Inequalities inequalities B Milanovic June-July 2004 1

Inequalities, inequalities…. . B. Milanovic, June-July 2004 1. Briefly: the three concepts 2. Concept 1 and Concept 2 inequality 3. What happened to global inequality: new 1998 data 4. Openness and within-country inequality

Sources • Book: Worlds Apart: Global and International Inequality, 1950 -2000 coming out in May 2005 (Princeton University Press) ORDER IT NOW!! • Website: www. worldbank. org/research/inequality • Email: bmilanovic@ceip. org

1. Definitions Different types of inequality

How they differ?

2. Inequality between countries Coverage: number of countries and share of world population

·About 140 countries included; about 6200 country/year GDPs ·almost 100 percent of world population and world GDP (in current dollars) ·current countries projected backward (NEW) SIMA World Bank data used to get benchmark 1995 $PPP GDP per capita; then these GDP per capita projected backward and forward using countries’ real growth rates (78% of data from WB sources; others mostly from national SYs; some from Penn. World Tables, UN sources)

According to Concept 1, countries' performances have diverged over the last two decades Unweighted inter-national inequality, 1950 to 2000 And it is not only because Africa is falling behind

Downwardly mobile world The Four Worlds defined The Rich: All countries with the GDP per capita equal/greater than the poorest WENAO The Contenders: With GDP per capita at least 2/3 of the poorest WENAO ( they can catch up within a generation) The Third World: With GDP per capita between 1/3 and 2/3 of the poorest WENAO The Fourth World: With GDP per capita less than 1/3 of the poorest WENAO

Overall upward and downward mobility 1960 -78 and 1978 -2000 1960 -78

The border countries and their GDP per capita levels (in $PPP, 1995 prices)

Why Concept 1 inequality matters • Are poor countries catching up as we would expect from theory? • Are similar policies producing the same effects or not? (Rodrik: convergence of policies, divergence of outcomes). Why? • Migration issues • Countries are not only interchangeable individuals (random assortments of individuals); they are cultures. Divergence in outcomes is elimination of some cultures. Perhaps it’s good, perhaps not.

3. Moving to Concept 2: its relevance and irrelevance • Once we have Concepts 1 & 3, Concept 2 is redundant. • But we have imperfect grasp of Concept 3 inequality => Concept 2 provides a check on “true” inequality (its lower bound) • We use it to approximate “true” inequality. Think, at the limit, of each individual being a country

The mother of all inequality disputes

How are Concepts 2 and 3 related? • In Gini terms: • where Gi=individual country Gini, π=income share, yi = country income, pi = popul. share, μ=overall mean income, n = number of countries • In Theil:

Inequality between population-weighted countries According to Concept 2, there is convergence among countries…

. . . or maybe there is not

Alternative Concept 2 calculations • Alternative growth rates for China (official-World Bank, Maddison, Penn World Tables) • Breaking China, India, US, Indonesia and Brazil into states/provinces (but redistribution within nations) • Breaking China into rural and urban parts (Kanbur -Zhang data set) • What PPP to use (Geary-Khamis, EKS, Afriat)

Implied China’s GDP per capita in different years According to different sources PWT 6. 1 Maddison World Bank 1952 568 627 262 1960 662 785 497 1966 773 879 534 1978 899 1142 754 1988 1703 2119 1676 1999 3319 3803 3867 2000 3642 na 4144

Concept 2 inequality for different versions of China’s GDP per capita

…and breaking China and India into their provinces/states. Inter-national population-weighted inequality: with China and India replaced by their provinces and states

How much has Concept 2 inequality changed (Gini points; 1985 -00)?

• . 4 Distribution of ln. GDPPP pc in 85 and 00 (WB numbers; states, R/U for China) • 0 • . 1 • kdensity lngdp • . 2 • . 3 2000 • 6 • 7 • 8 • kdensity lngdp • x • 9 • 10 • kdensity lngdp Gini: 60 (in 1985); 57 (in 2000) • 11

Distribution of ln. GDPPPP per capita; provinces/states and countries (1985, 2000)

Concept 2 between 1980 and 2000 Contributes to decline Reverses decline (equilibrating factors) (disequilibrating factors) • Inclusion of provinces/states of • Higher (old) income China, India, Brazil, level in China Indonesia, US (even if (Maddison) 1. 5 points many within • Inclusion of themselves are rural/urban break up of diverging!) 0. 5 point China 0. 5 points Result: we shave off half of the Concept 2 decline

4. Global inequality Number of income and expenditure-based surveys by region Note: “Expenditure” or “consumption” survey is used inter-changeably. Common sample: 86 countries.

Coverage: around 90% for population, 96% for GDP

What does Concept 3 say? World international dollar inequality in 1988 and 1993 (distribution of persons by $PPP and $ income per capita) Note: Gini standard errors given between brackets.

Decomposition of global income inequality, 1988 -1998 (common-sample countries; distribution of persons by income/expenditure per capita)

What explains the 1988 -93 increase in inequality? Key changes in inter-country terms between 1988 and 1993 (in Gini points) • • • Slow growth of rural incomes in populous Asian countries compared to rich OECD countries. The pulling ahead of urban China vis-à-vis rural China and India. The urban-rural ratio increased by a half in China and it went up in India too. The mean rank of population in urban China increased from 53 rd percentile to 62 nd while the mean ranks of populations in rural India and China stayed within 1 and 2 percentile of 1988. The “hollowing out” of the world’s middle class—a problem with Latin America and Eastern Europe since respectively the early 1980’s and early 1990’s; eg. the mean income rank of Russian population decreased from 80 th to 73 rd percentile.

And what explains the 1993 -98 decrease in inequality? Key changes in inter-country terms between 1993 and 1998 (in Gini points) The three factors that all worked toward increasing inequality between 1988 and 1993, behaved very differently over the next five-year period. • One of them (rising income distance between rural and urban areas in China and India), continued almost unabated. • Another—income distance between rural India and China and the West— reversed, contributing to inequality decrease. • And the third, the crisis in transition countries, moderated, and basically no longer affected world inequality very much.

The key determinants of global inequality Interaction between 1. the rich countries of the West, 2. urban incomes in China and India 3. rural incomes in these two countries The ratio between (2) and (3) has been rising, and is unlikely to moderate. Moreover, while China and India are the most important examples of the trend, the urban-rural gap is rising in several other Asian countries (Bangladesh, Indonesia, Thailand). But as (2) catches on (1), world inequality is reduced. The crucial “swing” factor then becomes the ratio between (3) and (1): what happens to rural incomes in China and India vs. incomes of the rich world. If the former catch up, world inequality goes down; if they do not, world inequality tends to rise.

SENSITIVITY ANALYSIS FOR CONCEPT 3 The three building blocks for the calculations… ● national distributions available from Household surveys, ● mean incomes again available from Household surveys or from National accounts (GDP per capita) and ● PPP exchange rates. …and problems with each of them

How Concept 3 changes with various definitions: what we expect A>B because HS/NA decreases in GDP (Little doubt that HS data should be used in distribution analysis, as it is done for individual countries; but people have recently used GDP per capita) C>B because of increasing rural/urban disparity

Would conclusions change if we used GDPs per capita instead of HS means? Yes to some extent. The increase in 1988 -93 is less sharp. World income inequality in 1988, 1993 and 1998 (common-sample countries; $PPP) IMPORTANT NOTE: Relatively high HS/NA ratios for China and India in 1988 (compared to later years); explain relatively low global inequality in 1988

Problems with the use of GDP • Property incomes chronically underreported in HS are allocated to all (but the poor hardly receive any of them!) • So are other parts of GDP: undisbursed profits, build-up of inventiories (all corporate income that is not distributed to HHs) • Underreporting and undersurveying is strong among the rich; but the entire difference btw. GDP and HBS mean is allocated to all (in proportion to their income) • Distribution of social transfers (H&E) is often prorich too

• Basically, the problem is that the gap between GDP and household mean is caused by underestimating top incomes • But that gap is allocated ACROSS the entire income distribution

And what happens if we do not change PPPs? That is, use CPIs to convert all incomes, express them all in domestic prices of 1988 and then use the 1988 PPPs. A milder increase between 1988 and 1993, and practically unchanged between 1998 and 1993 (or even continued increased if measured by the Theil index) Global inequality calculated using 1988 PPPs and incomes expressed in 1988 domestic prices (full-sample countries) Standard errors between brackets.

My preferred measure (Gini with R/U split + HS data) with 95% confidence interval

Everyone agrees global inequality is high--but recent trend is a matter of dispute

Intuitively, what is a Gini of 64 -66; how big is it? Is it “grotesquely high” as UNDP Human Development Report says? Or is it that we don’t know what “optimal inequality” is (as an economist said).

"Inequality transition"? • Lucas and Firebaugh view: global inequality has peaked; Why? • Permanent effects of industrial revolution • Policy convergence => Income convergence • Historically, Concept 2 inequality drove global inequality since IR; for the last 30+ years has been on the decline; then Concept 3 must follow.

But. . . • Technological revolution continues (not only one discrete big bang…), differences may be accentuated (speed of tech. inventions = > speed of dissemination) • Policy convegence did not result in income convergence • And all hangs on the break in Concept 2 trend which depends on one country and one particular set of growth numbers for it.

Poor, middle class and rich in the world According to three inequality concepts Note: Columns (1) and (2) based on GDP per capita. Column (3) based on data from household surveys (full sample; 122 countries in 1998). Brazil and Portugal always included in the higher group (respectively middle-income and rich).

Poor people in poor countries? How many are they? Almost 4 billion. Rich people in rich countries? About 700 million. Poor people in rich countries; rich people in poor countries? About a hundred million each. Brings us to almost 5 billion people? So, where is the middle? Correspondence between poor countries and poor people in the world (in million people; 1998; household survey data) Note: Full sample countries (122 countries). Poor below mean income of Brazil, or social assistance eligibility in the West (about $PPP 10 per capita per day.

5. Openness and within-country inequality Decile shares: deciles from 321 surveys, 1988 -1998 (SURE and GMM/IV estimates; openness and govt exp. instrumented) Equations (for 10 deciles and more than 100 countries, 1988, 1993, 1998) • j yij = income of i-th decile of j-th country in constant PPP dollars, mj = mean household survey income of j-th country OPEN = (exports and imports)/GDP DFI = (direct foreign investment)/GDP FD = financial depth = (M 2/GDP) DEM = democracy RINT = real rate of interest EXP = government expenditures/GDP ln. INF=rate of inflation

A look at the data: decile shares (unweighted)

The openness variable (trade/GDP in %)

Results • At very low level of GDP per capita, openness reduces the share of the bottom 7 deciles, increases the share of the top two • The negative effect of openness is lessened for richer countries as the interaction term between openness and mean income is positive • Turning point. At income level of Spain and Israel, there is a reversal: openness increases relative share of the bottom and middle deciles; the share of the top declines.

Average level of income and change in income share by decile • Calculated at average decile shares (bottom 2. 6%; second 4%, top 30%), Δtrade=0. 2.

Further results • Higher real rate of interest reduces the share of bottom 8 deciles, increases the share of the top two • How strong the effect: each percentage point increases income of the rich by 1/3 of a percent • Democracy increases the share of the middle income deciles (4 th to 8 th), reduces the share of the top • Inflation reduces the shares of bottom and middle deciles

• Government expenditures raise the shares of the bottom groups. A 10% increase in G raises the share of the bottom decile by 0. 2 percent (1/10 of its income) • Financial deepening raises the share of the bottom groups • Only the middle income countries behave as we would expect from neoclassical theory (in poor countries, openness Gini, in rich it Gini)

Why Dollar & Kraay do not find the effect of openness on inequality? • Largely, because they use a wrong measure: trade/GDP in PPP terms • It is a wrong measure for poor countries if we want to study inequality • Chinese barbers' output is assessed at international prices, but they are paid CHINESE wages (not internatonal) • So for inequality, the latter matters!

Measurement of openness USA Three measures of openness: 1) trade/GDP in current $ 2) trade/GDP in constant $ 3) trade/GDP in constant $PPP

But in poor countries, hardly any increase in openness according to the third (PPP) measure India China Wrong measure for the study of inequality!

To wrap up…moving towards an integrated approach to growth and distribution? • Connection between endogeneous or neo-classical growth theories; convergence literature; Quah’s “twin peaks”; transition matrices; global inequality… • If we know that determine growth and withinnational distribution => we have distribution of countries by their Ypc and distribution of people • Global distribution of national incomes (Concept 2) or people (Concept 3) is directly derived

But we should not expect a “grand theory” and let’s end up with Tocqueville’s warning… I hate. . . these absolute systems which make all the events in history depend on primary causes, linking one to another by an inevitable chain, and which, so to speak, take out people from the general history of mankind. I find them narrow in their pretended grandeur, and false under their guise of mathematical truths. I believe, whatever the view of the writers who have invented these sublime theories to nourish their own vanity and to facilitate their work, that many of the important historical facts cannot be explained but by the accidental circumstances, and that many others remain inexplicable. And that finally, chance, or rather that mixing of the secondary causes, which we thus call, since we do not know how to tell them apart, explain a lot of what we see on the world stage. But I strongly believe that chance does not do anything which is not prepared in advance. The existing reality, nature of the institutions, state of mind of people, customs, are the raw materials with which chance constructs the facts which surprise and awe us.