CHARLES UNIVERSITY
FACULTY OF SOCIAL SCIENCES
Institute of Economic Studies
Anton´ın Galle
Income inequality analysis in developing countries in the SDG Framework
Bachelor thesis
Prague 2020
Author: Anton´ın Galle
Supervisor: doc. Ing. Tom´aˇs Cahl´ık CSc.
Academic Year: 2019/2020
Bibliographic note
GALLE, Anton´ın. Income inequality analysis in developing countries in the SDG Framework. Prague 2020. Bachelor thesis (Bc.) Charles University, Faculty of Social Sciences, Institute of Economic Studies. Thesis supervisor doc. Ing. Tom´aˇs Cahl´ık CSc.
Range of thesis
Number of characters with spaces: 57 778
Abstract
This thesis analyses the effect of urbanization, secondary education, employ- ment in agriculture and GDP on income inequality in developing countries in context of the Sustainable Development Goals programme. A panel data analysis on unbalanced panel of 99 countries in time period 1991 - 2017 is performed. Significant negative effect of education on income inequality has been found in the whole time period and the effect of employment in agri- culture changed from positive to negative during the time period, suggesting that action directed at helping farmers in developing countries may deliver desirable effects.
Abstrakt
Tato pr´ace analyzuje vliv urbanizace, sekund´arn´ıho vzdˇel´av´an´ı, zamˇestnanosti v zemˇedˇestv´ı a HDP na nerovnost pˇr´ıjmu v rozvojov´ych zem´ıch v r´amci pro- gramu C´ıle udrˇziteln´eho rozvoje. Pr´ace obsahuje anal´yzu panelov´ych dat na nevyv´aˇzen´em panelu 99 zem´ı v obdob´ı 1991 - 2017. V cel´em ˇcasov´em obdob´ı byl zjiˇstˇen v´yznamn´y negativn´ı dopad vzdˇel´av´an´ı na nerovnost pˇr´ıjmu a vliv zamˇestnanosti v zemˇedˇelstv´ı se bˇehem zm´ınˇen´eho ˇcasov´eho obdob´ı zmˇenil z pozitivn´ıho na negativn´ı, coˇz naznaˇcuje, ˇze mezin´arodnˇe koordinovan´e akt- ivity zamˇeˇren´e na pomoc pro zemˇedˇelce v rozvojov´ych zem´ıch mohou pˇrin´est ˇz´adouc´ı ´uˇcinky.
Keywords
Income inequality, Sustainable Development Goals, Urbanization, Second- ary education, Panel data analysis
Kl´ıˇ cov´ a slova
Pˇr´ıjmov´a nerovnost, C´ıle udrˇziteln´eho rozvoje, Urbanizace, Sekund´arn´ı vzdˇel´av´an´ı, Anal´yza panelov´ych dat
Declaration of Authorship
I hereby proclaim that I wrote my bachelor thesis on my own under the leadership of my supervisor and that the references include all resources and literature I have used.
I grant a permission to reproduce and to distribute copies of this thesis document in whole or in part.
Prague, 5 May 2020
Signature
Acknowledgment
I would like to thank my supervisor, doc. Ing. Tom´aˇs Cahl´ık CSc. for his guidance. Also, thanks to my family for their support and patience.
Bachelor thesis proposal
Author Anton´ın Galle
Supervisor doc. Ing. Tom´aˇs Cahl´ık CSc.
Proposed topic Income inequality analysis in developing countriesin the SDG Framework
Research question and motivation
In the year 2015 the UN announced a set of goals targeting poverty, climate change and protection of nature. These goals are similar to Millenium De- velopment Goals, but now there are twice as much of them and they are much more ambitious. In the 10th Sustainable Development Goal the UN aims to achieve and sustain income growth of the bottom 40 per cent of the population at a rate higher than the national average for every country.
This thesis is motivated by the interest if this ambitious goal can be fulfilled for developing countries and if it were not to be fulfilled for every country, how many countries can be expected to reach this target.
Simon Kuznets shows that if a low-income country gets richer, income in- equality grows up to some point, where additional income starts to decrease inequality (”Kuznets Curve”). However, his work is more than 60 years old now and recent development, mainly in Africa, contradicts his conclusions, as was shown in the working paper ”Growth is good for the poor” by D.
Dollar and A. Kraay in 2002. This thesis will also discuss possible reasons why the Kuznets Curve does not work in the 21st century anymore.
Contribution
The purpose of this thesis is to discuss feasibility of Sustainable Development Goal regarding income inequality and predict income inequality in develop- ing countries by 2030. Results will be supported by econometrical analysis using income data from the DataBank provided by World Bank.
Methodology
This thesis will estimate the feasibility of fulfilling the goal for selected De- veloping Countries by using econometrical analysis of panel data. As the input, it will use data tables from WorldBank: income share of lowest 20%
of population of a country and income share of second lowest 20% from years 1998 ? 2018 for every listed country. Then it will predict the share of bottom 40 per cent of every selected country depending on two para- meters: yearly change in urbanization and net enrolment rate to secondary education, providing estimate of their effect on inequality.
Outline
1. Introduction
2. Sustainable Development Goals 3. Theoretical background 4. Methodo- logy 5. Prediction Method 6. Results and their discussion 7. Conclusion List of academic literature
S. Kuznets, Economic Growth and Income Inequality (1955)
P. Vanhoudt, An assessment of the macroeconomic determinants of inequal- ity (2000)
D. Dollar and A. Kraay, Growth is Good for the Poor (2002) D. Checchi, Education, Inequality and Income inequality (2001)
F.A. Cowell and D.G. Champernowne, Economic Inequality and Income Dis- tribution (1999)
List of tables and figures
Table 1: Fixed effects assumptions 12
Table 2: Random effects assumptions 12
Table 3: Relationship between SDG indicator 10.1 and Gini 23 Table 4: Estimated coefficients in period 1991 - 2007 26 Table 5: Estimated coefficients in period 2008 - 2017 30
Figure 1: Scatterplot of Ginis and income of bottom 40% 22 Figure 2: Inflation-adjusted prices of crude oil in USD 25 Figure 3: Development of Gini coefficient in Tanzania 33 Figure 4: Development of Gini coefficient in South Korea 33 Figure 5: Development of Gini coefficient in Mexico 34
Frequently used abbreviations
Abbreviation Meaning
FE Fixed Effects
GDP Gross Domestic Product MDG Millenium Development Goal
RE Random Effects
SDG Sustainable Development Goal
UN United Nations
Contents
1 Introduction 1
2 Literature review 3
3 Goal 10: Reduce inequality within and amongst countries 9
4 Econometrical analysis 11
4.1 Analysis of panel data . . . 11 4.2 Forecast method . . . 14
5 Specification of data 15
5.1 Income share held by lowest 40% . . . 15 5.2 Gini coefficient . . . 16 5.3 Urbanization, workers in agriculture, education and GDP per
capita . . . 16 5.4 Missing data . . . 18 6 Relationship between SDG target and GINI 21
7 Empirical panel data model 23
8 Results and their interpretation 25
8.1 Earlier period . . . 25 8.2 Later period . . . 29 8.3 ARIMA forecast . . . 31
9 Conclusion 35
References 37
Appendix 42
1 Introduction
During first fifteen years of twenty-first century the United Nations pursued Millenium Development Goals - a set of goals targeted at improving people’s living conditions, especially in poorer regions of the world. Based on this programme’s success, leaders of all member nations all over the world ap- proved a similar, but bigger and more ambitious plan, called Sustainable Development Goals, or SDGs, which follow the same format - several areas of future improvement measured by specific indicators during 15-year time period.
The plan is expected to be fulfilled in 2030 and consists of 17 devel- opment goals, with each one of them having a set of clear indicators of success - 169 indicators in total. Most of the goals are in some way tar- geted against poverty, as the General Assembly of the UN considers it ”the greatest global challenge and an indispensable requirement for sustainable de- velopment” (United Nations, 2015). The UN publishes yearly short reports for every goal, available on the website dedicated to SDGs, and also yearly
”Sustainable Development Goals Report” summarizing overall progress in all goals, available on the webpage of the Statistics Division of Department of Economic and Social Affairs of the UN.
The amount of scientific works, essays and even newspaper articles dis- cussing the SDGs in last few years is so high that it can be considered a buzzword among development economists and politicians. While some crit- ics (e.g. Bj¨orn Lomborg) consider the agenda to be ”implausibly optimistic and inefficient” and others (e.g. David Cameron) unnecessarily broad, many believe that it is the right way to improve conditions in developing and least developed countries. This claim may be supported by the fact that these seventeen goals were formulated after thorough discussion on the Rio+20 conference in 2012, where every country had the opportunity to propose adjustments to them (unlike the MDGs, which were not based on the out- come of any large international meeting), so leaders of countries of the UN in principle agree with them, or at least agreed with them in 2012. Also,
coordinated international action in the framework of MDGs was successful in many areas of interest, so there is a reason to believe similar plan for next 15 years could also deliver desirable results.
The main focus of this thesis lies on the 10th SDG: Reduced Inequalities, particularly on the question whether three variables (education, urbaniza- tion and GDP), which have been proposed as main drivers of inequality in previous works, do influence income inequality in developing and least de- veloped countries. Even though the SDG programme is officially followed by all UN member states, several goals, including Reduced inequalities, are focused mainly on developing countries where majority of people in need of help live. However, available data on income inequality, especially on vari- ables used by the UN as indicators of success, are somewhat sparse. One goal of this thesis is to use data from last three decades and test if hypo- theses presented before still hold, and second goal is to shed light on some causes of income inequality in recent times.
This thesis begins with a literature review, where the works that analysis presented in this thesis is based on are listed. After the review a chapter 3 which is dedicated to the Sustainable Development Goal 10: Reduced In- equalities follows. In chapter 4 theoretical basis for used models is explained and in chapter 5 data used by those models are described, along with ex- planation how missing data were treated and the solution of the problem of insufficient data provided on the core indicator of fulfillment of 10th SDG.
Chapter 6 describes why another measuerment than the UN’s approved in- dicator is used. Chapter 7, in which the empirical model is described follows, and finally in chapter 8 the results of the panel data models and forecast of future development of income inequality in three selected countries are presented. Last chapter concludes findings of this thesis, followed by the list of developing and least developed countries as classified by the UN in Appendix.
2 Literature review
The general topic of income inequality and its causes has been thoroughly studied for decades. All analyses of this problematic face the constraint of limited available data, which is true especially for investigating income inequality in developing and least developed countries, where the process of collecting accurate data is complicated. Several authors focused on a single developed country with abundant data, such as Thomas Piketty (2003), who performed a detailed analysis of determinants of income inequality in France during almost whole 20th century. However, literature describing contemporary situation in developing and least developed countries is not very broad.
The basics of studying income inequality were laid by Kuznets (1955), who published analysis of the relation between income of whole society and income of its poor members. He used data for three countries: the United States, the United Kingdom and Germany (without their overseas colonies) for the time period 1870 to 1947, when these countries experienced trans- ition from mostly rural countries with agriculture-based economy to heavily urbanized countries with most jobs in industrial sector.
The author observes a trend towards more equality, present in all three countries, differing only in its speed and time period. Variables he lists are different for each country, but all show gradual increase in poor citizens’
share of income before taxes, which gains momentum especially after the First World War. He also indentifies reasons why it may be surprising - given that income-yielding assets tend to accrue in ownership of the rich, and available jobs were moving from the agricultural sector to the industrial sector, which he considers less stable of the two. The study aims to explain this trend by several changes in society, namely the shift of political prefer- ences towards greater redistribution of wealth, changes in demography and shift in typical sources of income from assets (especially land) to services.
The last section of this study compares income inequality trends in de- veloped and developing countries. Income inequality has been found to
be significantly larger in developing countries. This fact is explained in the study by lack of a middle class in those lands at that time and exist- ence of privileged minorities in colonial territories like Rhodesia or relatively newly (in year 1955) independent countries like India. The resulting rela- tion between income and inequality is shown as the widely known inverted U-shaped ”Kuznets curve”.
Vanhoudt (2000) evaluates Kuznets’ conclusions and builds on his work.
In his paperAn assessment of the macroeconomic determinants of inequality he proposes a hypothesis that one variable - per capita income (or GDP per capita) is not enough to determine income inequality, but there is some set of countries’ attributes that have a large influence on income inequality, and the Kuznets curve still holds. In Vanhoudt’s model, income of individuals (and therefore also income of the poor) cannot be considered exogenous like in Kuznets’ study.
This model uses labour, physical capital and human capital as inputs to the production function, which has the Cobb-Douglas form. It resembles neoclassical growth model. Income inequality is modelled as the difference between wage of unskilled worker, whose human capital is 0, and wage of skilled worker, whose human capital is h, while each worker is either skilled or unskilled and all workers in a category are indistinguishable. The eco- nomy converges to its steady state. Vanhoudt shows that if relatively small part of workers are skilled, the convergence towards the steady state causes inequality to increase, but from a certain share of skilled workers the ap- proaching of the steady state results in decrease in inequality. This share is unique to each economy, and cannot be generalized for all countries, as Kuznets proposed (he used terms ”agricultural” and ”industrial” instead of Vanhoudt’s ”unskilled” and ”skilled” workers, but the idea remains very similar.)
According to Vanhoudt, the real factors influencing income inequality are the exogenous variables in the neoclassical model - propensity to increase
physical capital (savings) and population growth. As for propensity to in- crease human capital (education), its effect is not statistically significant at the 0.05 level, but from that it cannot be concluded that the effect is not present. As expected, propensity to save has statistically significant negative effect on inequality and population growth has significant positive effect.
Dollar and Kraay (2002) in their study Growth is good for the poor ex- amine which economic policies influence income share of the poorest 20 % of countries’ inhabitants. Firstly they confirm the hypothesis that rising overall income results in rising incomes of the poorest quintile approximately 1:1, meaning that in absolute terms, growth is indeed ”good for the poor”. To achieve this confirmation they used the pooled OLS and 2SLS methods on a sample of 418 observations of logarithm of income per capita as explanatory variable and income of poorest 20% as dependent variable.
Authors also test if Kuznets’ hypothesis (growth in poor countries favors their richer inhabitants) is supported by data from years after his paper was published, and discover that the effect proposed by Kuznets is not present, meaning that even in low-income countries income growth does not signific- antly reduce poorest 20% income share.
Following that, authors provide analysis of determinants of income share of the poor, with the focus on inflation, share of exports and imports on GDP of a country (as a proxy for globalization), government expenditures, finan- cial development and strength of property rights. According to regressions provided by this paper, none of these indicators exhibit effects statistically significant on the 5% level, which might be surprising. This study con- cludes by stating that empirical evidence does not support many stylized facts about determinants of income of the poor and emphasizing the effect of total income growth on income growth of the poor. Authors recommend generally effective pro-growth policies, e.g. trade openness or fiscal discip- line to policymakers who aim to reduce poverty.
Cobbinah, Erdiaw-Kwaise and Amoateng (2014) studied urban poverty in developing countries. They conducted their research one year before the start of Sustainable Development Goals programme, but the concept of ”sus- tainable development” was established before. In the study they summarize many other research papers and put them into context. In first part the authors explain what can be considered sustainable development and what possible problems can cause a deviation from the sustainable path. They observe that despite many efforts from several international organizations, reality in many developing and least developed countries is in stark contrast to those organizations’ plans. In the paper it is pointed out that most diffi- culties faced by developing coutries can be attributed to two causes: poverty and rapid urbanization.
Following that, authors explore different definitons of poverty from the point of view of developing countries’ inhabitants. They argue that any simple measurement (like poverty line at 2 USD per capita income for a day) fails to explain such complex phenomenon fully, so they offer several different variables, that better describe poverty like citizens of developing coutries feel it. Amongst those variables, the most prominent is unemployment and infant mortality. From both kinds of measurement, Africa south from the Sahara and, in lesser extent, South East Asia come out as regions where the need of improvement is the most pressing. Authors also examine effects of urbanization on poverty and possible links to sustainable development, and discover that the process of city sprawling is different in developed and developing countries. The main difference is the lack of any urban plan and fast spreading of a few large aglomerations in most developing countries, as opposed to planned and balanced city sprawling in developed world. According to the paper, such urbanization creates a high risk of amplifying already present problems as lack of clean water, environmental hazards and social exclusion of certain neighbourhoods (”slums”).
The study concludes by statement that both poverty and unguided urb- anization remains present in many developing countries. Authors state that
international action to relieve poverty may benefit from poverty-urbanization approach and emphasize the need to incorporate local inhabitants into decision- making process. They also point out that until 2014 there was little evidence of success of any policy intervention and call for further research in this field.
Ayiar and Ebeke (2019) discuss the relationship between inequality of op- portunity and inequality of outcome. In their working paper, inequality of opportunity is synonymous with intergenerational mobility, measured by the difference in income of fathers and sons when they reach their peak earnings (around 50 years of age). They are interested in the effect of intergenera- tional mobility and income inequality on the GDP growth. Many models in many papers work with income ineqality without including intergenerational mobility with mixed results. Authors present hypothesis that inequality of opportunity is the missing link between income inequality and GDP growth, i.e. that the adverse effect of income inequality on GDP growth is signific- antly larger in societies with strong intergenerational rigidities, where income inequality persists across generations more and is harder to diminish.
The authors offer three explanations why this effect may be present. Ac- cording to them, intergenerational rigidities may hamper the access of lower classes to quality education, which creates a net loss of human capital, im- pair the ability of the economy to overcome negative external shocks or cause worse access to available financing for enterpreneurs from ”outsider” class.
The study does not aim to discover which channel is the strongest or if they all are present, but it examines the overall effect of these three channels together.
At the core of the working paper is a panel data model, in which the dependent variable is the 5 - year nonoverlapping average of real per capita GDP growth in each country (5 - year average is used instead of yearly data to diminish or filter out sharp effect of shocks) in time t, while independ- ent variables being log of income per capita in time t-1 (first lag), income inequality measured by Gini in time t-1, interaction term between income
inequality and intergenerational mobility (the variable of focus of the paper), country-specific and time-specific dummies and a matrix of covariates, which contains investment in percent of GDP, log of average years of schooling (as a proxy for human capital) and trade openness. Purpose of this matrix is reducing possible omitted variable bias caused by attributing the effect of variables that determine GDP in neoclassical model to intergenerational mo- bility or income inequality. At the end of the paper the possibility of income inequality being endogenous is discussed, but sufficiently rejected.
The regression confirmed authors’ hypothesis. The restricted model - only with income inequality as independent variable - provides inconclusive results, as authors expected. Regression with intergenerational mobility shows that its effect on GDP growth is statistically significant, therefore inequality of income harms countries’ economic growth much more if it is also accompanied by inequality of opportunity. It also shows that even income inequality by itself has significant negative effects on economic growth, but its magnitude differs across countries with different levels of inequality of opportunity.
After presenting their results, Ayiar and Ebeke also propose several policies to address the causes of inequality of opportunity, or ”level the playing field”. Amongst other advices, they recommend widespread international action focused on keeping students longer in education system, especially provide secondary education to lower strata of society (according to them, investments towards tertiary education do not reach students coming from families that need help the most).
They also point out that liberalization of labour market would help to increase intergenerational mobility by decreasing structural unemployment.
Last but not least, policies that would assist starting businesses in secur- ing sufficient funding would not only help low-income enterpreneurs, but also promote innovation and spread of good ideas. If all these three issues were solved, all three possible channels of maintaining intergenerational ri- gidities could be closed and programmes directed towards decreasing income
inequality would be more efficient than now.
3 Goal 10: Reduce inequality within and amongst countries
Income inequality can be considered to be one of most pressing issues in development economics. Any policymaker interested in reducing impacts of poverty also has to bear in mind its negative effects. Most of countries ex- perience a period of steady economic growth since 2010, but the distribution of its benefits varies widely between countries. The United Nations acknow- ledge that income inequality tends to persist and grow in past several years (United Nations, 2019). According to the SDG Fund, richest 10% of world population earns 40% of income in 2019, and inequality increased by 11% in developing countries since 1990 after correcting for population growth (SDG Fund, 2019).
While researchers of income inequality in developed countries usually fo- cus on income share of richest few percent (or even one percent) of earners, those interested in developing or least developed countries focus on the bot- tom few deciles. The first target of 10th goal demands that all countries should”by 2030, progressively achieve and sustain income growth of the bot- tom 40 per cent of the population at a rate higher than the national average”
(UN, 2015). This formulation corresponds with the approach focused on developing countries, where most of world’s impoverished people live.
Like whole SDG agenda, actions aiming to help to fulfill 10th goal are directed mainly towards developing and least developed countries, which are not as likely to help themselves as developed countries. These actions include - but are not limited to - programmes of the SDG Fund, which are implemented exclusively in developing and least developed countries.
According to Progress Reports published in first four years of the pro- gramme, the results appear to be mixed. Progress Report 2019 provides certain insight: from 92 countries with available data, 50 were fulfilling the
criterion during the time period 2011 - 2016. However, majority of countries with available data are developed, therefore countries which are the most important for this goal are mostly missing from this analysis. The Progress Report itself recognizes this limitation and points out the need for better collection of data.
Despite its limitations, this analysis offers some valuable information - it identifies countries where the share of income of poorest 40% was increas- ing significantly even before SDGs were implemented. Those countries are interesting for researchers in a different way than those where the share was stagnating or decreasing (or data were unavailable). It is clear that in those parts of the world something has been done right, but it is much less clear what, and many papers are discussing the possibilities.
On the other side, countries that were not on the right track during this period are the ones that 10th goal is directed towards. Such countries are situated all over the world, but certain regions are better off than others.
Countries that are problematic from the point of view of goal fulfillment are often located in the Middle East, in Africa and in Latin America (United Nations, 2019). Especially the two latter regions are plagued by large income inequality at least in last half a century (as can be seen for instance from the SWIID database created by Frederick Solt.)
While several developed countries also suffer from substantial rise of in- come inequality, for instance Greece, Spain, Portugal and Italy, the unsat- isfactory situation there arose only in last decade and its probable cause is mismanaged Eurozone debt crisis. There exists broad recent literature about this topic and these countries are not the focus of this thesis.
4 Econometrical analysis
4.1 Analysis of panel data
The panel data analysis, whose results are described in section 6, was per- formed in the software R (version 3.2.2) using package plm (Croissant and Millo, 2008), which is very user-friendly and useful for estimating all four methods listed above. All estimations and tests except the Breusch - Pagan test are performed by functions supplied by the package plm or those from base R.
Models included in this thesis work with panel data in a way presented in the bookEconometric Analysis of Cross Section and Panel Data written by J. Wooldridge (2010). The main advantage of using panel data is the fact that it allows the model to deal with heterogeneity caused by unob- served differences between countries. It is also beneficial that panel data include both the time and cross-sectional dimension, so they are well suited for studying dynamics of change, while also making use of all information available in data sources.
The general expression of the panel data model can be written as
yit =βxit+ci+uit (1) whereyit stands for the dependent variable in unit i at time t (in this case value of the Gini coefficient in country i in year t),xitstands for the vector of independent variables (GDP,urbanization and education),β is the vector of coefficients measured by the model, ci is unobserved effect specific for each cross-sectional unit (each country) and uit is the idiosyncratic disturbance specific for each cross-sectional unit in each time period.
In this thesis four methods of estimation are used: Pooled OLS, first differences, fixed effects and random effects. Amongst commonly used panel data estimation methods, I consider FE or RE to be the best for the model (in next section it will be decided which one), because they have several advantages: The data satisfy their assumptions reasonably well, they are
not very challenging to interpret (as opposed to, for example, methods based on maximum likelihood function), but they are efficient enough to preserve almost all information from the data in the process of estimation (as opposed to first differencing). Pooled OLS and first differences are included mainly for comparison with random effects or fixed effects.
In order to correctly estimate the coefficients corresponding to dependent variables in the model, several assumptions need to be satisfied. These assumptions are listed for example in the book Econometrical analysis of cross section and panel data by J. Wooldridge (2010) in chapter 10. For the sake of clarity, they are listed in tables 1 and 2.
Table 1: Fixed effects assumptions.
FE.1 We have random sample from cross sections
FE.2 Each explanatory variable changes over time and no perfect linear relationships exist among the explanatory variables.
FE.3 E(uit|Xi, ci) = 0 for all time periods (strict exogeneity) FE.4 V ar(uit|Xi, ci) =V ar(uit) =σ2(homoskedasticity)
FE.5 Cov(uit, uis|Xi, ci) = 0 for eacht different froms (no autocorrelation)
Table 2: Random effects assumptions.
RE.1 - 5. All assumptions like in FE
RE.6 E(ciXi) =β0 (constant expected value of unobserved effects) RE.7 V ar(ciXi) =σ02 (constant variance of unobserved effects)
Under assumptions FE.1, FE.2 and FE.3 is estimation of the model yit =βxit+ci+uit by the method of fixed effects unbiased and consistent (Wooldridge, 2010). If assumptions FE.4 and FE.5 were also fulfilled, then the FE estimator would be best linear unbiased estimator. Therefore, in the case of individual heterogeneity being correlated with at least some explan- atory variable, estimates provided by fixed effects would be the most reliable.
On the other hand, if individual heterogeneity turned out to be uncorrelated
also be unbiased, consistent and on top of that more efficient.
Several tests need to be performed to detect which potentially unwanted properties of the data have to be adressed and to decide which estimator is the most trustworthy. Firstly, the Durbin-Watson test (the variant for panel models) was carried out. Its null hypothesis is that time series are stationary and based on the results of it the null hypothesis can be rejected, so there is a strong reason to believe that autocorrelation is present in these data during whole period 1990 - 2017.
This is unpleasant, but not entirely unexpected, given that income in- equality is known to be persisting for long time, even across generations (Ayiar and Ebeke, 2019). The adaptation used in this thesis is using autocorrelation-robust standard errors for all panel data models, which for- tunately are easily implementable inplm package in R.
Next useful test is Breusch-Pagan test for heteroskedasticity. Accord- ing to its results the null hypothesis that the data are homoskedastic can not be rejected, however this finding does not confirm homoskedasticity of the data, only cannot confirm that the data are heteroskedastic. Anyway, heteroskedasticity- and autocorrelation-robust standard errors are used to correct for both these problematic properties of the data.
Further testing is handy for deciding which estimation method yields most valid result. F test for individual effects (pF test in R) is used to distinguish between pooled OLS model and fixed effects model. Because analyzed countries are very diverse, it is hard to believe that there are no individual effects, and F test confirms that, so there is a strong evidence that estimation from pooled OLS model is not to be trusted (for both time periods for which the models are estimated, and for the whole period 1991 - 2017 too). It is still included in result tables, but it should be borne in mind that fixed effects is much more trustworthy.
Then it only remains to decide between fixed effects and random effects.
To answer this question, the Durbin-Wu-Hausman test is performed. It
evaluates consistency of random effects estimation. Because fixed effects es- timation is known to be consistent, but inefficient, random effects estimation may help to increase efficiency. However, if the random effects estimation proves to be inconsistent, then fixed effects are the better option. In case of data used in this thesis, random effects proved inconsistent in all three cases - in both time periods used in models and in the whole data overall, so fixed effects are the most trustworthy estimator.
There is a slight inconvenience that because robust standard errors were used, the distribution of test statistic in the Durbin-Wu-Hausman is not χ2 anymore, so the difference in the distribution might influence the test results.
Therefore, the test results should be taken with a grain of salt, but in the tradeoff between some efficiency gain at the price of potentially inconsistent estimator, fixed effects still come out better.
If random effects proved to be consistent, another test would be performed to decide between pooled OLS and random effects, however tests recommend fixed effects over both, so the last test is not necessary.
The Breusch - Pagan test (for heteroskedasticity) is not supplied by the package plm, so a function from the package lmtest(Kuznetsova, Brockhoff and Christensen, 2017) is used.
4.2 Forecast method
In cases when forecasting future development of a variable based on economic intuition is not sufficient, the ARIMA (autoregressive integrated moving av- erage) method can be used. It consists of three parts: the forecasted variable can be regressed on its previous p values (hence the AR part) and d-times differenced to eliminate non-stationarity (I part). The MA part means that regression error is modelled as the moving average of lastqerrors. Therefore, specific models are labelled as ARIMA (p, d, q).
Hyndman and Khandakar (2008) introduced an algorithm which is used to select best combination of p, d and q for given data.1 This algorithm was
1Exact description can be found in the paperAutomatic time series forecasting:the forecast package
used for the model and it recommended using ARIMA (1,1,0) for all three countries which are compared in last section of this thesis.
5 Specification of data
All data have been downloaded from their sources on 10th February 2020.
Brunei, Ivory Coast, the Republic of Kongo, Saint Thomas and Prince and Taiwan are not included despite being in category of developing countries, because there are no data about any of variables of interest available. Data for Sudan until 2009 include both today’s Republic of Sudan and South Sudan, data after the split of this country have been omitted.
5.1 Income share held by lowest 40%
Data about income share of lower two quintiles of developing and least de- veloped countries’ population originate in the WorldBank online database of indicators. Income share of lowest 40% is a simple sum of two published datasets - about lowest quintile and about second lowest quintile. They are available for individual countries on the yearly basis and are collected by na- tional surveys of representative households adjusted for household sizes, as is stated in those datasets’ metadata. Empirical model in this thesis uses only datapoints from countries categorized as ”developing economies” or ”least developed countries” by the UN (as listed in Appendix A) from years 1990 - 2017.
Usefulness of these data is very limited, because they suffer from several problems. The core issue lies in their sparseness; they contain only 17%
of values, remaining values are missing. Most of values are also provided from Latin American countries, on the other hand Africa is represented only by 58 datapoints. This is likely caused by the fact that most developing and least developed countries do not conduct surveys about their citizens’
incomes every year and many of them did so only once or twice in the period of interest of this thesis.
Accuracy of values may also not be high, because representative house- holds of developing or least developed countries may represent only those households that the government agency conducting the survey is capable of reaching, which might disproportionally exclude the truly poorest people.
Given that income share of lowest two quintiles of countries’ population is a key indicator of evaluating progress of a Sustainable Development Goal, it can be surprising that data about such important metric are so problematic.
This thesis deals with those limitations in following sections.
5.2 Gini coefficient
Gini coefficients for all the analyzed countries have been obtained from The Standardized World Income Inequality Database (Version 8) available at Harvard Dataverse. This dataset pools sources from OECD databases, WorldBank indicators, Eurostat, regional international organizations (e.g.
the CEDLAS), national statistical offices and others. More information about how were these data created is given by its creator in the paperMeas- uring Income Inequality Across Countries and Over Time: The Standardized World Income Inequality Database (Solt, 2019). It offers several versions of estimates of the Gini coefficient (this thesis uses pre-tax, pre-transfer version of Gini). Even though 40% observations from the countries and timespan in question are still missing, it provides much more information about income inequality than WorldBank’s quintile surveys.
5.3 Urbanization, workers in agriculture, education and GDP per capita
Data about variables with a supposed effect on income inequality come from WorldBank’s online database of indicators. While some of the limitations of WorldBank’s data remain true also for these variables, because the data come from similar surveys, the main problem - missingness of data - is not as restraining as in case of income of the lowest 40%. Education is repres- ented by net enrollment for secondary education in this model and it is the
sparsest of the three explanatory variables - 32% of datapoints is missing, mostly from countries from the Middle East and Central Africa.
Urbanization data can be considered most reliable of the four and gener- ally without major flaws - there are no missing values and the measurement is likely to be precise, because almost all countries tend to keep exact records where their citizens live, notably either for tax purposes or for purposes of compulsory military service. The only caveat of using this variable is the fact that different countries use slightly different definitions of urban areas, but for vast majority of people those differences do not change their status of belonging to the group of urban (or non-urban) population.
Shares of workers in agriculture were obtained as simple averages of male and female agriculture workers provided by WorldBank (for simplicity, it is assumed that in every country and every year half of the population is male and half is female. The ratios of course were not exactly 50:50 everywhere for all the time, but deviations caused by the differences in male/female ratios are smaller than WorldBank’s rounding error, so it is impossible to include so small numbers into models.) There are no missing datapoints.
According to details of this dataset provided by the WorldBank, the data may be slightly understated, because like definition of an urban dweller, definition of agriculture worker slightly varies between countries - somewhere only employees are considered to be workers, some countries include self- employed farmers. Also, it is possible that some contries do not count unpaid family members, so in some cases whole family of farmers may be counted as one person, because only one person sells the products, one person pays taxes and therefore the government does not know or care there are more people there (developing countries tend to face this problem much more than developed).
However, counting employees in other sectors suffers from the same prob- lems, so it is likely that some workers would be missed for every sector.
Because the variable is the share of workers in agriculture, these mistakes
may cancel out or at least decrease enough to be smaller than the rounding error in the data.
GDP data are of similarly high quality - because it is such important measurement for many purposes, it is relatively accurate. According to metadata of the dataset, GDP of a country is determined simply as gross value added of all residents of the country plus product taxes minus sub- sidies, calculated in thousands of current U.S. dollars2 , divided by country’s population on the 1st of June of a particular year. 21 datapoints are missing (for Djibouti, Somalia, Syria, Tanzania and Venezuela).
5.4 Missing data
Data used for panel models form an unbalanced panel, meaning that differ- ent countries contribute by different amount of observations. In theory, there are three ways of how the data can be missing. The best case is ”missing completely at random” (MCAR). It occurs when the missingness of data- points is completely unrelated to information in the dataset. In that case any statistical inference from that data is unbiased and the missing data can be safely ignored (Grilliches, 1986).
The second best case is ”missing at random” (MAR) - perhaps slightly confusingly, it occurs when data are not missing at random, but the reason why they are missing can be explained by those data where there is com- plete information. If the missingness of particular datapoints was related to information in them (e.g. if countries with GDP under some threshold did not supply data), it would mean that data are ”missing not at random”
(MNAR). That is the worst case, because then it would cause biased results of estimation on such data.
It is easy to check if data are MCAR - for example by Little’s MCAR test
2Original data are in units of dollars, not in thousands. However, working with units can cause prob- lems for some statistical softwares, which tend to replace very small numbers with 0. This replacement then causes some estimation methods, namely fixed effects, to fail. Stata does not seem to suffer from this problem, but R does.
(Little, 1988). In case of data used in this thesis it is not necessary, because from a glance at the data it is clear that missingness is correlated with the year of that observation - the oldest (1991 - 1994) and the newest (2015 - 2017) data are missing much more often than data in the middle of ex- amined period. Fortunately there does not seem to be any clear relationship between missingness of values and the values themselves. Phenomenons that were searched for include: existence of ”cutoff points”, where values above or below a certain value would be missing, relation between GDP in years when there were available data and missingness in other years and relation between Gini coefficient in years with available data and missingness.
No patterns were found - data for all variables cover wide range of values, poorer countries report data of similar quality (in terms of missingness) as rich countries and no relation between income inequality and data availab- ility was detected. The only irregularities that have been spotted are those linked to events happening in certain countries - for example, no data about income inequality in Iraq are available from the period of Saddam Hussein’s rule, but immediately after he has been ousted in 2006 all data became available. No data are also available for Libya since the beginning of the civil war in 2011. However, such occurrences are rare and it may be even beneficial that they are missing, as the effects of revolutions and wars on income inequality may be incorrectly attributed to explanatory variables in the panel data model.3
There is no statistical test that would distinguish between MAR and MNAR - it is impossible to do that just from the available data (the differ- ence lies in those datapoints that are missing), but because there is no clear pattern why certain data are missing except for time, the data can be safely considered to be MAR and therefore the results of the estimation is not expected to be biased. If measurement for any variable in a certain country
3Not all periods of turmoil lead to missing data, for example Venezuela diligently reports data for all variables of interest of this thesis during all hardships that the country experienced during last 30 years.
If there is any effect of such events on income inequality, it will appear in the regression as idiosyncratic error.
in a certain year is missing, then this datapoint is omitted from estimation, even if other variables are not missing.
6 Relationship between SDG target and GINI
As it has been said in the section about the SDGs, when evaluating fulfill- ment of the SDG: Reduced inequalities, the UN’s choice of first indicator is Growth rates of household expenditure or income per capita among the bot- tom 40 per cent of the population and the total population. Unfortunately, comparing income of bottom two quintiles of population with the whole pop- ulation was not a very common way of income inequality measurement until 2015, so most of developing and least developed countries did not start to collect data about income distribution by quintiles until this goal was made public (or do not collect it altogether). Available data about this variable are not sufficient for performing any meaningful analysis that would be only based on them. Therefore, some other variable that would describe income inequality is needed.
The most used measurement of income inequality is the Gini coefficient.
It was developed as early as 1912 by Corrado Gini and is defined as the ratio between area under Lorenz curve and area under equality line in the plot of total income earned cumulatively (on y axis) by a share of population (on x axis) times 100. As it turns out, there is a strong relationship between Gini coefficient and the UN’s indicator (after all, they both measure income inequality). Figure 1 shows a scatter plot of datapoints from those countries where the values of income share of lower two quintiles of population was available and 3 shows results of simple OLS regression, where the depend- ent variable is the UN’s indicator and the explanatory variable is the Gini coefficient. There seems to be clear indication that these two measurements are highly (negatively) correlated, so it can be said that on average, a one percent increase in Gini coefficient corresponds to 0.386 percent decrease in income share of lower 40% of population.
Figure 1: Scatterplot of Ginis and income of bottom 40% (in percentages)
Furthermore, given that the goal stated in the 10th SDG states that the condition of fulfillment is only the increase in lower two quantiles’ share of income, the approximate of the indicator by the Gini coefficient is even more appropriate, because for the evaluation of the indicator, a stable ratio Gini:
share of lower 40% is not needed - just the information that if one of those variables increases, the other decreases is sufficient to evaluate which way a country is heading in terms of fulfillment of the UN’s indicator. Therefore, the possibility of a evaluation mistake caused by replacing the UN’s indicator by Gini coefficient is even smaller, so in models in this thesis the replacement is performed.
Table 3: Relationship between SDG indicator 10.1 and Gini
Dependent variable:
income of lowest two quintiles Gini −0.386∗∗∗
(0.010)
Constant 43.880∗∗∗
(0.486)
Observations 783
R2 0.645
Note: ∗p<0.1;∗∗p<0.05;∗∗∗p<0.01
7 Empirical panel data model
I estimate two separate models: one model for the time period 1991 - 2007 and second one for the years 2008 - 2017. Since the WorldBank Database provides data about share of agricultural workers only from 1991 on, the earlier period starts at that year.
There are several reasons why I chose to distinguish between earlier and later periods. As can be seen in Table 4 and Table 5, coefficients for variables that this thesis is interested in significantly differ between those models. The exact position of the break in time is arbitrary to some degree, but it is at some timepoint around the end of the first decade of 21st century.
The first - and the most obvious - reason for the separation into two models is the global financial crisis beginning in 2008. There is evidence that some countries pursued policies which had disproportionally affected the poor, while other (e.g. Viet Nam) started to massively redistribute wealth (Fiorio and Saget, 2010, Mendoza, 2011)4.
4Even though Fiorio and Saget analyze only the UK and the USA, the situation in developing countries is likely to be the same or worse, given the general vulnerability of poor to shocks to the economy. All transmission mechanisms of inequality mentioned in the paper are present in developing countries too.
Second reason for splitting the models is the spread of new communica- tion methods to most of developing and least developed countries. Life of inhabitants of the most remote areas of those countries did not change, but in a large parts of developing countries the Internet became available shortly before 2010 or shortly after (according to the data provided by WorldBank database, only 8,3% of inhabitants of developing countries had access to the Internet in 2004, but the share increased to 15,4% in 2008 and 25,7%
in 2012 ). This new phenomenon could have negated a part of cities’ ad- vantage against rural areas, since the Internet access was spreading to the countryside instead of being available only in big cities.
It should also be mentioned that oil prices were on average substantially lower in years 1991 - 2007 than from the year 2008 on, as can be seen in Figure 2. It cannot be said that oil prices should influence inequality in developing and least developed countries upwards or downwards - it affects many variables and its effects on economy is complex and varies country to country, but splitting the time period roughly into cheap oil phase and less cheap oil phase may reduce any omitted variable bias caused by possible effect of oil prices on inequality.
The models differ only by the time period. Both models are specified as Ginii,t =α+β1urbani,t+β2educi,t+β3agrii,t+
+β4GDPi,t+β5(GDPi,t)2+ci+ui,t
where Gini represents pre-tax pre-transfer Gini coefficient, urban denotes percentage of urban population,educstands for net enrollment for secondary education in percent,agri represents percentage of people whose main job is in agriculture and GDP is self-explanatory. ci is unobserved heterogeneity and ui,t is the idiosyncratic error.
These variables are selected mainly to test if the effects described by Kuznets (1955) and Vanhoudt (2000) - inverted U-shaped relationship between income and income inequality - are present in contemporary period, too. If it was indeed true and income inequality would tend to decrease with total
to fulfilling the 10th SDG by 2030 just by reaching that threshold and then maintaining economic growth. Secondary hypothesis to test is that human capital from Vanhoudt’s model (expressed by enrollment to secondary edu- cation) indeed does significantly contribute to decrease of inequality.
Figure 2: Inflation-adjusted prices of crude oil in USD
8 Results and their interpretation
8.1 Earlier period
Table 4 summarizes the results of regression obtained by pooled OLS, FD, RE and FE estimation for the earlier period (1991 - 2007). In parentheses are robust standard errors, because serial correlation has been detected.
Heteroskedasticity has not been detected, but given that the time-dimension of the panel is so short, the Breusch-Pagan test may not be considered strong enough to rule out its presence. The Hausman test rejects its null hypothesis (that both RE and FE are consistent) very strongly (at p-value 10−15), so its results support FE estimation, rather than RE. However, this test does not use robust standard errors, so the serial correlation and possible heteroskedasticity may influence its results. Therefore, it cannot be clearly said that one method is definitely better than the other based on statistical inference.
Table 4: Period 1991 - 2007, estimated coefficients
Dependent variable:
Gini mkt
(Pooled OLS) (FD) (RE) (FE)
urban −0.036∗ 0.034 0.033 0.021
(0.021) (0.029) (0.025) (0.029)
educ −0.127∗∗∗ −0.0003 −0.042∗∗∗ −0.037∗∗∗
(0.021) (0.007) (0.009) (0.010)
agri −0.237∗∗∗ 0.025∗ 0.024∗ 0.039∗∗∗
(0.018) (0.013) (0.013) (0.013)
GDP −0.263∗∗ 0.054 0.179∗∗∗ 0.0002∗∗
(0.115) (0.041) (0.068) (0.0001)
GDP2 0.003 −0.0001 −0.001 −0.000
(0.002) (0.001) (0.001) (0.000)
Constant 64.372∗∗∗ −0.099∗∗∗ 44.808∗∗∗
(1.510) (0.022) (1.471)
Observations 423 353 423 423
R2 0.308 0.021 0.515 0.127
Note: ∗p<0.1;∗∗p<0.05;∗∗∗p<0.01
Perhaps somewhat surprisingly, the variable urban proved to be stat- istically insignificant at p-value 0,05 in all methods of estimation. It is contrasting with the findings presented by Cobbinah, Erdiaw-Kwaise and Amoateng - according to them, increasing share of urban population in de- veloping countries should increase income inequality. Results of estimation of model in this thesis do not show any evidence of such phenomenon. A possible explanation may be that majority of the change in urbanization is caused by immigration into smaller cities or orderly neighbourhoods of large cities, not into slums, and therefore overall income inequality is not very affected by the existence of slums (which are very noticeable and negative social phenomena are rife there, but there is not a large amount of them compared to the whole urban population.)
The most suitable method, FE estimation, shows that the variable agri seem to have small, but significant, positive effect on income inequality. This is suggesting that citizens’ job, rather than place of living (urban or rural), affects inequality level in a country. Wages in agricultural sector are gener- ally low in developing countries, often only slightly above minimum wage (as Bhorat, Kambur and Stanwix (2014) show on the example of South Africa), so countries with many people employed in agriculture tend to also have large amount of impoverished citizens. In context of Kuznets’ theory, such results suggest that today’s developing and least developed countries are already on the downward-sloping part of the Kuznets curve. If we consider Vanhoudt’s model accurate, then there are enough skilled workers in the developing countries that approaching steady state causes income inequality to decrease.
As expected, the variable educ turned out to be significantly negative.
This is in line with wide consensus and empirical evidence5 that education generally (and secondary education specifically) reduces income inequality.
In context of Vanhoudt’s model (see Literature review) that means that
5For example Abdullah, Doucouliagos and Manning (2015), who performed meta-regression analysis on 64 studies on this topic, or Wasim Qasi et al. (2016), who provides a concrete example by analyzing the relation between income inequality and school enrollment in Pakistan.
the amount of ”skilled” workers in economy is large enough that education decreases inequality instead of increasing it. Nevertheless, the effect of sec- ondary education is smaller than it may be expected - a 10% increase in net enrollment results in decrease of only 0,37 points of Gini coefficient, so education is clearly not an universal solution for excessive income inequality.
It is likely that secondary education resembles a double-edged sword - while it can help children from poor families escape poverty by giving them an opportunity to find a good job, it may deepen already existing social strat- ification if more students from relatively richer families have access to the schooling.
In contrast with Kuznets’ hypothesis about inverted U-shaped curve de- scribing the relation of total income and income inequality, no such phe- nomenon has been found for the earlier period.
8.2 Later period
Results of regression on data since 2008 differ in two main aspects, as can be seen in 5. The most striking difference seems to be the reversal of the sign of coefficient in the variableagri. Again, the effect is not huge, but now it is negative, meaning that countries with large share of people employed in agriculture now show smaller income inequality. Timewise it correlates with the increase in developed countries’ interest in the wellbeing of farmers in developing countries (either in the form of international organizations’
help, such as programmes of the SDG Fund, or in the form of NGO efforts, such as the Fair Trade arrangement.6 Of course, correlation does not imply causation, but it could be a hint that the programmes may actually work.
The second outstanding difference is that in the later period a significant effect of GDP appeared. GDP was found to have negative, slightly con- vex effect on inequality (the results show that the point where the positive quadratic term prevails over the negative linear term is at about 38 500 USD per capita. No developing or least developed country except Singapore and Qatar reached so high GDP.)
Like in previous period, variable urban is not significant on p-value 0.05.
6Efectiveness of SDG Fund programmes has been shown on certain specific cases, e.g. the programme promoting production of quinoa in Peru (SDG Fund, 2018), but to exactly evaluate the overall effect of those programmes, further research is needed).
Table 5: Period 2008 - 2017, estimated coefficients
Dependent variable:
gini mkt
(Pooled OLS) (FD) (RE) (FE)
urban −0.004 0.055 0.016 0.054
(0.023) (0.060) (0.032) (0.046)
agri −0.140∗∗∗ −0.005 −0.086∗∗∗ −0.045∗ (0.023) (0.018) (0.023) (0.024)
educ −0.074∗∗∗ 0.006 −0.057∗∗∗ −0.027∗ (0.021) (0.011) (0.014) (0.014)
GDP −0.032 −0.661∗∗∗ −0.478∗∗∗ −0.693∗∗∗
(0.095) (0.109) (0.114) (0.130)
GDP2 −0.0004 0.009∗∗∗ 0.006∗∗∗ 0.009∗∗∗
(0.002) (0.002) (0.002) (0.003)
Constant 54.344∗∗∗ −0.154∗∗∗ 54.927∗∗∗
(2.061) (0.037) (2.227)
Observations 372 300 372 372
R2 0.130 0.124 0.634 0.315
Note: ∗p<0.1;∗∗p<0.05;∗∗∗p<0.01
8.3 ARIMA forecast
In previous section of this thesis, panel data analysis was used to assess which variables affect the level of income inequality in developing countries.
Nevertheless, the issue is too complex to forecast future development only based on these variables. Generally low values of R2 for all models indicate that a large part of the data-generating process for income inequality lies elsewhere than in these four variables. Therefore, ARIMA forecast method was selected to approximate future income inequality in selected developing countries. Detailed description was provided in section 4.
One country from each continent where developing countries are located was selected for forecast as a representative: Tanzania from Africa, South Korea from Asia and Mexico from Latin America.7 The results of the fore- cast can be seen in 3 (Tanzania), 4 (South Korea) and 5 (Mexico). It should be noted that forecasting for longer period, especially for the year 2030 (the expected end year of Sustainable Development Goals), results in very wide confidence intervals with almost no predictive value. Because of this, the figures show forecast for 6 years from the latest datapoint available.
If no major change happens, the forecast suggests that income inequality in both Tanzania and South Korea will probably slightly increase. Therefore, these two countries do not seem to be on the right track to fulfill their part of 10th SDG.
While Tanzania managed to achieve spectacular GDP growth especially in the last 15 years (according to data from WorldBank database, GDP per capita more than doubled between 2005 and 2020), Gini coefficient only stabilized at around 40. In a few years it would become clear if it was a first step of reversing the unfavourable trend which started as early as 1970s, so Tanzania can report compliance with the goal in 2030, or the trend prevails.
7The choice was driven partly by how big the populations of those countries are (all of them have over 50 million inhabitants in 2020), partly by absence of any significant turmoil during last 50 years and partly by availability of the data about inequality - the SWIID database provides uninterrupted yearly data at least from 1970 for all three. To some extent they represent their whole continent, especially time series for Latin American countries strongly resemble the Mexican one. Asia and Africa are a bit more diverse.
Kinyondo and Pelizzo (2018) analyzed the causes of income inequality in Tanzania very thoroughly and recommend enacting of minimum wage and focusing on decreasing inflation, so income inequality could be decreased.
In time series concerning South Korea two periods of sharp increase in income inequality can be seen. The first corresponds with Asian financial crisis in late 1990s, the second with the subprime mortgage crisis of 2008.
According to Kwang-Yeong and Ju (2014), that is no coincidence: the in- crease in income inequality is fueled by neo-liberal labour market reforms enacted in response to those crises in combination with rapid ageing of South Korean population and changes in family structure (favouring Western-style small families instead of large multigenerational clan-like structures). These trends are embedded deep in Korean society, so it is hard to assume they will go away in next decade. 8 Absolute levels of income inequality in South Korea are not as high as in most African countries or even in Latin Amer- ica, but the trend is certainly unfavourable, as both ARIMA forecast and Kwang-Yeong and Ju’s paper show. South Korea is one of developing coun- tries for which the successful fulfillment of 10th SDG is the most unlikely.
The case of Mexico is different from the other two - the forecast for this country is slighly favourable. Of course it must be borne in mind that the absolute level of Gini coefficient is much larger in Mexico than in two other compared countries, and it may be easier to get progressively better if the inequality was higher to begin with. Mexico is currently making good progress toward fulfillment of the goal, but given that income inequality is notoriously high in Latin America, including Mexico, faster progress would definitely benefit the poorer 40% of Mexicans. (If Mexico continued to de- crease Gini coefficient by the same speed as in last 10 years, it would reach the level of today’s Tanzania approximately in 2050 and today’s South Korea around 2080.)
8The labour market laws would be easier to change than aging of the population, but another landslide victory of the neo-liberal Democratic Party of Korea in the 2020 general election does not suggest that
Figure 3: GINI coefficient in Tanzania
Figure 4: GINI coefficient in South Korea
Figure 5: GINI coefficient in Mexico
