increase of the capital intensive good cannot offset the rise of the price of capital. Therefore their wages also decrease and match the wages of type C workers (Haskel et al., 2012).

So what do we learn from this. The first conclusion is that this mechanism is a potential source for widening inequality, since only high skilled, high talented workers profit from trade. Also it’s important that this model emphasizes that not only the sector of employment matters but also differences in workers within sectors matter. The main conclusion shows how big the distributional effects can be. Only high skilled, high talented workers win, while most people see their incomes stagnate (Haskel et al., 2012). This is something we see happen in the case of the United States, though this trend is less pronounced in the European countries.

(9)

It is difficult to find this causal explanation in the empirics. The main prob-lem is in disentangling the effects of technology and trade effects1. Economists agree that wages in the manufacturing industries in the US have dropped in recent years but they do not agree about the impact trade has had on factor prizes. According to Kletzer NAFTA generated 24-27% of lost jobs in the manu-facturing industry in the US between 1993 to 1999 that can be related to import competition (Kletzer, 2004, p729).

2.2 Redistribution

As we have seen in the section above, trade can have distributional effects on groups of workers within countries. To account for this a case for redistribution can be made.

There are two main economic normative theories of free trade that give very different emphasis to redistribution. The utilitarian approach tells us we should avoid ‘distributive complications’. The focus is on the efficiency gains by free trade on the whole economy. Too much emphasis on distribution would make the analysis to complicated and are also assumed to be less relevant than efficiency gains. In contrast, there are scholars that do give great importance to the distribution of welfare due to free trade. This falls in line with the liberal thought that people make choices with their own interest in mind. It also follows the social welfare theory by Pareto. Social welfare theory by Pareto tells us that only policy that makes at least one better off and nobody worse is a Pareto welfare improving policy. Implication of a free trade policy would be Pareto inefficient if it generates losers. The utilitarian approach can be satisfying when the losers of international trade are a very small portion of the population. Ass this group grows it gets more relevant from an economic as well as a political point of view. Policy makers should be interested in their electorates and thus should be concerned with individual effects of their policy (Kapstein, 2000).

In line with this thought we come to the following argument. If we want to enjoy the efficiency gains of free trade while addressing the individual effects of free trade we can redistribute financial funds from losers to winners. This way every one can gain from free trade. Redistribution as a form of compensation can be divided in two categories. If a worker gets displaced due to trade he will first get unemployment compensation. This is a temporary measure to level out the immediate effects of job loss. After this period people can sometimes make a claim on ‘active labour market programs’ (ALMP’s). These are programs which focus on workers who are displaced and try to get people to a new job.

The effects of the latter are discussed by Card, Kluve and Weber (2010). They find that short term effects are mostly small but midterm effects are generally positive. The effect also varies between programs. Subsidized jobs seem to perform worse than other ALMP’s. The problem of recent studies is that they focus only on the benefits of ALMP’s and do not address the cost side.

(10)

An important question is whether compensation schemes work for the losers and if they are welfare or Pareto improving. Also it is important to find which programs are most effective. Davidson and Matusz (2006) try to address this in their paper. They try to make a model that can identify the most efficient program for different groups. When tariffs are lifted and free trade is allowed for this implies lower wages in the importing sector, which we have extensively addressed in the previous section. There are two rules which determine the quality and effectiveness of a compensation program. The program should be targeted on a specific group. Within this group the program should be most effective for the average worker of this group. If the compensation program satisfies these two rules, it will be social welfare maximizing2_.

In this model there is an important distinction made. Either people stay in the importing sector or they move to the exporting sector. This distinction between movers and stayers is important for the effectiveness of the policy. The stayers should be compensated using a targeted employment subsidy. For the movers a targeted wage subsidy is more effective (Davidson & Matusz, 2006). This will be made clear in the following section.

Figure 3: Redistribution Model (Davidson & Matusz, 2006)

Consider figure (3). This model assumes full employment that is divided by two sectors. The model has some similarities with the extended HOS-model but can better describe how workers can be compensated. Subscript 1 indicates the

2_{Still some would be worse off since we target the average worker, therefore allowing for}

(11)

low tech sector and subscript 2 indicates the high tech sector, which correspond respectively to the labour intensive and capital intensive sectors in the HOS-model. Workers differ in terms of ability a ∈ [0, 1] and are paid there marginal products w(a). The assumption is made that ability matters more in the high tech (knowledge intensive) sector. This is similar to the HOS-model, though ability is now a continuous variable in contrast to the three fixed talent levels before. Superscript TD indicates the wages under a tariff regime, this is the initial situation. FT indicates the wage structures under a free trade regime. We start analyzing the model under a tariff regime. The intersection of wT D

1 (a) and wT D

2 (a) indicates a critical cut off. The worker with ability aT D is the marginal worker. He or she is indifferent between choosing sector one or two. All workers with a higher ability level than the marginal worker will choose to work in the high tech sector whereas workers with a ability level lower than the marginal worker are better of in the low tech sector (Davidson & Matusz, 2006). Now consider if we allow for free trade. We assume that this is a western country, thus the country imports low tech goods. When the tariffs are relieved this will have a downward pressure on the low tech wages. The high tech wage function will shift upwards because of free trade, just as was discussed in the HOS-model. Because the wage functions have shifted a new cutoff level is now at aF T_{. Here we start to see why people will change sectors. All workers with} aF T _{< a < a}T D _{now change from the low tech sector to the high tech sector.} This group can however be divided in two groups. All workers with a ability higher than_ea will see an increase in their real wages due to free trade, while workers with a ability level below _ea are worse off in comparison to the tariff regime. A third group, the so called stayers, are the workers with a a < aF T will also be worse off (Davidson & Matusz, 2006).

Now that we have identified the ’losers’ from trade we can start to identify which worker to target. First consider the moving group. We want to com-pensate all movers, that is workers with aF T _{< a < a}T D_{, for their losses. If} the policy focuses on the representative worker, the average mover with â, all movers with a < â are under-compensated while a > â are over-compensated (Davidson & Matusz, 2006).

(12)

Figure 4: Redistribution model (Davidson & Matusz, 2006)

With the target worker identified we can start to see which compensation scheme is the most efficient. Consider figure (4). There are two types of compensation considered. Firstly, workers can be compensated using a employment subsidy, (η). This employment subsidy is independent of the wage rate. The second possibility is compensating workers using a wage subsidy, (w). This increases in a workers ability. Using these subsidies distorts the market and creates inef-ficiency’s. We want only the workers with a > aF T _{to be incentivised to move.} When the subsidies are implemented this is not the case for both redistribution schemes. But we can clearly see that aw is closer to the desired outcome than aη. Therefore for the movers a wage subsidy is the best option. The same re-versed logic can be applied for stayers, for this group a employment subsidy is most efficient (Davidson & Matusz, 2006).

The flaw off this model is that it does not take into account adjustment costs. These can be considerably high. Also if we allow for these extra costs people can get trapped in the wrong sector. Use of active labour market programs can be a good policy to battle this problem. Therefore we would expect ALMP’s and more conventional unemployment programs to be correlated.

2.3 Government size and openness

In the past most research in this field has been done on the correlation between government size and openness. Although this seems similar to my research it is

(13)

not exactly the same. Redistribution programs are only a small part of the entire government expenditures. Research in this field is mostly done by international (macro) economists and political economists. Economist use a more macro approach while political scientist use a more micro bottom up approach. These two approaches can help to create a more in depth picture of the effects of trade on the size of redistribution programs.

One of the most prominent research on this topic is done by Rodrik (1998). In this paper he shows the existence of a positive correlation between the size of the government and the countries openness to trade. The argument for this correlation is that countries that are more exposed to trade have to cope with the risk involved in this. That’s why the size of governments of open countries are bigger than that of less open countries. Rodrik develops a channel through which this correlation comes about. The risk of openness can be divided in two separate sources of risk. The terms of trade risk comes from fluctuations in the import and export prices. The relevance of this risk type is confirmed by Mendoza (1997). Higher uncertainty about the terms of trade reduces social welfare. They find the effects of this kind of risk are higher than we would expected by conventional business cycle effects. This shows the importance of government intervention to reduce this risk. The second type of risk comes from the risk involved in being to reliant on one export product. Rodrik uses an index of the product concentration of exports as a measure of reliance on the amount of export products. Countries that rely strongly on a very small set of commodities have more exposure to external risks of trade (Rodrik, 1998). This relation is also confirmed by empirical literature. However, this relation is not as straight forward that Rodrik puts it. There seems to be a cut-off value. When countries are diversified below this threshold, trade is a source of risk as Rodrik predicts. But if the product concentration is above this threshold trade is actually reduces risk (Haddad, Lim, Pancaro & Saborowski, 2018).

The distinction he makes between developed and undeveloped countries is especially relevant. More advanced countries have an ability to protect their citizens from these external risks through social insurances. Rodrik finds that in developed countries that social security spending is highly correlated with openness. In less developed countries this correlation is not so strong. In this case the relationship between government size and openness works through gov-ernment consumption (Rodrik, 1998). An alternative explanation would be that undeveloped countries do not feel the negative effects of trade in the way it is felt in the developed world. In the undeveloped world labor is abundant so we would expect wages to increase, redistribution is thus less relevant for these countries. The focus in this paper is mainly on the developed world so we would expect a more direct positive relationship between social security and openness. Rodrik also addresses the problem of political support which was put forward in the introduction. ”(. . . ) scaling governments down without paying attention to the economic insecurities generated by globalization may actually harm the prospects of maintaining free trade” (Rodrik, 1998, p.1029). He thus gives great importance to the government to mitigate the risk exposure of trade to its citizens.

(14)

The argument of Rodrik is challenged by alternative explanations why open countries have big governments. Alesina and Wacziarg (1997) argue that this correlation is mainly due to a country size effect. Larger countries can afford to have a smaller government and they are also less open to trade than smaller countries. This is quite surprising since Rodrik (1998) checks for this explana-tion and finds no significance when including a country size variable. So the academic discussion remains indecisive on this particular point. In line with the hypothesis of my research they do find a more direct relationship between openness and government financial transfers (Alsesina & Wacziarg, 1997).

Although their conclusions differ, the research by Alesina and Wacziarg (1997) and Rodrik (1998) are similar in their research structure. They share a macro approach to the problem. Recently this has had some critique. The argument made is that the relation between the size of the government and the openness of a country is actually linked by a micro causal link. The compen-sation hypothesis works as follows. Losers from international trade want to be compensated. These people then tend to vote for parties that want to increase the welfare state. This then leads to a bigger government (Walter, 2010). So the more open the country is the more people demand compensation, the bigger the government gets. Walter (2010) has examined this empirically and finds in the case of Switzerland that this argument holds. Recently there has been more emphasis for the micro foundations of macro phenomena so this research is a helpful to take in account.

2.4 Integration of literature

The literature above learns us a couple of things. Firstly, free trade can generate winners and losers. An extension of the HOS-model can also explain why only a small proportion of the population of developed countries benefit. Since this is the case redistribution gets more relevant. Redistribution can be done in several ways. The way it’s done is crucial for the effectiveness of the policy. Also when a country gets more open, more redistribution is needed to compensate everyone. This is where the correlation between redistribution and openness comes in to play. Previous research shows that government size and government expenditures are indeed correlated as we would expect.

(15)

3 Hypothesis

We expect countries that are more exposed to the world economy to be more likely to have a big government size. This can be because of several reasons. Firstly, the more open a economy is, the more it is exposed to external shocks. To level this out countries that are more open would also have bigger social safety nets, which we will refer to as the risk channel. Secondly, displacement of workers is more likely in open economies because of the factor prize equalization effects of free trade. To compensate this, higher social expenditures would be expected, which we call the compensation channel. Thirdly, according to the compensation hypothesis, losers are likely to ask for compensation from the government.

Hypothesis 1 (H1): Countries that are more exposed to the world economy have bigger social security programs.

A second more precise measure are the active labour market programmes that target displaced workers more directly. Therefore we would expect this measure to be even more correlated with openness.

Hypothesis 2 (H2): Countries that are more exposed to the world economy spend more financial funds on active labour market programs.

Also we expect that the correlation between government size and openness is mainly due to redistribution rather than to level out external shocks.

Hypothesis 3 (H3): Correlation between government size is mainly due to the compensation channel

4 Research method

4.1 Model

In this paper the relation between the effect of the exposure to the world econ-omy on the size of social expenditures is measured using four different regres-sions. To make sure we get a better fit we will use some extra control variables, following the method of Rodrik (1998). For all regressions we use panel data for a variety of countries over a period of some decades. These relations will be examined using the following regression models. Where subscript i indicates a country and subscript t indicates a year.

The first regression is the most broad regression and is similar to the re-gression done by Rodrik (1998). The dependent variable is the Government Size (GOVSize). The measure for this particular variable is the government expenditures as a proportion of GDP. This measure is retrieved from the World

(16)

Bank database. This is a very rough measure because government expenditures include all expenditures by the government. This can be a problem because we would not expect healthcare expenditures to be directly related to openness.

GOV SIZEi,t= αi,t+ β1Opennesi,t+ β2GDP P erCapitai,t+ β3Depi,t + β4U RBi,t+ εi,t (1) The second regression focuses on a smaller aggregate. To examine the direct effect of openness on redistribution we regress a measure for fiscal redistribu-tion on openness. The dependent variable here is the Fiscal Redistriburedistribu-tion From Transfers Working Population (FRTransfersWP) as a proportion of GDP. This measure is retrieved from the Leiden Redistribution Database (Caminada & Wang, 2017). We specifically use the redistribution for transfers and for the working population because this is how and where we would expect the redis-tribution to take place. These data are available for 47 LIS countries. This paper focuses on the effects of free trade in developed countries therefore some countries are not included.

F RT ransf erW Pi,t = αi,t+ β1Opennesi,t+ β2GDP P erCapitai,t+ β3Depi,t + β4U RBi,t+ εi,t (2) The third regression is done to measure the effect of openness on social ex-penditures as an aggregate. The dependent variable we use is the aggregate social expenditures (AGGSE) as an proportion of GDP. This measure is re-trieved from the OECD Social Expenditure data set (Adema, Fron & Ladaique, 2011). The problem of this measure is that this is an aggregate which also in-cludes expenditures on social security expenditures on elderly people which we would not expect to have a direct relationship with openness.

AGGSEi,t = αi,t+ β1Opennesi,t+ β2GDP P erCapitai,t+ β3Depi,t+ β4U RBi,t+ εi,t (3)

To address this the fourth and last regression is done on an even more specific variable. As was argued in the literature review and the hypothesis we would expect active labour market programs to be used more in countries with a high level of openness. ALMP’s can be an effective tool to help people from one job to the other and thus are now widely applied to combat the effects of free trade. Therefore we use the size of ALMP’s as an proportion of GDP. These data are also retrieved from the OECD Social Expenditure data-set (Adema, et al., 2011). As the name of the data suggests all OECD countries are used here. This is a nice feature because this includes mostly developed nations.

(17)

ALM P si,t= αi,t+ β1Opennesi,t+ β2GDP P erCapitai,t+ β3Depi,t+ β4U RBi,t+ i,t (4) Using these different regressions allows us to compare the results and the fit of the model. This can help us to understand better if and how the mechanism found by Rodrik works. We should be able to separate one mechanism from the other better. How the regressions fit into one model can be seen in figure (5). The blue arrows show the direct relationships proposed above in regression form. The yellow arrows indicate a possible relation between the different regressions. ALMP’s are a part of AGGSE. It could also be that AGGSE are a part of government expenditures. However, in the particular measure we use it is not entirely clear if spending on social security is included in the measure (therefore this relation is indicated with a question mark)3_{. This indicates that if for} example ALMP’s are correlated with openness, AGGSE could be correlated with openness as well since a part of AGGSE are expenditures on ALMP’s.

Figure 5: Model

4.2 Independent variables

The main independent variable is a measure of openness. To measure this variable we use the most standard method which is also used by Rodrik (1998). Openness is measured by taking the sum of the exports and imports as an proportion of GDP using data from the World Bank data set.

Inspired by the research of Rodrik several control variables are used. GDP per capita, urbanization ratio and dependency ratio are added to the regression

3_{Long Definition: General government final consumption expenditure (formerly general}

government consumption) includes all government current expenditures for purchases of goods and services (including compensation of employees).

(18)

because they are likely to be correlated with the dependent variables. This can then improve the fit of the model.

4.3 Checks of robustness

Since we want to examine the causal mechanism between openness and social security expenditure we should do additional checks of robustness. Firstly, a check for outliers is executed and outliers will be left out, this is the case for countries with openness higher than 200. Secondly, there is a chance for omitted variables bias. For example the correlation between openness and government size could be a spurious one. Since we have panel data a fixed effect regression is also an option. This kind of panel data regression controls for omitted variables (Stock & Watson, 2015). Thirdly we have to check if difference in approach leads to different outcomes. Therefore we will also use the same method as Rodrik (1998) to make sure that the approach does not influence the results.

We also conduct some tests to see if there are any econometric problems where we have to control for. We could expect our data to be heteroskedastic and auto-correlated. This is all to make sure our estimators are unbiased and remain significant when correcting for these possibilities.

With these checks done we can be fairly sure about the robustness of our results. Though the main question remains, how can we distinguish our channel from that of Rodrik. A big part of this can be explained by comparing the models. However we should do additional econometric analysis to allow for a more thorough investigation.

5 Data Analysis

5.1 Comparing the models

We start the analysis by running the regressions proposed in the model sec-tion before we conduct a deeper analysis. Consider the regression table below. Observations are seen as outliers if countries have an openness to GDP ratio above 200, this is the case for Luxembourg and some observations of Ireland. We determined this threshold by looking at the the scatter plots. To check this statistically we used the hadi’s distance and the BACON method. Because both methods are similar they also come to more or less the same results (Weber, 2010). At the 5% level all observations with openness>230 or openness>208 should be considered outliers depending on the method used. This is the case for 19 or 23 observations respectively. Since 208 is considerably close to my initial estimate, the threshold remains 200. All regression are done using fixed effects models. This has the advantage that the potential for omitted variables is limited, though we lose cross country differences. However we cannot use ran-dom effects for models 1,2 & 4 since the Hausman test tells us that this would not be appropriate (Model 1 P<0.000, Model 2 P<0.0052 Model 4 P<0.004).

(19)

For model 3 we could use the random effects model since P<0.0749 so we cannot reject the null that the difference in the coefficients is not systematic. In the regression table below the results of model 3 are also from the fixed regression, for comparing purposes and the exclusion of omitted variables. All independent variables are centered and the independent variable GDPPerCapita is linearly transformed by dividing it by a 1000. All dependent variables and all inde-pendent variables are in percentages of GDP (except for GDPPerCapita). This allows us to better interpret the coefficients.

Table 1: All countries

(1) (2) (3) (4)

GovExp FRFromTransfers AGGSE ALM

Openness 0.0178* 0.0208** 0.0352*** 0.00231* (1.68) (2.06) (3.02) (1.84) GDPPerCapita -0.145** 0.0929*** 0.0919*** -0.00704** (-2.31) (3.15) (3.07) (-2.05) DependencyRatio 0.0726 -0.0339 0.0816** 0.00276 (1.20) (-0.94) (2.22) (0.69) UrbanizationRatio 1.004*** -0.00522 0.340*** 0.00426 (10.24) (-0.07) (6.98) (0.69) cons 12.75*** 10.88*** 16.15*** 0.575*** (9.26) (29.79) (43.89) (12.21) N 330 266 317 215 R2 _0.347 _0.172 _0.410 _0.032 adj. R2 0.267 -0.011 0.334 -0.164 t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01

Consider regression table (1)4above. From these first results we can see that in model 1 and 4 openness cannot explain differences in the dependent variables convincingly. It is quite surprising that in model 1, openness can only explain the level of government expenditures weakly. This is not in line with the find-ings by Rodrik (1998). Why this is the case cannot be said with certainty. One explanation could be that this analysis uses a more recent period than Rodrik, though it could also be attributed to a different approach by Rodrik than is the case here. Rodrik makes use of averages in openness and government expendi-tures while in this research yearly data are used. To test if this is the cause of the difference, we conduct a regression with averages. We create a average over the period 2013-2015 for all variables. This creates the possibility to run a standard OLS regression since there is no time variable anymore. This allows us to keep

(20)

the cross country differences which we could not use in the panel regression be-cause random effects were not appropriate. Consider the regression table below.

Table 2: Rodrik’s method

(1) (2) (3) (4)

GovExp1315 GovExp131 GovExp1315 GovExp1315

Robust Robust Openness1315 0.0301∗∗ 0.0301∗∗∗ 0.0296∗∗ 0.0296∗∗ (2.37) (3.77) (2.21) (2.31) GDPPerCapita1315 0.0722∗∗ 0.0722 -0.00682 -0.00682 (2.51) (1.58) (-0.21) (-0.12) DependencyRatio1315 -0.00632 -0.00632 0.236∗ 0.236∗ (-0.08) (-0.09) (1.74) (2.03) UrbanizationRatio1315 0.110∗∗ _0.110∗∗ _0.127∗∗ _0.127∗∗ (2.66) (2.60) (2.16) (2.56) cons 17.97∗∗∗ 17.97∗∗∗ 19.48∗∗∗ 19.48∗∗∗ (36.92) (36.92) (29.59) (36.91) N 48 48 29 29 R2 0.470 0.470 0.430 0.430 adj. R2 0.421 0.421 0.335 0.335 t statistics in parentheses ∗_{p < 0.1,}∗∗_{p < 0.05,}∗∗∗_{p < 0.01}

Model 1 and 2 use a larger variety of countries while model 3 and 4 use only developed countries. Model 2 and 4 use robust standard errors. Surprisingly the openness variable is significant in all models at the 5% level. Why does this variable turns significant when using this method? The most likely cause of the difference in outcomes is when using fixed effects we lose cross country differences. The differences between countries is likely to be bigger than within countries over time, which will lead to better results. Though we should not forget that with the other models a fixed regression using panel data does provide us with significant results. This is at least an indication that the results of Rodrik are maybe not as robust.

Model 2 uses fiscal transfers as the dependent variable. We would expect that countries that are more open, to redistribute a bigger proportion of their wealth. This is supported in this model because openness is significant (P<0.05). This variable measures the level of fiscal redistribution for the entire population. Beforehand we expected that fiscal redistribution in relation with openness was mostly taking place within the working population. This is not the case5_{. This}

(21)

indicates that most redistribution is taking place from the working population to the non-working population. This is a disturbing result because this would mean that only displaced workers are compensated and not workers that see their wages drop due to trade.

In model 3 the openness variable turns out to be highly significant (P<0.01). Actually all variables included in model 3 are significant. This raises a impor-tant question. Active labour market programmes are a part of the aggregate social expenditures. Also ALMP’s and conventional social expenditures would be expect to go hand in hand, as was made clear in the literature about re-distribution. In the ALM model however openness is only weakly significant (P<0.10). This can be due to two reasons. Firstly this can be due to a lower level of observations in model 4 (N=215) than in model 3 (N=317). we check for this possibility. When running Model 3 on the observations of model 4 the openness coefficient remains significant but does drop a bit (N=317 t=3.02 N=213 t=2.34)6_{. The possibility that the lower amount of observations in the} ALM model is the reason why its less significant can therefore not be ruled out. Secondly, it could also be the case that another part of the AGGSE is highly correlated with openness which leads to this specific result. To check for this possibility we conduct regressions on specific parts of the AGGSE to see if one of these is actually highly significant. One possible part of the AGGSE, unemploy-ment expenditures could be a big piece of the explanation. We therefore conduct a regression to see what happens when we regress unemployment expenditures on the same independent variables as before. We find that the dependent vari-able is insignificant here as well7_.

Figure 6: Active Labour Market Programs

6_{See Appendix, Table 8} 7_{See Appendix, Table 7}

(22)

We see in figure (6) above that there are lot of observations that are very low. This can affect the results. In some countries active labour market programs are rarely used, and thus variation is quite low. In figure (7) below we see that relationship between AGGSE and openness is a more close relationship.

Figure 7: Agregrate Social Expenditures

5.2 Interpretation of results

Model 2, 3 and the model with government expenditures when using averages are all significant. To get a feeling of the implications of the results we look what happens when a country gets more open by 100% points, ceterius paribus. An increase of 100% points in openness would lead to 2% point more fiscal redistribution in terms of GDP, 3.5% points more aggregate social expenditures in terms of GDP and some 3% points more government expenditures. These values all fall in the same range. Also it is interesting to see that government expenditures only increase by 3% points while social expenditures increase by 3.5% points as percentage of GDP, so the increase in government expenditures is smaller than the increase in social expenditures.

Is a 3.5% point increase in social security expenditure much? Minimum AGGSE in the data-set is 5% and maximum AGGSE is 31%. This places the results in a perspective. Only a small piece of the variation can be explained by openness. This indicates that redistribution in reaction to trade is taking place, but is maybe not sufficient. We cannot determine this because the specific costs for losers due to trade are not known.

(23)

5.3 Probing deeper

We conclude from the regression above that the level of Aggregate Social Ex-penditures can be partly explained by the openness of the economy of a country. However from this relationship we cannot be certain which causal mechanism underlies this finding. It can be due to two reasons. The argument Rodrik puts forward is that more exposed countries are more vulnerable to external shocks, to mediate these shocks the government expenditures should be higher. The mechanism that this research puts forward is that more open countries have more displacement of workers due to trade. To compensate these workers, social expenditures tend to be higher. But which mechanism is the most convincing and how can we check this statistically? The fact that AGGSE are significantly and positively correlated with openness but government expenditures are not is a first indication that the mechanism proposed by Rodrik is not the right mech-anism. Because if it was the case that governments tend to be big to mitigate external shocks, we would expect that government expenditure would matter as well, not just the social expenditures.

To try to distinguish the compensation channel and the risk channel we try to run a different regression where we regress tax revenues on openness and GDP growth. If the compensation channel is at play we would expect that tax revenues are positively correlated with openness. In line with the risk channel we would expect that tax revenues would react strongly to economic growth. Consider the regression equations below. Data is extracted from the OECD database, so includes only OECD countries.

T axP I/T OTi,t= αi,t+β1Opennesi,t+β2GDP Growthi,t+β3Depi,t+β4U RBi,t + i,t (5) From table (3) below we can conclude that in both regressions, both openness as well as GDP growth can not significantly explain tax revenues. Why this is the case can not be said with certainty. It could be due to omitted variables but since we use fixed effects this should not be a big problem. Still though the very low R2 _{indicates that the model explains variations in tax revenues very} poorly. This model thus cannot help us with distinguishing the compensation-from the risk channel.

(24)

Table 3: Tax Regression (1) (2) TaxPI TaxTOT Openness -0.00317 0.000735 (-1.35) (0.15) GDPGrowth -0.00687 -0.0110 (-0.58) (-0.43) DependencyRatio 0.0455∗∗∗ 0.0744∗∗ (3.07) (2.33) UrbanizationRatio -0.0610∗∗∗ 0.125∗∗∗ (-3.26) (3.09) cons 11.13∗∗∗ 20.04∗∗∗ (7.30) (6.08) N 762 762 R2 _0.043 _0.024 adj. R2 _-0.007 _-0.027 t statistics in parentheses ∗_{p < 0.1,}∗∗ _{p < 0.05,}∗∗∗_{p < 0.01}

5.4 Checking for econometric problems

There are some concerns about the results above. To check for these concerns we check if our data is auto correlated, or heteroskedastic. The Wooldridge test is used to check for serial auto-correlation. We reject the null that we have no first-order auto-correlation (Model 1: F (1, 32) = 149.325, p < 0.001), Model 2: F (1, 28) = 47.4, p < 0.001), Model 3: insufficient observations , Model 4: F (1, 1) = 91.362, p < 0.1). A likelihood-ratio test is conducted to check for heteroskedasticity. We also cannot be certain our data is homoskedastic (Model 1:χ2(32) = 284.30, p < 0.001, Model 2: χ2(44) = 143.03, p < 0.001, Model 3 χ2(32) = 160.02, p < 0.001, Model 4, χ2(32) = 214.29, p < 0.001). Therefore we have to use a regression with Driscoll-Kraay standard errors. To check if this effects the results we rerun the regressions with Driscoll-Kraay standard errors.

(25)

Table 4: Correcting for heteroskedasticity & autocorrelation

(1) (2) (3) (4)

Openness 0.0178 0.0208∗ 0.0352∗∗∗ 0.00231∗∗∗ (0.65) (1.92) (3.33) (4.49) GDPPerCapita -0.145 0.0929∗ 0.0919∗∗ -0.00704∗∗ (-0.98) (1.90) (2.05) (-2.62) DependencyRatio 0.0726 -0.0339 0.0816∗ 0.00276 (1.34) (-0.97) (1.97) (1.22) UrbanizationRatio 1.004∗∗∗ -0.00522 0.340∗∗∗ 0.00426 (6.63) (-0.07) (7.09) (0.67) cons 12.75∗∗∗ 10.88∗∗∗ 16.15∗∗∗ 0.575∗∗∗ (4.59) (23.42) (30.87) (7.42) N 330 266 317 215 t statistics in parentheses ∗_{p < 0.1,}∗∗_{p < 0.05,}∗∗∗_{p < 0.01}

Consider table (4) above. Accounting for heteroskedasticity and auto corre-lation does change the significance of the estimators. In model 1 the t-statistic drops to t=0.65. The significance of model 2 remains almost the same. The openness coefficient turns more significant in both model 3 and 4 (both are significant at the 1 % level). This is another indication that the compensation channel is more convincing since the model of government expenditures is not significant anymore, even at the 10% significance level.

6 Conclusion

The main goal of this research was to establish an answer to the following research question. What is the relationship between the level of exposure of the economy to the world economy and the level of redistribution?. By running regressions on different types of aggregates of redistribution we tried to find the answer. Although this gave us some results we cannot give a short and decisive answer to the research question. We will look at all three hypotheses to give a complete picture.

First of all this research has found a robust correlation between various mea-sures of redistribution and openness, especially in the case of aggregate social expenditures. Also, when using a method of averages government expenditures and openness are correlated as well. Therefore we can conclude that there is evidence for my first hypothesis that countries that are more exposed to the world economy tend to spend more on social security.

(26)

Since ALMP’s are considered to be a good tool to help people who are displaced (Card et al., 2010), we would expect them to be used to combat the distributional effects of free trade. At first sight we conclude that openness and spending on ALMP’s was only weakly significant. This could be attributed to a lower amount of observations. Also when we correct for heteroskedasticity and auto correlation we find a significant positive correlation between openness and the size of ALMP’s, at the 1% significance level. Although there were some concerns, we conclude that there is some evidence in favor of the second hypothesis. Countries that are more exposed to the world economy seem to spend more financial funds on labour market programs.

This brings us to the last hypothesis. Is there enough evidence that we can conclude with some certainty that the correlation between government size and openness is mainly due to the compensation channel. There is some evidence that does point to this. First of all in the regression table (1) openness in the model of government expenditures do not enter the regression significantly but in the aggregate social expenditures it does (at the 5% level). Though when using averages for government expenditures, openness turns significant. Rodrik (1998) also acknowledges in his paper that for developed nations the correlation between the size of the government is mainly due to spending on social security. However this could still be due to the risk channel. This research could not establish a definite prove that the reason this correlation exist is due to redistribution. Though it is at least a possibility that this mechanism is at play.

7 Discussion

The main discussion point has come forward in the previous section. The reason why government size and openness are correlated can not yet be determined. Future research should focus on disentangling the compensation- and risk chan-nel. This is a priority since this may clarify if redistribution is actually taking place because of compensation of trade or it is only because governments want to level out risk involved in a open economy. This leads us to the second point of discussion.

In this research we were able to find that open countries tend to redistribute a greater proportion of GDP. However this research can not determine if enough funds are redistributed to compensate losers of trade. Future research must try to measure the effects of trade which would make it possible to make an estimation of the lost welfare for the losers. This can then be compared with the amount of compensation, to see if the level of redistribution is sufficient. This kind of research is problematic because disentangling the effects of trade and technology is really hard. Although this kind of research comes with its challenges it is important to understand if negative opinions on free trade are grounded. This is where the motivation of this research started. But this thesis could not provide the full answer.

(27)

research we focused on developed nations. Since trade has very different effects on developing nations we would expect outcomes to be very different as well.

We also want to point out the statistical and econometric challenges of this research. First of all the statistical procedures can greatly influence the results as was seen in the comparison between the fixed effects panel data regression and the OLS regression with the government expenditure model. Both econometric approaches have their pros and cons but this is something to consider. Also feature research could make use of an alternative approach to establish more evidence for a causal relationship. Using instrumental variables could help to establish a more robust result.

Last of all there are some remarks for policymakers. As was set out earlier, policy makers should be more concerned with the distributional effects of free trade. In particular, they should first think about how to deal with the creation of losers before pushing for liberalizing trade further. This research also saw that spending on ALMP’s is still quite low in comparison with more conventional types of social security programs. More focus on ALMP’s could prove useful since research shows that the effects are generally positive.

(28)

8 References

Adema, W., Fron, P., & Ladaique, M. (2011). Is the european welfare state really more expensive?

Card, D., Kluve, J., & Weber, A. (2010). Active labour market policy evalua-tions: A meta-analysis. The economic journal, 120 (548).

Davidson, C. & Matusz, S. J. (2006). Trade liberalization and compensation. International Economic Review, 47 (3), 723–747.

Davidson, C., Matusz, S. J., & Nelson, D. R. (2007). Can compensation save free trade? Journal of international Economics, 71 (1), 167–186.

Haddad, M., Lim, J. J., Pancaro, C., & Saborowski, C. (2013). Trade open-ness reduces growth volatility when countries are well diversified. Canadian Journal of Economics/Revue canadienne d’´economique, 46 (2), 765–790. Haskel, J., Lawrence, R. Z., Leamer, E. E., & Slaughter, M. J. (2012).

Global-ization and us wages: Modifying classic theory to explain recent facts. Journal of Economic Perspectives, 26 (2), 119–40.

Kapstein, E. B. (2000). Winners and losers in the global economy. International Organization, 54 (2), 359–384.

Kletzer, L. G. (2004). Trade-related job loss and wage insurance: a synthetic review. Review of International Economics, 12 (5), 724–748.

Krugman, P., Obstfeld, M., & Melitz, M. (2015). International Economics. Pearson Education.

Mendoza, E. G. (1997). Terms-of-trade uncertainty and economic growth. Jour-nal of Development Economics, 54 (2), 323–356.

Reuters (2016). Survey shows plunging public support for ttip in u.s. and ger-many.

Rodrik, D. (1998). Why do more open economies have bigger governments? Journal of political economy, 106 (5), 997–1032.

Stock, J. & Watson, M. W. (2015). Introduction to econometrics 3rd ed. Essex: Pearson Education.

Walter, S. (2010). Globalization and the welfare state: Testing the microfounda-tions of the compensation hypothesis. International Studies Quarterly, 54 (2), 403–426.

Wang, J. & Caminada, K. (2017). Leiden lis budget incidence fiscal redistribu-tion dataset on income inequality.

Weber, S. (2010). bacon: An effective way to detect outliers in multivariate data using stata (and mata). Stata Journal, 10 (3), 331–338(8).

(29)

9 Appendix

Table 5: Developed Countries Only

(1) (2) (3) (4)

Openness 0.00971 0.0201∗ 0.0335∗∗∗ 0.00236∗ (0.96) (1.78) (2.78) (1.78) GDPPerCapita2 -0.203∗∗∗ 0.105∗∗∗ 0.109∗∗∗ -0.00767∗ (-3.38) (2.88) (3.31) (-1.97) DependencyRatio 0.314∗∗∗ -0.00953 0.129∗∗∗ 0.000758 (4.58) (-0.18) (2.68) (0.13) UrbanizationRatio 0.661∗∗∗ -0.0324 0.297∗∗∗ 0.00605 (6.03) (-0.38) (5.00) (0.81) Constant -39.47∗∗∗ 11.59∗∗ -14.98∗∗∗ 0.168 (-5.08) (2.07) (-3.61) (0.35) Observations 300 211 294 195 R2 0.377 0.169 0.397 0.032 Adjusted R2 0.299 0.026 0.321 -0.167 t statistics in parentheses ∗_{p < 0.1,}∗∗_{p < 0.05,}∗∗∗_{p < 0.01}

(30)

Table 6: Checking for Working Population Bias (1) (2) FRFromTransfersWP FRFromTransfers Openness -0.00869 0.0208∗ (-0.93) (2.06) GDPPerCapita 0.0272 0.0929∗∗ (0.99) (3.15) DependencyRatio -0.0698∗ -0.0339 (-2.08) (-0.94) UrbanizationRatio -0.0330 -0.00522 (-0.51) (-0.07) cons 13.47∗∗ 9.679 (2.81) (1.87) N 266 266 t statistics in parentheses ∗_{p < 0.05,}∗∗_{p < 0.01,}∗∗∗_{p < 0.001}

Table 7: Unemployment Expenditures

(1) (2) AGGSE UnemploymentEXP Openness 0.0352∗∗ 0.00459 (3.02) (1.78) GDPPerCapita 0.0919∗∗ -0.0147∗ (3.07) (-2.42) DependencyRatio 0.0816∗ 0.00446 (2.22) (0.53) UrbanizationRatio 0.340∗∗∗ 0.0137 (6.98) (1.33) cons -16.03∗∗∗ _-0.208 (-4.02) (-0.24) N 317 239 t statistics in parentheses ∗_{p < 0.05,}∗∗ _{p < 0.01,}∗∗∗_{p < 0.001}

(31)

Table 8: Using the observations of Model 4 (1) AGGSE Openness 0.0341*** (2.34) GDPPerCapita 0.0220 (0.58) DependencyRatio 0.0846* (1.94) UrbanizationRatio 0.350*** (5.20) cons -15.02*** (-2.97) N 213 R2 _0.307 adj. R2 _0.165 t statistics in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01

Winners and Losers from trade: Does redistribution actually happen?