Wps 1q35 POLICY RESEARCH WORKING PAPER 1935 Explaining the Increase Since the beginning of transition to market economy, in Inequality during inequaiity has increased in all the Transition transition countries. The factors driving inequality up: increasing wage inequality (as Braniko Milanzotvic workers move from a relatively egalitarian state sector to a less equal private sector), and the rising share of income from self- employment and property (both very unequally distributed). Social transfers have failed to dampen the increase in inequality because they have remained, as under socialism, unfocused. The World Bank Development Research Group June 1998 POLICY RE.SEARCH WORKING PAPER 1935 Summary findings The transition from planned to market economy has unequal sources of income both before the transition and witnessed one of the biggest and fastest increases in now. In addition., some of the released state sector inequality ever recorded. On average, inequality in workers remain unemployed. Their incomes decline. Eastern Europe and the former Soviet Union increased Increased inequality is thus accompanied by the from a Gini coefficient of 25-28 (below the OECD "hollowing out" of the middle class (where the middle average) to 35-38 (above OECD average) in less than 10 class is defined as the former state sector workers). One years. In some countries, such as Bulgaria, Russia, and part of state sector workers moves to higher incomes as Ukraine, the increase in inequality has been even more workers in the private sector or entrepreneurs; another dramatic, outpacing the yearly speed of Gini increase in remains jobless. the United Kingdom and the United States in the 1980s The model is contrasted with the actual developments by three to four times. in six transition economies: Bulgaria (over 1989-95), What are the factors pushing inequality up? Milanovic Hungary (1987-93), Latvia (1989-96), Poland (1987- constructs a simple model of transition defined as the 95), Russia (1989-94), and Slovenia (1987-95). In all removal of restriction on private sector development. As countries, wage inequality has increased (in some, like the private sector becomes free, it attracts workers who Russia, dramatically); income from self-employment has leave the shrinking state sector. Wage inequality in the remained as unequal as before but its share in total private sector is greater than in the old, relatively income has risen, and the importance of social transfers egalitarian state sector. This is one of the forces pushing in total income has increased, but its focus on the poor inequality up. The second is the growth of income from has not improvec. self-employment and property, both of which are fairly This paper - a product of the Development Economics Research Group - is part of a larger effort in the group to study social issues in transition economies. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Grace Evans, room MC3-568, telephone 202-458-5734, fax 202-522-1153, Internet address gevans@worldbank.org. The author may be contacted at bmilanovic@worldbank.org. June 1998. (41 pages) ,The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center EXPLAINING THE INCREASE IN INEQUALITY DURING THE TRANSITION' Branko Milanovic World Bank, Policy Research Department Introduction To explain the change in inequality that has occurred in all transition economies,2 we need first to present the economic changes in terms of what happened to incomes of different social classes, and then to translate this into the 'language" of personal incomes. This is, of course, a common problem of moving from social classes and thus factoral income distribution to personal income distribution (for a recent overvlew see Atkinson, 1995). The 'translation" is cumbersome but necessary if we want to combine (i) a discussion of economic forces that shape income distribution with (ii) how these forces 'played themselves out"in the arena of personal income distribution. To look into (i) we have to use the standard economist's 'Yit of tools"which is designed to study the behavior of individuals who are either workers or capitalists or fanners or transfer-recipients. In doing this we 'ag"individuals by assuming that they have no other sources of income but one. This is the approach I use in Sections 1 and 2 where I present first, a simple model of factoral income distribution during the transition, and second, a numerical illustration of change in inequality. 'Tagging"is, of course, wrong: a capital-owner may (and presently often does) work; a workman may own stocks and engage in entrepreneurial activity; a pensioner may work part-time or lease some of his assets. Thus when we move to personal income distribution, the point of view changes: we no longer 'tag" individuals but simply add all their sources of income, adjust total income for personal and household characteristics (household size, age of children etc.), and 1 The findings, interpretations, and conclusions expressed in this paper are entirely those of the author. They do not necessarily represent the views ofthe 'World Bank, its Executive Directors, or the countries they represent. I am gatel to Jan Ratkowski for vely helpfi.l comments. 2 See Milanovic (1998, Chapter 4). :1 then rank everybody according to his/her adjusted household income. People are no longer workers or capitalists or pensioners; they just have incomes, some of which may be in the form of wages, or profit, or pensions. In terms of disaggregation of an index of inequality we go from the disaggregation by recipients to disaggregation by income sources. The latter is the point of view I adopt in Section 3 where I present empirical evidence on income inequality and factors that were responsible for its increase. In the concluding Section 4, I contrast the conclusions drawn from the empirical evidence with the conclusions drawn from the model and numerical illustrations in Sections 1 and 2. 1. Change in sector shares and factor incomes Consider a following simple model of an economy before and after the transition. Pre-transition. The economy is composed of two sectors. First, a small private sector whose size is limited through various legal restrictions (e.g. a limit on the number of workers that can be employed; no access to credit; banned sectors of operation etc.); and second, a large state sector. The private sector (consisting mostly of the self-employed) maximizes the average product per worker (aplp)3 where subscript p denotes private sector. It employs OP workers (Figure 1). Since its growth is limited by legal barriers, it is not allowed to maximize the aplp curve. The average income in private sector is therefore y. Since there is no unemployment, the state sector must employ the rest of the labor force (NP)4 and its wage (ws) is established at the point where the demand for labor (mpls) intersects the vertical P line.5 Normally, we would ' All per capita magnitudes are written in lower case. 4The state sector share should be read from right to left. 5A more usual assumption (see e.g. Blanchard and Keeling, 1996; Commander, Tolstopiatenko and Yemtsov, 1997, Appendix 1) has been to assume that the private sector equalizes marginal product of labor and the wage, while the state sector maximizes the average product of labor. This was rationalized by arguing that the private sector was a 'honnal"private sector, and the state sector was labor-managed. However, this approach fails to acknowledge that almost the entire private sector before the transition was self-employed or cooperative, and that average product was therefore a better maximand. Further, only in Yugoslavia, and to some extent in Poland, could the old state sector be described as labor- managed. 2 expect that since ws1 (see 9Thbe entire private (self-employment) income before the transition is denoted by Y. After the transition, we still denote the entire private sector income by Y; however, it is now equal to Wp (private sector wages) plus B (capitalists' profit). '7 Figure 1), and y<1 (since average pension is a fraction of average wage), the overall inequality before the transition becomes GINI = Gyty py+ - Ws [(,B-1) pypw + (y-f3) pypt + (I-y) pwpt] + L The typical population shares before the transition were approximately 60% of household heads employed in the state sector, 20% self-employed or employed in the private sector, 20% pensioners. Assuming further that the average pension was 1/2 of the average wage (y-=0.5) and average self- employment income 3/2 of the average wage (,B-1.5),10 the GlIN becomes (5) GINI=0.06Gy+0.06+0.04+0.06+L_0.18+L where we have assumed, based on some empirical evidence, that Gy=0.3. L (overlapping) inequality can stem only from the self-employed who might "overlap"with other social groups. This is so because all pensioners and all state sector workers have respectively yws or ws: there can be no "overlapping"between them. Moreover, there is some evidence that the self- employed not only 'bverlapped"with some other recipients, but even bracketed them. The self-employed before the transition included both poor farmers, barely above the subsistence level, and rich entrepreneurs who, thanks to connections and luck were able to make large profits." These two extreme groups among the self-employed were particularly in evidence in countries that combined a large private agricultural sector divided in many small plots like Yugoslavia, Poland arnd Hungary, and also allowed some non-agricultural private enterprise. Equation (5) shows that under the ultra restrictive assumptions of equal wages and equal pensions (for respectively all state-sector workers and all pension recipients), the overall Gini in pre-transition countries would have been between 18 and 20, where this extra 'iudge"above GINI=1 8 comes from the presence of both relatively poor and relatively rich among the self-employed. 1O So that the overall average income ,u (0.6*1+0.2*0.5+0.2*1.5) is equal to the average wage. The latter group produced even a few instances of '%ocialist"millionaires like Jan Kulczyk, the founder of Inter Fragrances, a large Polish perfume exporter, or Emo Rubik of the Rubik cube fame in Humgaiy. For the evidence of an inverted U shape of the private sector share in disposable income, see Milanovic (1992, p. 21). 8 Gini of 20 is approximately the value rncorded in Czechoslovakia, the most egalitarian among former socialist countries. Supporting indirectly the ultra restrictive assumption of equal wages and pensions, it was argued with some plausibility by Vecernik (1986; 1993) that before the transition disposable household income in Czechoslovakia could have been well approximated by multiplying individual household demographic characteristics--number of the employed, number of pensioners and number of children--with constants (which are, by definition, equal for everybody).12 Implicitly, inequality of wage or pension distribution could be treated as negligible. If we now abandon the ultra restrictive assumptions of Gv=Gt=0, and replace Gv=0 with (the empirically-based) Gw=0.25;13 and similarly Gt=0 with the empirically-based Gt=0. 15 14 (while still ruling out overlapping between workers and pensioners: that is, still assuming that all workers have higher incomes than all pensioners), the GINI increases to about 27. With the overlap component, it would be about 28. Mosi. of the increase is due to inequality among wage earners (see Table 1). Indeed the Gini between 24 and 28 was the level observed in Poland, Romania, Slovenia, Hungary, Bulgaria and most of the republics of the former Soviet Union before the transition. 15 In conclusion, assuming 'ideal-typical"conditions of socialist economies before the transition and breaking down equation (4) we find that: (i) about 16 Gini points of inequality (see Table 1) was due to the differences in average incomes between the self-employed, workers in the state sector and pensioners; '2 Household income, approximated by the formula. co + cl*number of employed members + c2*number of pensioners + c3*number of children, could explain more than 60% of actual variation in Czechoslovak incomes (see Vecernik 1993). 3 Atkinson and Micklewright (1992, p.81) give the earnings' Ginis in 1986-87 as 20 in Czechoslovakia, 22 in Hungary, 24 in Poland, 28 in the USSR. Similarly, Redor (1992 [1988]) gives the following values: Hungary (1980) 21; Poland (1980) 23; Czechoslovakia (1979) 20; USSR (1964) 24. For more details see Figure 5 below. 14 Ginis of pensions (calculated across pension recipients) ranged in East European countries between 12 and 18 (World Bank PRDTE data base). 5 See Atkinson and Micklewright (1992, p.1 13); also Milanovic (1997, p.44). 9 (ii) about 9 Gini points was due to inequality among state-sector workers; (iii) about 2 Gini points was due to inequality among the self-employed and pensioners; (iv) and the rest (1-2 Gini points) was due to "overlapping." Table 1. Break-down of inequality before the transition Formula Value (in Gini points) (A) Inequality among state Gw rw pw 9 workers (B) Inequality among the self Gy 7ry py 1.8 employed (C) Inequality among pensioners Gt irt pt 0.3 (D) Differences in average P incomes ((y-ws) py pw +(y-t py pt + (Ws-t pwpt) -16 Total without overlapping [A to D ] ~27 Overlapping 1-2 points Total GINI about 28 Assumptions: pw=60/O, py=20%, pt20o%; Ws=--; pension/wage=0.5; self-employment income/wage=1.5; Gw=0.25; Gy=0.3; Gt=O.15. Hence, rw=60%/o, ini30%, 7Wt=10%, ±=I. Transition. How are these values affected by the transition? There are two factors driving inequality up: (i) the change in the composition of the employed; and (ii) the increase in inequality amongst both the private sector and state-sector workers. On the offsetting side, a movement of some workers from the state sector to transfer recipients (unemployment) where inequality is less, will tend to lower the overall GINI. First, from Section 1 we know that the shares of the private and state sector will change during the transition. The former will grow; the latter will decline. Since inequality in the private sector was originally greater (Gy-0.3 vs. Gw=0.25) and is likely to remain so, the shift in employment will tend to increase the overall GINI. (A qualifier 'tend"is needed because the relationship, as shown in footnote 16 below is more complicated.) Second, inequality in the new private sector (Gy) will rise.16 16 The evidence (see Rutkowski (1996) for Poland; Vecernik (1994a) for the Czech republic; Vodopivec and Orazem (1995) for Slovenia, and Rutkowski (1995) for Bulgaria) shows not only that private sector incomes become more unequally distributed (a thing which we might expect), but that state sector 10 Combining the 'hew class"of capitalists that has emerged during the transition with private-sector workers, and rewriting equation (4) where subscript y refers to both of them, while subscript t denotes now both the pensioners and the unemployed, (4) GN = Gy7ry py + Gw7cw pw + CGt t pt + + W [(O-1) pypw +(0-y) pypt + (I-y) pwptl + L Al we can 'isolate"the following causes of increased inequality: (i) Assume that Ah percentage of state sector workers 'bnoves"to the private sector. This reduces pw from 60% to (60-Ah)%, and increases py from 20% to (20+Ah)%. Since Gy>Gw, GINI will rise. 17 wages' inequality rises too. However, in order to keep matters relatively simple, I shall ignore the higher inequality of Ws. t7 If Ah is sufficiently small, the change in the non-overlap component of the GMNI will be 8GNI/6h =-(Gy y py - Gw ws pw) + - [(P-1) pw - (,-1) py + (,Y) pt - (1Y) pt]= 2ws w (GypBpy-Gwpw)+ - [(P-1) (pw - py) + (J-1) pt]= 2ws S(A (Gyp3py-Gwpw)+ - [(3-I) (pw - py+pt)] (4A) where we use the fact that 7ry= f3wspy/4 and 7xw= wspw/p. The sign of the first RHS term in (4A) is unknown (in general case). The second term will be positive if there is a difference in average incomes between the groups (i.e. if ,B>1). If the difference between the average incomes is small (p close to 1)., the equation (4A) may be negative. In that case, G1NI will increase due to an increase in the overlap component. The overlap component will increase because the share of the more 'dispersed"(unequal) sector increases. Using the values from our example, however, SG1NISh = 0.18. Thus with 8h =0.2 (as assumed below), the change in the Gini, due to the non-overlap component alone, will be 0.036 or 3.6 Gini points. 11 (ii) The changed composition of employment will affect income shares: 7ry will increase and 7cwwill decline. Since Gy>Gw, GINI will rise. (iii) Gy will also increase driving overall GINI up.18 Let now twenty percent of labor force move out of state into private sector (Ah=0.2), and ten percent of labor force become unemployed (Nu=O.l). The share of private sector in the population (household heads) increases to 40% and in total income to 57%; the share of state sector in population and in total income drops to 30%; the share of transfer-recipients (pensioners and the unemployed) in population increases to 30% and in income to 14%. 19 f and y for simplicity stay unchanged.20 The overall GINI without the overlap component rises to about 33; with the overlap component probably to 35-36. This is approximately the level of inequality recorded now in the Baltics and the Balkans. Inequality in Central Europe is lower (about 25) and in Russia and Ukraine higher (about 45). 18The change in GINI will beAGMNI=y pyAGy > 0. 9 Before the transition, overaU income was 100 wage units (this can be seen from the RHS of equation 1). After the transition, the overall income is 105 wage units: 40 persons in private sector earning 1.5 average wageplus 30 persons in state sector eaming average wageplus 30 transfer recipients eaming 1/2 average wage = 105 wage units. 20 Note than P=1.5 wS is still possible despite the fact that wp=ws (see Figure 2). This is because the private sector income includes not only wages but net profits as well, i.e. the rectangle wpNp plus a part of the area (net profits) between mplp and wp in Figure 2. 12 Table 2. Break-down of inequality after the transition Formula Value (in Gini points) (A) Inequality among state Gw itw pw 2.2 workers (B) Inequality among private Gy nry py 9.1 sector (C) Inequality among transfer- Gt rt pt 0.6 recipients (D) Differences in average I incomes ((y-ws) p3, pw +(y-t py pt + (wSt) pwpt) Total without overlapping [A to D] =33 Overlapping 2-3 points Total GINI about 35-36 Assumptions: pw30°/o, py=40/o, pp=30%; Ws=1; pension or unemployment benefitlwage=0.5; self- employment income/wage=l.5; Gw=0.25; Gy=0.4; Gp=O. 15. Hence, aw=29%, ny=57%, 7tp14%, p=1.05. If we compare Tables 1 and 2, we can see the sources of increased inequality. The shift of workers from the relatively egalitarian state sector to the private sector and into unemployment increased inequality by I Gini point (sum of A, B and C in Table 2 vs. the sum ofA, B and C in Table 1).2' Had inequality in the private sector remained the same, there would have been no increase. The differences in average incomes --primarily due to the fact that the weight (py pt/g) attached to (the large) gap between the average private sector income and average pension has increased substantially--account now for 21 instead of 16 Gini points. The greater inter-group differences have therefore added some 5 Gini points to inequality. The greater inter-group differences are not due to the changed average relative incomes between the groups (which by construction I assumed constant), but to a changing composition among the employed, and between the employed arid the transfer-recipients. In essence what occurred was the "hollowing out" of the middle. While before the transition, 60% of the household head were employed at the average income (=state sector wage) while 20% each were either eaniing more (the self- employed) or less (pensioners), the transition cut down this "middle class" to half as a large chunk of them moved to the private sector and some became 21 The ffist t lm in Table 1 sum to 11.1 Gini points; the same lines in Table 2 sum to 11.9 Gini points. 13 unemployed. Thus both the "rich" and the "poor" increased, while the middle shrunk. In conclusion, this simple example illustrates that the bulk of the increased inequality (about 5 Gini points) was due to the changing composition of employment and the emergence of unemployment ("hollowing out of the middle") and about 1 Gini point to the increased inequality in the private sector. A further Gini point increase might have been caused by the greater income overlapping of private sector people with state-sector workers and transfer- recipients (the latter two groups are still assumed not to overlap).22 3. Inequality by sources of income23 The analysis so far has been couched in terms of income recipients. This is from the point of view of exposition of what happened during the transition an easier way to proceed. We can simply say that some persons who used to work in state sector have transferred to private sector. But this "tagging", as mentioned in Introduction is neither.realistic (people have numerous sources of income), nor are income distribution data normally presented in that way. Thus we need to move to a study of income sources (wages, private-sector income etc.) rather than individuals, that is, to the disaggregation of the GINI by factor incomes. The formula for the decomposition of the GINI also gets simpler as the overlap term disappears.24 We can now write disposable incomefor each person as the sum of wages (w), cash social transfers (t), and non-wage private sector income (p). The Gini coefficient of disposable income is then formally equal to the weighted average of the concentration coefficients Ci of the three individual sources (wages, transfers, private sector income) where weights are their shares (Si) in total income (equation 6):2 22 The greater overlap of private sector people is simply due to their greater nunbers. Since they are distributed across the entire income spectrum and there are more of them, the overlap component of the GINI will increase. 23 Parts of this section are published in Chapter 4 of Milanovic (1998). 24 Unlike in the previous case, GINI can be exactly decomposed. 2The concentration coefficient captures both inherent inequality with which a given income source is distributed (source Gini coefficient) and the correlation of that source with the overall income. Thus, an 14 (6) G > 3 SiCi = S.C. + StCt + SpCp The change in the Gini between two dates (before and after the transition) can be written as: (7) AG = Y3=AAsi Ci + Acm. Sw + AC, St + ACp Sp + 3=IASi ACi The first term on the RHS shows the change in Gini due to the changing shares of different income sources; the next three terms show the change due to changing concentration coefficients of income sources; and the last term is an interaction term. What happened to factor shares? Consider first what happened to Si's during the transition. Table 3 shows the shares of wages, pensions, other social transfers, and non-wage private incomes before the transition (1987-89) and in 1993-95 in six transition economies: four in Eastern Europe, two in FSU. For simplicity of presentation, I have selected only the end-years for each country. The yearly data are shown in Figure 3 (left panels). All data are calculated from the countries' Household Budget Surveys. The sample is limited to the countries where I had access to the fairly detailed 26 successive annual or quarterly income distribution data.27 inherently unequal source like social assistance with a high Gini coefficient will have a low or negative correlation with overall income (because most of social assistance recipients are poor), and its concentration coefficient will be low or negative. When we use the term "concentration" of the source we have in mind not only its inherent inequality but also how it correlates with the overall income.The exact definition of the concentration coefficient of the source i is Ci =Gi R, where Gi=Gini coefficient of the source and R cov[i, rank(income)] ratio of covariances between source i and ranking of cov[i, rank(i)] recipients according to total income, and source i and rankig of recipients according to source i. rank () is a rank function taking values from I to N (total rnumber of recipients). If the two ranks coincide, R=1, and Cw=Gw. Since in cov[i,rank(i)] both i and rank(i) uniformly increase, its value will be greater or equal than that of cov[i rank(income)]. Therefore, R-<. 26 Individuals divided into deciles according to per capita household disposable income plus income composition by each decile. 27 The list of surveys and the discussion of data quality is in Annex 1. 15 Wages are defined as all labor earnings including those from second jobs, fringe benefits (in cash or in kind) and from either state or private sector. (Private sector wages were, of course, very rare before the transition). Pensions include all types of pensions (old-age, survivor, invalidity). Other social transfers are all non-pension cash social transfers like family benefits, unemployment allowance, sickness benefits, 28 scholarships, social assistance. Finally, non-wage private incomes are a mixed bag. They include self-employment net income, value of home consumption, private gifts and remittances from abroad, net interest 29 and entrepreneurial income, income from the lease or rental of assets etc. This source does not include wages earned in private sector but does, in principle, include distributed business profits. 28 Some, however, might be included in wages (if paid out by enterprises). 29 In cases of high inflations, however, this source is not included. In HBSs, it is always showri in nominal terms. But, in high inflation conditions most or all of nominal interest income compensates for the depreciation ofthe principal. (Often when real interest rates are negative, not even hat is accomplished.) Strictly speaking, we would need to include as income only the positive real interest portion. When inflation is very small or zero, nominal and positive interest portion are almost the same; but with high inflation it would be a much worse error to treat the entire nominal interest income as real income han to disregard it altogether. 16 Table 3. Composition of population disposable income before the transition and "now" (1993-96) (in percent; calculated from HBSs) Countries (years) Wages Non-wage Pensions Other social private income transfers pre "now pre "now" now" -r "now"_ Bulgaria (89-95) 57 47 22 31 17 18 5 4 Hungary (87-93) 60 50 14 16 19 19 7 15 Poland (87-95) 55 44 24 24 17 27 5 6 Slovenia (87-95) 67 57 20 25 17 22 2 4 Eastern Europe 60 50 20 24 17 21 5 7 Russia (89-94) 74 55 5 23 8 18 7 3 Latvia (89-96) 82 50 12 24 8 18 3 9 FSU 78 53 9 23 8 18 5 6 Sources: see Annex 1. The regional means are unweighted averages. Rows sum to 100. a/ Gross income for pre-transition years. Since personal income taxes were negligible, the difference between gross and disposable income is very small. The data in Table 3 show: . The share of wages has declined in all countries. Its decline has been sharper in Russia and Latvia (where initially wages were more important as an income source) than in Eastern Europe. In the East European countries, the share of wages declined everywhere by about 10 percentage points. In Russia, the wages' share dropped by almost 120, and in Latvia by 30 percentage points. FIGURES 3 AROUND HERE . In Eastern Europe, non-wage private sector income increased everywhere by a few percentage points only.30 In Russia and Latvia, by contrast, this source of income increased its share by almost 15 percentage points. 3 As mentionned, non-wage private sector income includes (i) self-employment and small enterprise income including that in agriculture, (ii) v;lue of home consumption, (iii) gifts and remittances, (iv) property and rental incomes. Before the transition, the entire private sector income belonged there (as in our description of transition in Section 1). After the transition, many of successful self-employed and small enterprises grew into "regular" private firms. Their wage payments, as well as those of ab novo private companies and the privatized SOEs, are now included together with other wages. This explains 17 Figure 3 Composition of disposable income Concentration coefficients of wages, (in percent) social transfers and private-sector income Poland 5a a 40 29 -acns* t .pensioNS., 3D pensions 20 a other tiansfels -20- 97 BB 89 9O 91 92 93 94 95 87 81 a9 91 9S 92 93 94 95 Slovenia ___________________________ ~~(b) p\ en s inin s 40 . 60 20 ; olhei transfers 11 8B 89 90 91 92 93 94 95 17 aa 09 91 91 92 93 94 95 Figure 3 (continued) Composition of disposable income Concentration coefficients of wages, (in percent) social transfers and private-sector income Bulgaria (c) IE~~ ~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~ 50 p_- 50 \ ~~~~~~~~~~~~~~~~~~~~~~~~~~~earnings 4 20 - ,,,,,,,,,,,,,,,, ,,,.,,.......... .......,e.ios. ....-1 -.,.esin, .......... -- , , -- .....- ot~~~~~~~~~~iher Iran nier oiler tian ero L , , , , , , , , -IO I I IIIII 7 UB 89 90 9 1 92 3 94 9' 99 90 91 92 93 94 95 Hungary (d) 20 pensiI~~~~~~~~~~~~~~~~~~~~~~~~hs~4 10 e0 -ne I9a0 9 2 - 8 - .... 89 90 ......91 9 19 - frs;: -10_ " '- a~~~~~~~~~~ -2 _ 87 a8 89 90 91 92 93 87 88 89 90 91 92 93 Figure 3 (continued) Composition of disposable income Concentration coefficients of wages, (in percent) social transfers and private-sector income Russia (e) 40~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ privat9 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ __privat 40 -pernsionrs 40 -5 hSI,flS oi L sother iransers 2 0 seensions ~~~~~~~~~~~~~~~~.... I.. .... . .. . .. -40 B9 90 91 92 93 94 09 90 91 92 93 94 Latvia (f) 100 50 60 a0 ernins g~ 40 - 30 " 20- 40- 20~~~~~~~~~~~~~~~~~~~~~~~¶ 20~~~~~~~~~~~~~~~~~~~~~~~~~~ siher trios. ~~~~~other transfer I I I I IC~~~~~~~~~~~~~~-1 07 BB 89 90 91 92 93 94 95 95 07 BB 89 90 91 92 93 94 95 95 Note: More unequally distributed income source has a higher coefficient. The conicentration coefficient also shows how much a given sourc 'pushes' up the overall inequality. Sources: Household Budget Surveys (see Annex 1). . The share of pensions increased without exception in all countries. The increase again was sharper in Russia and Latvia which started with a lower pre- transition share. * Non-pension social transfers increased in all countries except in Russia. Their growth has been particularly dramatic in Hungary where in 1993 they accounted for 16 percent of population disposable income (10 percent in the form of various family benefits, 4 percerLt as unemployment allowances). .Overall, the income compositions in Eastern Europe, on the one hand, and Russia and Latvia, on the other, are much more similar now than before the transition. Wages in all countries are around 1/2 of total disposable income, non- wage private sector accounts for one-fourth of disposable income and so do cash social transfers. But because the initial starting point was further from the current outcome for Russia and Latvia, the changes in income composition in these two, countries were more dramatic than in Eastern Europe. What happened to factor concentration coefficients? Consider next Ci's from equation 6. 3 In all countries, the concentration coefficient of wages went up (Table 4). The increase was substantial for the East European countries, averaging about 10 points (or 50% of the initial value), and truly "gargantuan" for Russia and Latvia. The concentration coefficient of wages in both countries is in excess of 50. A concentration coefficient can increase either because the Gini of the source (wages in this case) goes up or the correlation between the income sources (wages) and disposable income rises. Without additional information (i.e. individual data) we cannot disentanigle the two.31 Whatever the cause, higher concentration of wages clearly puts an upward pressure on the Gini. why this source of income -which now accounts for only a part of total private sector income-- has not grown more substantially. In addition, in case of Poland for example, there was a sharp real, and relative (in relation to overall country average income), drop in farmers' incomes after the transition. 31 See however an attempt in Section 4. 18 Table 4 Concentration coefficients before the transition and "now" (1993-96) (in percent; based on HBSs; individuals ranked by per capita disposable income) a/ Wages Non-wage Pensions Other social private income transfers ______ _pre "now" pre "now"5 pre "now"l pre "now"5 Bulgaria (89-95) 21 34 38 37 11 13 - 6 2 Hungary (87-93) 25 35 30 26 14 21 -13 -16 Poland (87-95) 25 31 37 40 17 36 -10 -13 Slovenia(87-95) 20 26 18 21 22 21 - 4 -19 Eastern Europe 23 32 31 31 16 23 - 8 -12 Russia (89-94) 28 55 18 44 -20 30 8 14 Latvia (89-96) 23 50 16 43 34 9 - 7 7 FSU 25 52 17 43 -- 20 11 Sources: see Annex 1. The regional means are unweighted averages. - indicates that the country differences are so large that averaging is meaningless. a/ Gross income for pre-transition years. Since personal income taxes were negligible, the difference between gross and disposable income is very small. * Non-wage private income has kept the same concentration coefficient in Eastern Europe overall; the changes in individual countries are small too. Before the transition, private sector income had a substantially higher concentration than wages in Bulgaria, Poland and Hungary (30's vs. 20's). Since wages' concentration coefficients have risen, the two (wages, and non-wage private sector income) now have about the same concentration. In Russia and Latvia, private income's concentration, like that of wages, has increased substantially. But it has risen less than the concentration of wages and is still lower. . The average concentration coefficient of pensions in Eastern Europe has increased substantially (from 16 to 23) mostly on account of changes in Poland and Hungary. The improvement in the average ratio between pensions and wages32 has led to pension income (and pensioners) being distributed across the entire income spectrum--much more so than in the past when they were more 32 Not shown here. 19 concentrated among the middle to poorer segments of the population.The data for Russia are suspect, particularly a very high concentration coefficient for 1994. In Latvia, the differentiation and concentration of pensions have declined following the introduction of "flat" pensions in 1992 (see Figure 3f right panel). oThe change in the concentration coefficient of other (non-pension) social transfers is important in its own right, because the brunt of anti-poverty policy (particularly in conditions of massive income declines) falls on these transfers. By their nature, they are either explicitly targeted on the poor (like social assistance) or implicitly so (like unemployment benefits). In Eastern Europe, their targeting has improved from being very mildJy pro-poor in absolute terms (-8) to more strongly so (-12). Almost the entire improvement in targeting is due to the introduction of unemployment benefits (Milanovic, 1997, chapter 5). Since (i) the unemployed are often among the poor and are easily identifiable, and (ii) the rules for benefits' eligibility are relatively clear and are followed by the unemployment offices, unemployment benefits have been focused on the poorer segments of East European population.33 In Russia and Latvia, targeting has deteriorated. Decomposing the -overall GINI change. The outcome of the changes in Si's and Ci's is the change in overall GINI. Table 5 shows the decomposed change in the Gini between a pre-transition year (1987 or 1989) and 1993-6. The laLst column shows the overall increase in the GMN between the two end periods. (The pre-transition year, whether 1987 or 1989 is fixed by the data availability. Since income distribution in the late 1980's was stable, the exact year does nolt matter much. This, however, is not the case for the choice of the transition year because now inequality does change. To avoid "pinning" all conclusions to one --end point year-- I present in Annex 2 the year-after-year GINI decomposition for the entire period. The year-to-year GINIs for East European countries are given in Figure 4.) We can make three conclusions. 33 Point (ii) for example does not hold for social assistance. Also, total spending on unemployment benefits exceeds spending on social assistance in all East European countries. 20 First, the change in the composition of income has had very little to do with increased inequality. In the only country where it did have a significant impact (Russia), it contributed to reduce inequality, i.e. the composition of income in 1994 was more favorable to equality than in 1989.34 This is chiefly because social transfers, which were the most equally distributed income source in Russia before the transition, increased their share in overall income (see Table 3). In other countries, changed income composition added or subtracted only about 1 Gini point to total inequality. Second, higher concentration coefficients of wages (in all countries) drove the overall Gini up. It was the most important factor behind the increase in inequality. The increased wage concentration was responsible for between 3.5 and 8 Gini points increase in Eastern Europe, and for huge 16-18 Gini points increases in Latvia and Russia. In the latter two countries, the increase was due not only to a greatly increased concentration coefficient of wages, but also to a very high initial 1988 share of wages in income. Thus the weight attached to a more unequal concentration of wages is greater (see the term Sw in equation 6) than if the original share were low. 34 Had, of course, the concentration coefficients of various sources remained at their pre-transition levels. 21 Table 5: Decomposition of the change in the Gini coefficient between 1993-96 and before the transition (in Gini points) Due to: Change in Change in concentration of: composition of income Country (end years) Wages Social Out of which: Non- Interaction Overall Gini change (between the end-years) Country (end years) ~~~~~transfers wage te private sector Non-pension Pensions transfers Bulgaria (89-95) +1.4 +7.8 +0.9 +0.4 +0.4 -0.4 +0.3 +10.0 (from 21.7 to 31.7) Poland (87-95) -1.7 +3.4 +3.5 +3.2 -0.1 +0.8 +0.9 +7.0 (from 25.0 to 32.0) Slovenia (87-95) -0.2 +3.6 -0.6 -0.1 -0.4 +0.4 -3.8 +2.6 (from 19.8 to 22.3) Hungary (87-93) -1.3 +5.9 -0.6 +1.4 -0.2 -0.6 -1.3 +2.2 (from 20.7 to 22.9) Russia (89-94) -3.4 +17.8 +5.1 +3.9 +0.4 +3.0 +1.2 +23.6 (from 21.9 to 45.5) Latvia (89-96) -1.6 +15.0 -1.5 -2.0 +0.5 +1.4 -3.3 +10.0 (from 22.6 to 32.6) 22 FIGURE 4 AROUND HERE The increased non-wage private sector income concentration was responsible for 3 Gini points increase in Russia and about 1.5 points in Latvia, while its impact was negligible in Eastern Europe. Third, the effect of transfers on inequality was not uniform across the countries. In Bulgaria, Slovenia, and Hungary, the concentration of transfers did not change. In Latvia, better targeting of transfers reduced inequality by 1.5 Gini points. In Poland and Russia, on the contrary, greater concentration of transfers increased inequality. This was due to a greater concentration coefficient of pensions. Non-pension transfers, because of their small initial size, did not anywhere have much impact on the overall change in inequality. 3 The "stylized" facts are illustrated well on the example of Bulgaria in Figure 3c. The rising concentration of wages (from around 20 to 35) contributed strongly to inequality. The concentration coefficient of private sector income which was high already before the transition remained at the same level while the share of private sector income in total increased. The rising share also pushed up the overall inequality. Pensions' concentration and share both remained unchanged thus leaving inequality unchanged. Finally, non-pension transfers were too small (less than 5 percent of total income) to make any difference to the overall GINI. Polish, and to a lesser degree Slovenian, results, illustrate a different story (see Figures 3a and 3b). Although wage concentration increased markedly, the most dramatic developments were in the area of social transfers: their rising concentration, and the rising share in overall income. In 1995, pensions had the same concentration coefficient as wages and non-wage private sector income. Russia represents a unique case of a country where all income sources' concentration coefficients are higher now than before the transition: they all pushed overall inequality up (see Figure 3e, right panel). The only factor that moderated the increase in inequality was a shift toward more equally distributed income sources: 35This conclusion differs to some extent from Cornia's (1994, p.39) observation that "the relative importance of redistribution [via transfers] has grown... Targeting of these [social] transfers has generally improved or remained sufficiently progressive." 23 Figure 4 Gini coelffcients in Eastern Europe (based on individuals ranked by annual household per capita disposable income) 3 5 BUL PO 15 I I I I B7 98 89 90 91 92 93 94 95 Sources: see Annex 1. transfers and non-wage private sector income which prior to the transition had smaller concentration coefficients than wages (see Table 3). As we saw before, transfers, and in particular pensions, either left inequality unchanged (Slovenia, Hungary and Bulgaria),36 or contributed to its increase (Poland and Russia). The only exception among our countries is Latvia where "improved" (more pro-poor) concentration of pensions was due to the introduction of almost flat pensions in 1992 (note the steep downward-sloping line for pensions in Figure 3f, right panel). 3 4. The conclusions: contrasting the model and the empirical evidence The model and numerical simulations from Sections 1 and 2 led to several predictions regarding the change in factor shares and inequality during the transition. We shall consider three predictions: regarding (i) the changes in the income shares, (ii) mechanism that underlies increased inequality, and (iii) "hollowing out" of the middle class (state-sector workers). First, we expect, of course, a declining share of state-sector income. The model also "predicts" that, after the transition, the share of transfers in total income will be greater. This is because the government will have to pay, in addition to unchanged relative pensions,38 unemployment benefits. Indeed, this is what the evidence from seven transition economies in Table 3 confirms. There is not a single economy where the share of cash social transfers has not risen. This is also true separately for pensions, and for all other transfers.39 Second, one cause of increased inequality during the transition lies, according to our numeirical simulations, (1) in the transfer of labor from the relatively low-inequality state sector to the high-inequality pnvate sector. In the empirical part, we saw that (2) the concentration coefficiernt of wages which includes both private and state sector 36 Note the relative stability of pensions' concentration coefficients in Slovenia, Hungary and Bulgaria (Figures 3b, 3c and 3d right panels). 37 The concentration coefficient of pensions decreased from 34 in 1989 to -4 in 1995. 38 Compared to state wages (state wages are the numeraire). 39 With the exception of Russia where non-pension transfers have declined. 24 wages increased everywhere and, moreover, that its increase was the most important element driving GINI up. Is (2) consistent with (1)? Note first that since Cw now includes private sector wages (while before the transition it did not), there is prima facie evidence that private sector wages had something to do with a higher CW.40 Second, a higher Cw may be due either to a higher Gini coefficient of wages (which is what wie would expect from our model) or to the increase in the correlation coefficient between wages and overall income. We can write this as: Cw = Gw Rw Before the transition Cw was between 23 and 25; after the transition it rose to 32 in Eastern Europe and about 50 in Russia and Latvia (see Table 4). Since we know that before the transition Gwwas between 25 and 30 (see Figure 5), Rw must have also been close to 1 (more exactly between 0.85 and 0.9). And, indeed it could be expected Ihat the link between one's wage ranking and one's overall income ranking (see the formula of Rw in footnote 25 above) was fairly high because wages represented about 60 percent of total income in Eastern Europe and. even 80 percent in the FSU. But, if Rw before the transition was close to 1, there was no much "room" left for Rw to increase during the transition. If, for example, Rw went up from 0.9 to 1, that would "push" Cw upwards by only about 2-2.5 points (from Cw=25 to Cw=27.5). On the other hand, we know that Cw increased by 9 points in Eastern Europe and more than 25 points in Russia and Latvia. Consequently, most of the increase of the concentration coefficient of wages must be due to the increased Gini coefficient of wages, i.e. to the greater inequality among the wage-earners. The empirical evidence, is therefore consistent with the implication contained in the numerical example in Section 2. FIGIJRE 5 AROUND HERE Third, according to our model, the most important source of increased GIN1 is the "hollowing out of the middle", thatt is increasing inter-group differences (see Table 2). Since the empirical evidence does not deal with recipients but with income sources, we do not have direct evidence on this. However, what one observes for several countries (Poland, Slovenia, Latvia, Flungary) is both an increase and a gradual closing of the gap between the concentration coefficients of non-wage private sector income, 40 It is, of course, also possible even if not probable that Cw increased only because the concentration coefficient of state sector wages increased. 25 Figure 5 Gini coefficients of wages, 1956-1994 30 - Poland 25- 20- 15 - 1 1 l l l l l-7l , , , l l , l 56 61 66 71 76 81 86 91 Source: Calculated from the countries Statistical yearbooks (various issues) from the distribution of monthly wages (World Bank PRDTE data base). Note: There is a break in the Hungary series in 1988 due to introduction of gross (instead of net) wages. and wages (see Figure 3). At the beginning of the transition, the non-wage private sector income had typically higher C's than wages. But during the transition, a part of private sector income "moves" into wages as some small businesses and the self-employed decide to expand and join the "regular" capitalist private sector. This "switch" dampens the increase in the recorded share of non-wage private sector (which, as we have seen, has remained unchanged in Eastern E'urope), and since potentially the most profitable and unequal part "moves" out dampens too the increase in this source's concentration coefficient. By the same token, the "switch" combined with the widening of wage distribution in the state sector and in the ab novo private sector, increased the concentration of wages. Thus the two sources, non-wage private income and wages, came to have similar, albeit higher than before the transition, Cs. Because the "hollowing out" hypothesis is couched in terms of people with specific sources of income (e.g. statev sector workers) who get pulled in two opposite directions--some do very well (e.g. become entrepreneurs), and others very badly (e.g. become unemployed)-- the increase of the concentration coefficients does not directly address this hypothesis. The "hollowing out", however, is consistent with the increased wage and private-sector inequality. But more research is needed: one needs to study issues of polarization, rather than inequality--a topic that is beyond the scope of this paper. 26 ANNEX 1 DESCR1!PTION OF THE SURVEYS USED AND DATA PROBLEMS 1. Description of the surveys used For all East European countries and Latvia, the data come from the official surveys conducted by the countries' statistical agencies (CSO). For Russia, the 1989 is the official survey; the 1994 survey is the World Bank and Goskomstat Rossii jointly sponsored Russian LongitudinalMonitoring Survey, RLMS (see Table 1). For Poland, Bulgaria and Slovenia all survey instruments are the same, that is the surveys for each individual country have exactly the same design year after year (barring some improvements: e.g. the Polish surveys became fully representative in 1993). For Hungary, the 1987 and 1993 survey instruments are the same (Household budget survey that is normally conducted once every two years, but whose 1991 results were not published). The 1991 survey is a microsimulation of a large 1987 Income survey conducted by CSO. For Latvia, the 1989 survey is a Living Standard Survey; 41 the 1992-93 surveys, however, are the unrepresentative quota-sample Soviet Family budget surveys. Finally, for 1995-96, I use the new and representative New Latvian Household Survey. For Russia, the 1989 data come from the old Soviet survey; the 1992 and 1994 are from the representative RLMS (Round 4). Almost all surveys are annual. The shortest ones are the 1995 and 1996 Latvia surveys which are quarterly. Out of the total of 34 surveys used, 29 are annual, 3 are semi-annual, and 2 are quarterly. 41 Living standards survey was conducted once every five years. 27 Table 1. Characteiistics of the surveys County Source of data; survey Period Period of analysis Data reported in: Representative Income concept Income Other problems (number conducted by: covered survey includes with imcome or of home expenditure surveys) .cption def_itions Poland CSO, Household budget 1987-93; 1987-92 and Grouped data for the period 1987-92 Fully representative Disposable after Yes None (8) surveys (Budzety 1995 1995 yearly, (published inthe annual Budzety since 1993; before 1993; gross before gospodarstw domowych) 1993 frst six gospodarstw domowych, Central policy, Army and (the difference months. Statistical Office, Warsaw). non-agricultural between gross and Individual data for 1993 and 1995. private sector disposable was omitted. minimal; less than 1%) Hungaiy CSO, Household budget 1987, 1989, Yearly. For 1987, grouped data as published Yes Disposable No None (3) surveys for 1987 and 1993; 1993 in CsaladiKoltsegvetes 1987, Income survey Central Statistical Office, Budapest, microsimulation for 1989 1989, pp.78-9, 102-3, 126-7. For 1989, detailed grouped data from Kupa and Fajth (1990). For 1993, i_ndividual data available. Slovenia CSO, Household budget 1987-95 Yearly. Grouped data. For 1987-91 data Yes Disposable Yes Incorrect definition (9) surveys. published inAnketa o potrosnji of income domacinstava, Federal statistical (adjustment made) Office, Belgrade (all data presented a/ by republics). For 1992-95 personal communication by the Statistical office of Slovenia (Mrs. Irena Krizman and Mrs. Alenka Kajzer). Some results published also in Statistiene infonnaciye, and Rezultati raziskovanja st.684/1997 Statistical Office of Slovenia, Ljubljana. Bulgaria 1989-94 CSO, Household 1989-95 1989-94 yearly, For 1989-94 grouped data reported Yes 1989-94 gross Yes Slightly incorrect (7) budget surveys; 1995 1995 first six inBiudzeti na domakinstvata v income (the definition of Gallup survey months. Republika B 'Iganya, National difference between income (adjustment Statistical Institute, Sofia. For 1995, gross and made) a/ individual data available. disposable minimal). 1995 disposable. Latvia CSO, 1989 Living 1989, 1992, 1989, 1992-93 Grouped data. For 1989 data Not before 1995 1989-93 gross Not in 1989: slightly 28 (5) 1 standards survey, 1992-93 1993, 1995, annual; 1995-96 reported in Survey of Living (the Soviet-type income (the 11992- mcincorrect defintion Family budget surveys; 1996 quarterly Standards, Riga: Goskomstat branch-based difference between 1 93. 1 of income. a! 1995-96 New Family Latviiskoi SSR, 1990; for 1992-93 survey). gross and Yes for 1992-93 money budget suveys reported in Gimenes budzets, Riga: Representative disposable other income only. CSO; for 1995-96, personal since 1995. minimal). 1995-96 years. communication by the Latvian disposable. Statistical Offices (Mr. Edmunds ._____ Vaskis). Russia 1989 CSO; 1992 and 1994 1989, 1992, 1989 annual; For 1989 and 1992, grouped data 1989 not 1989 gross income Yes 1989 slightly (3) RLMS 1994 1992 quarterly reported respectively in Popkin representative the (the difference incorrect definition 1994 (Oct. 1993- (1992, Table 5) and Popkin (1993, Soviet-type branch- between gross and of income. a/ Feb. 1994) Table 20). For 1994, individual data based survey); disposable available. 1992 and 1994 minimal). 1992 and surveys 1994 disposable representative. income. Note: CSO=country's statistical office. HBS=Household budget aurvey FBS=Famnly Budget survey. RLMS=Russian Living Standards Monitonng Survey. a/ Disposable income calculated by deducting some revenue itms froi the incone conept ued by the CSO (and according to whichthe individuals and hoseholds we ranked). 29 Out of 35 surveys used, I had access to individual data for five. For all others I used the grouped data. The number of published groups varied between 10 and 20. The income groups were formed according to CSOs' definitions of per capita income. In two countries (Slovenia and Bulgaria) this led to some problems because the CSO-defined income included items that did not belong to income.42 In Slovenia, it included net withdrawals from saving accounts and personal borrowing; in Bulgaria, sales of assets and insurance compensations. These items had to be deducted from the CSO-defined income in order to obtain actual disposable income. Performing this operation on the grouped data (in distinction to individual data) implies that the measured income inequality becomes underestimated because we no longer, strictly speaking, estimate the Gini coefficient of disposable income but the concentration coefficient of disposable income. The problem is negligible in the case of Bulgaria because the "wrong" items account for less than 1 percent of the C SO-defined income; in Slovenia, however, they account for about 8 percent of the CSO-defined income. For Poland, Bulgaria, Latvia and Russia, for a number of years, the income concept used is gross rather than the disposable income.43 However, since personal income taxes were minimal because gross income excludes payroll taxes withdrawn at source which represented the largest chunk of personal taxes. In all cases, the difference between disposable and gross income was less than 1 percent, and thus using either of the two concepts would produce the same results. (This, of course, has changed now with the introduction of more substantial PIT system in Hungary and Poland.) Finally, in all cases except Hungary in 1993, home- consumption is included in income. The components of disposable income are standard. Disposable income is equal to all wage earnings (from primary and secondary jobs etc.) plus cash social transfer plus income from property and entrepreneurship plus received gifts plus value of home consumption. It excludes payroll and PIT taxes. 2. Comparing pre-transition and transition years: what are the biases? The comparison between Polish, Bulgarian, and Slovenian survey results over the period 1987-95 is straightforward and warranted. All the data come from essentially the same surveys, and no dramatic changes in the refusal rates (they went up though) or underreporting (it went up too) occurred The same is, to a large extent, true for Hungary whose 1987 and 1993 survey instruments are the same. 42 The same problem exists with the Bulgarian data but is negligible since the "wrong" items account for less than 1 percent of the CSO-deflned income. In Slovenia, the "wrong" items, however, account for almost 8 percent of the CSO-defined income. 43 Disposable income=gross income minus palroll and direct personal taxes (Pfl). 30 However, this somewhat optimistic assessment needs to be qualified. There is a change for which the users of surveys, and possibly the "producers" of surveys too, could not control. It is the change that accompanied the transition, and has nothing to do with the survey designper se. We can caJl it systemic or underlying change. Generally speaking, refusal rates have increased during the transition and particularly among the rich; coverage of wage and social transfer income that was nearly 100 percent before the transition has deteriorated (as earnings reported by the households which in the past used to be double-checked against the enterprise or pension authorities' records are no longer so checked); the omission or inadequate coverage of informal (and illegal) sector income before the transition has now become an even greater problem as such incomes have increased in absolute and relative terms. These are some of the problems for which we, as users of the surveys, cannot correct. The bottom line effect of these systemic changes --assuming an unchanged survey design-- is that incomes are now more underestimated than in the past. The direction of the bias in terms of inequality is less clear. In the past, surveys underestimated inequality by not accounting for many fringe benefits and perks received by the elite.44 Today, they might underestimate it by not covering those with high incomes who refuse to participate. But, it is up to each researcher to decide how strong an emphasis he or she wishes to place on these systemic (underlying) changes; how much he or she believes that they vitiate all pre-post comparisons. I would tend to believe that the underlying change in Eastern Europe was not of such a magnitude as to render, after appropriate caveats, the comparisons of inequality before and after the transition unreliable. On the other hand, the argument that such comparisons are much less reliable can be, I think, made with respect to some of the republics of the former Soviet Union. In the Soviet case, not only was the underlying change much more profound (witness the explosion of the informal sector), but the initial surveys were fundamentally flawed because they were not random surveys but basically surveys of the families of the employed (see the discussion below), among which the "average" households (e.g. both parents employed, and having 1 or 2 children) tended to be oversampled. To these households a quota of the pensioners and students living outside their homes was added. The pre-transition surveys in the former Soviet Union were biased, left out large segments of the population, and tended to show higher average incomes and lower inequality. Thus the recorded change in inequality in Latvia and Russia is almost certainly bound to appear larger than the actual change. 44 Subsidies were not included either; yet their effect was (with the possible exception of housing subsidies) to reduce inequality. 31 Survey biases before the transition. What can we more formally say about the survey biases before the transition? We shall consider four areas: survey design, uderreporting of income, the use of per capita vs. equivalent units, and annual vs. quarterly data. The very fact that these caveats are listed here indicates that we cannot do much (or anything) to remedy them. Yet they are worth listing for two reasons: to provide some caution when it comes to the interpretation of the results, and to delineate the areas that most clearly need improvement in future work. Survey design This is the problem of sampling inadequacy. The household surveys that we use have been criticized, rightly, for several biases. The Eastern European surveys were sample surveys. However, in several countries (e.g. Poland), they were not designed to be representative of the entire population but rather of individual socioeconomic groups (SEGs). This was probably the product of a Marxist view of society as composed of social classes and concern with intergroup equity. The data were thus representative of workers' households (in the state sector) or of pensioners, but they could not be easily combined to obtain an accurate picture for the whole population, essentially for two reasons. First, the sample shares of the groups that were included were not always proportional to their shares in the population (e.g., there were too many workers and not enough pensioners) and the results were not corrected for systematic differences in the rate of refusal to participate in the surveys. Second, some groups were left out of the surveys entirely. These groups included both those with high incomes (self- employed entrepreneurs, Army and police personnel) and those with low incomes (the institutionalized population, the unemployed). Income distribution was truncated at both ends. The Soviet data were even more problematic. Not only could data for SEGs not be combined, but the surveys were not based on a sample technique but on selecting households at their place of work (the so-called "branch [of production] approach"). Workers and farmers were chosen by their managers and asked to cooperate with statistical authorities. The results were biased: the employed were systematically overrepresented in relation to the non-employed (to correct some of the bias a quota of pensioners and students was added);45 workers in large enterprises and with a longer work record were more often selected that those working in small firms and with a shorter work record. Since the selection criterion was employment, larger households were undersampled.46 The survey was essentially a panel-with the same 4`The pensioners households were simply "added on" -i.e. statistical offices will be asked to add a quota of pensioners which was often below their true share in the population. 461n order to have unbiased results, the probability of selection of a larger household should be proportionately greater han the probabilty of selection of a smaller household. But when the criterion of selection is employment and the participation ra,tes are high, the two households (e.g. one with two adults 32 households staying in the sample year after year-but to further complicate matters, it was not explicitly designed as panel, and the identification numbers of the households were not systematically maintained. The panel nature of the survey further biased the results: since households were supposed to stay in the sample indefinitely, the share of older working households,47 presumably with higher than average earnings, was too high. In conclusion, there were two kinds of biases. First, a bias toward the sampling that is representative of various pre-defined socio-economic groups but not of the population as a whole with the result that people that could not "fit" into any of the main social groups were likely to be left out, and these people were often at greater risk of poverty than the average citizen. Second, a bias toward the "average" or "normal" households that existed only in the household surveys that followed the so-called "branch principle". These are Soviet and, it seems, Romanian surveys.4' The sampling selection was skewed in favor of "average" enterprises, "average" workers in terms of earnings, "average" skills, "average" family size (a couple with one or two children) etc. Thus even within a given social group (workers in state enterprises) income distribution was truncated. Underreporting of income The second problem has to do with income. The use of income, rather than expenditure, data tends to underestimate "true" welfare. This is because people tend to hide their sources of income and thus to underreport them.49 They are less careful when asked to remember their expenditures. An example of this tendency is shown in Figure 1, which gives income and expenditure data by ventiles (5 percents of recipients) for Poland in 1993. Individuals are ranked on the horizontal axis according to their level of household (per capita) income. An interesting fact is revealed by the situation of the lowest income ventile. The reported expenditures of the lowest ventile are twice its income and are equal to the expenditures of the fourth ventile. This indicates a possible measurement problem: the people in the lowest income ventile are in reality not very different from those who are (according to income) significantly better off. It seems that they either severely underreport their income or that their permanent income substantially diverges from their current income. But in any case, and three children, and another with two adults only) will have approximately the same probability of being selected. 47'Once households stopped working, they generally tended to drop out. 'In addition, Romanian results were doctored to such an extent that they are worthless for the years before 1989. Bulgaria also followed "the branch principle" in the 1970's, but abandoned it later. 49The underreporting problem exists in market economies too. It is particularly severe for self- employment and capital income. Atkinson, Rainwater and Smeeding (1995, Table A3) find, using LIS data, that self-employment income is underestimd (compared to the national accounts data) by between 10 percent (Canada) and 60 percent (W. Germany). Property income in almost all countries (US, IUK, Italy, Germany, Finland, Canada and Australia) is underestimated by a half 33 in our statistics they would be counted as poor. We thus impart an upward bias to the poverty rates. Generally speaking, the countries that have a greater share of the informal sector ("gray economy") and small-scale private sector will be more affected.50 Their data will systematically show lower incomes and higher poverty than the data in countries in which an overwhelming share of income is earned in the state sector, or in the wage-reporting (and thus tax-paying) private sector, or is received in the form of social transfers. Inter-temporal comparisons will be affected too. As the share of the gray economy rises with the transition, the problem becomes more serious. On the offsetting side, however, improvements have been made in the survey techniques and greater effort is now made to include such "gray" sources of income. For example, all countries except those that still stick with the Soviet-type surveys, now include the self-employed in their HBSs.5" Hungarian statistical authorities have been imputing tips, "fees" and "black" income. Finally, while the gray income often remains illegal (in the sense that people do not pay taxes on it), there is no longer a political compulsion to ignore it. The political compulsion existed in the past, with both households and the enumerators being keenly aware that such sources of income are not only illegal but also "politically incorrect." Both preferred to look the other way and ignore all non-official sources of income. This source of bias is now gone. WEighty-four percent of individuals belonging to the lowest income ventile are individual farmers, irners-workers, and the self-employed (outside agiculture). 51 In the past, they were left out of the surveys. 34 Figure 1. Income and expenditures per capita, Poland 1993 (individuals ranked by income per capita) 4000 3500 3000 2500 '~2000 -Expenditure g ~~~~~~~~~~~~~~~~~~~-Inoome 1500- 1000 500 - C ( ) (CD O C. O 0) 0 J X) v ) O r d 0 Ventiles Per capita versus equivalent units International comparisons of poverty are --to make an understatement-- complicated. In the particular context of transition economies, there are at least several problems that must be mentioned. First, the use of a per capita poverty line exaggerates poverty in any country compared to using an equivalent-scale derived poverty line. This is because with an equivalence scale the needs of additional family members (often children) do not rise in proportion as they do when we use a per capita measure. By implication then, the use of a per capita line exaggerates even more poverty in countries with larger average family size."2 For example, Marnie and Micklewright (1993) compare Uzbekistan and Ukraine using the same Soviet Family Budget Surveys (FBS) for 1989. They find that the larger household size in Uzbekistan accounts for 14 out of 38 percentage point difference in the headcount index between the two republics. There are several reasons why we use the per capita line: the data in all countries are published in that form (rather than being adjusted for household composition by the use of equivalence scales); the economies of scale in consumption under socialism were typically less than in market economies because the main source of such economies of scale (housing, 52In a recent paper Coulter, Cowell and Jenkins (1992) show that the poverty headcount charts a U- shaped pattern, first decreasing and then rising, as equivalence scale moves from 0 (full economies of scale: total household income alone counts) to 1 (per capita calculations). The same results are obtained by Forster (1993, p.21) in an empirical study of 13 OECD countries. 35 utilities, etc.) were heavily subsidized and this, still remains true although to a lesser extent; the use of per capita poverty comparison allows uls to move easily from such per capita comparisons to GDP per capita comparisons. Quarterly versus yearly data A final problem concerns the time period of data collection. Normally, the surveys are designed in such a way that households repoit (i.e. keep track of, or recall) income and expenditures for a quarter or a month. These data are then "blown up" for the entire year.5 Under conditions of high inflation, however, the data collected in different months represent wholly different real quantities of goods and services and cannot be summed up unless they are adjusted for inflation. The adjustments are sometimes not made by statistical agencies, and at times they are made inadequately (if, for example, inflation is understated). Under such conditions, quarterly data on income or expenditure distribution, from which we calculate poverty figures for several FSU countries, are to be preferred because they refer to a shorter time period and imply about the same command over real goods and services. The usual drawback of the short-period data, namely, t!hat they overestimate income inequality and poverty (there are more people with extraordinarily low and high incomes the shorter the time period), is then of less import than the advantage that the same reported money amounts represent approximately the same real quantities of goods and services. Comparison: before and after the. transition If we take that the "best" achievable household survey should be: (i) representative for the country as a whole, (ii) refer to annual income; (iii) use disposable income; (iv) include home-consumption, and (v) have income "correctly" defined, and consider the transition-year surveys no survey survey fulfills all five conditions but several come close. Bulgarian and Slovenian surveys have a slightly incorrect definition of income for which I could not fully adjust as I did not have the individual data; Hungary's 1993 survey does not include home- consumption in income;54 for Poland in 1993, 1 had semi-annual instead of annual data. Russian and Latvian 1994-96 surveys are quarterly. One may also be somewhat skeptical regarding the claim that the Russian RLMS survey is self-weighted. (Note, however, that even the "besil" achievable survey would still contain some possible biases--which are present in all the surveys, including in those in developed countries. The "best" survey would still probably understate the two tail-ends of income distribution (the poorest and the richest) who are typically undersurveyed, and it would also underestimate "3In order to increase the response rate (which was one of the main sources of bias), Polish household surveys began to require households to keep track of their income and expenditures for one, instead of three, months. The response rate increased from 65 to 80 percent (see Kordos, 1994). 541t includes it in expenditures, but it could not have been separated from other expenditures and thus could not be added to income. 36 some sources of income like those from property (routinely underestimated by up to 50 percent in developed countries)" and entrepreneurship.) Table 2 shows where the surveys falls short of the five requirements iisted above, and what this implies in terms of the bias when estimating inequality. Table 2. Survey defects and inequality bias: pre-transition and transition years Pre-transition Transition Change in inequaity bias Poland incomplete coverage of semi-annual in 1993; but annual Slight overestimate of the __________ recipients in 1995 increase Hungary no home consumption in 1993 Slight overestimate of the ._ _ _increase Slovenia income definition problem income definition problem None Bulgaria Minor income definition Minor income definition Negligible problem problem. gross instead of disposable gross instead of disposable income income Latvia quota, biased sample quarterly instead of annual data Overestimates increase (strongly) Russia quota, biased sample quarterly instead of annual data Overestimates increase __________ I (strongly) In case of Hungary, the absence of home-consumption will lead to a slight increase in inequality, because home-consumption is generally greater for the poorer households. In Poland, too, there should be a slight overestimate of the increase because the pre-transition surveys did not include the entire population (i.e. left out some well-off segments). For Bulgaria, the use of gross income instead of disposable will reduce measured inequality (to the extent that personal income taxes are progressive). However, since PIT is very small, as most taxes are deducted at source, the downward bias be negligible. The bias is more serious for Latvia and Russia. The old Soviet survey could not satisfy more than two (annual data and inclusion of home consumption) out of five requirements listed above.56 The choice of households to participate in the surveys was biased. The new improved surveys suffer from too short a period of observation (a quarter). 55 The so-called "non-eaned" income is understated by about 40 percent (compared to national accounts statistics) by the US Current Population Survey (see Michel, 1991, p.185). 56 In addition. since the difference between gross and net income was negligible, they could be said to satisfy the condition (iii). 37 ANNEX 2 Decomposition of annual changes in the Gini coefficient BULGARIA . base year 1989 1990 1991 1992 1993 1994 1995 Shares 0.5 1.4 1.7 1.7 3.3 1.4 Wages 0.6 -1.9 5.7 7.1 4.3 7.8 Non-wage private 0.9 1.7 2.3 2.9 -0.9 -0.4 Social Transfers -1.1 -0.4 -1.5 -0.0 0.1 0.9 Pensions -1.1 -0.3 -1.7 -0.4 -0.2 0.4 Non-pension 0.0 0.0 0.2 0.4 0.2 0.4 Total 1.1 1.9 8.3 11.7 12.5 10.0 Residual 0.2 1.1 0.1 -0.0 5.8 0.3 POLAND base year 1987 1988 1989 1990 1991 1992 1993 1995 Shares 0.3 -0.5 -0.6 -2.4 -3.2 -2.6 -1.7 Wages -0.0 3.0 2.2 2.3 3.9 4.1 3.4 Non-wage private -2.1 -3.1 -0.4 1.7 1.5 3.4 3.5 Social transfers 1.3 2.1 -0.4 -3.3 -3.8 -1.9 0.8 Pensions -1.6 -3.2 -0.4 1.9 1.8 1.0 3.2 Non-pension -0.3 0.2 0.1 0.0 -0.3 -0.5 -0.1 Total -0.7 1.0 0.5 -0.3 0.4 4.9 6.9 Residual -0.1 -0,5 -0.4 1.3 2.1 1.8 0.9 SLOVENIA base year 1987 1988 1989 1990 1991 1992 1993 1994 1995 Shares 0.0 0.1 0.0 -0.1 -0.1 -0.2 -0.2 -0.2 Wages -0.7 3.5 -0.7 2.2 2.3 3.2 1.5 3.6 Non-wage private 0.4 0.5 -0.2 1.7 1.6 2.2 1.5 0.4 Social transfers -2.8 -2.4 1.2 -1.4 -0.7 -1.4 -1.1 -0.6 Pensions -2.4 -2.5 1.2 -1.2 -0.5 -0.8 -0.2 -0.1 Non-pension -0.3 0.2 0.2 -0.1 -0.2 -0.3 -0.4 -0.4 Total -2.4 3.1 0.1 2.3 2.8 4.3 1.7 2.6 Residual 5.8 0.8 1.4 0.8 -0.9 -2.0 1.0 -3.8 38 HUNGARY _ base year 1987 1989 1993 Shares 1.2 -1.3 'Wages 1.4 5.9 Non-wage private 1i:7 -0.6 Social transfers -1.3 -0.6 Pensions -0.9 1.4 Non-pension 0.2 -0.2 Total 4.2 2.2 Residual 1.1 -1.3 LATVIA _ _ _ l base year 1989 1992 1993 1995 1996 Shares -1.1 -1.0 -1.7 -1.6 Wages 1.2 4.2 16.4 15.0 Non-wage private -2.2 -2.1 -2.6 -1.5 Social transfers 1.2 0.9 1.3 1.4 Pensions -2.2 -1.9 -2.9 -2.0 Non-pension 0.1 0.1 0.3 0.5 Total 0.9 1.3 9.4 10.0 Residual 1.7 -0.6 -4.1 -3.3 39 REFERENCES Atkinson Anthony B. (1994), "Seeking to Explain the Distribution of Income", Discussion paper WSP 106, Centre for Economics and Related Disciplines, London School of Economics, December. Atkinson, Anthony B. and John Micklewright (1992), Economic Transformation in Eastern Europe and the Distribution of Income, Cambridge: Cambridge University Press. Atkinson, Anthony B., Lee Rainwater and limothy Smeeding (1995), "Income Distribution in Advanced Economies: Evidence from the Luxembourg Income Study", Luxembour Income Study (LIS) Working paper No. 12, Luxembourg, October. Blanchard, Olivier and J. Keeling (1996), "A. Numerical Model of Transition", mimeo, May 30, 1996, p.20. Commander, Simon, Andrei Tolstopiatenko and Ruslan Yemtsov (1997), "Channels of Redistribution: Inequality and Poverty in the Russian Transition", paper prepared for the Conference on inequality and poverty in transition economies, EBRD, London, May 23-24, 1997. Comia, Andrea Giovanni (1994), "Income Distribution, Poverty and Welfare in Transitional Economies: A Comparison between Eastern Europe and China", Innocenti Occasional Paper Economic Policy Series No.44, Florence, Italy:Unicef, October. Coulter, Fionna A.E., Frank A. Cowell and Stephen P. Jenkins (1992), "Equivalence Scale Relativities and the Extent of Inequality and Poverty", The Economic Journal, 102:1067-1082. Forster, Michael F. (1993), "Comparing Poverty in 13 OECD Countries-Traditional and Synthetic Approaches", LIS Working paper No. 103, August. Kordos, Jan (1994), "Outline of the Current Budget Surveys in Poland", mimeo. Kupa, Mihaly and Gaspar Fajth (1990), "Hungarian Social Policy Systems and Distribution of Income of Households", Hungarian Household Panel Survey (TARKI) Budapest, nimeo. Marnie, Sheila and John Micklewright (1993), "Poverty in Pre-reform Uzbekistan: What Do Official Data Really Reveal", mimeo, August. Michel, Richard C. (1991), "Economic Growth and Income Equality since the 1982 Recession", Journal ofPolicy Analysis andManagement, 6:2, pp. 181-203. 40 MNlanovic, Branko (1992), "Income Distribution in Late Socialism: Poland, Hungary, Czechoslovakia, Yugoslavia and Bulgaria Compared", Socialist Economies Reform Unit, Income Distribution during the Transition, paper No. 1, World Bank, Washington. Milanovic, Branko (1998), Income, Inequality and Poverty during the Transition from Planned to Market Economy, Washington, D.C.: World Bank. Popkin, Barry (1992), "Poverty in the Russian Federation: Demographics and Coverage by Current Support System", mimeo, September 23, 1992. Redor, Dominique (1992; first published in French 1988), Wage Inequalities in East and West, Cambridge: Cambridge University Press. Rutkowski, Jan (1995), "Labor Markets and Poverty in Bulgaria", A background paper for the World Bank poverty assessment, mimeo, October. Rutkowski, Jan (1996), "High Skills Pay off: The Changing Wage Structure during Economic Transition in Poland", Economics of Transition, vol.4, May, pp. 89-112. Ve6ernik, Jiri (1986), "Mzdova a prijmova diferencijace: analyza nekterych problemu", Institute for Philosophy and Sociology, Prague, mimeo. Veoernik, Jiri (1993), "Recent changes in household income distribution: the Czech republic 1988-1993", mimeo. Ve6emik, Jiri (1994), "Changing Earnings Inequality under the Economic Transformation: The Czech and Slovak Republics in 1984-92", mimeo. Vodopivec, Milan and Peter Orazem (1995), "Winners and Losers in Transition: Returns to Education, Experience and Gender in Slovenia", The World Bank Economik Review, vol. 9, No. 2, May. 41 Policy Research Working Paper Series Contact Title Author Date for paper WPS1911 The Internationalization of Financial Stijn Claessens April 1998 R. Vo Services in Asia Tom Glaessner 33722 WPS1912 Pay and Grade Differentials at the Daon Filmer April 1998 C. Bernardo World Bank Margaret Grosh 31148 Elizabeth King Dominique van de Walle WPS1 913 The 1994 Currency Crisis in Turkey Cya Celasun April 1998 K. Labrie 31001 WPS1914 Distinguishing between Types of Jesko Hentschel April 1998 PREM Aclvisory Data and Methods of Collecting Them 87736 WPS1915 Distortionary Effects of State Trading Merlinda Ingco April 1998 M. Fernandez in Agriculture: Issues for the Next Francis Ng 33766 Round of Multilateral Trade Negotiations WPS1916 The Size, Origins, and Character James H. Anderson May 1998 P. Sintim-Aboagye of Mongolia's Informal Sector 37656 during the Transition WPSI917 Financial Liberalization and Financial Asli Demirguc-Kunt May 1998 P. Sintimn-Aboagye Fragility Enrica Detragiache 37656 WPS1918 How Does Foreign Entry Affect Stijn Claessens May 1998 R. Vo the Domestic Banking Market? Asli Demirgu9-Kunt 33722 Harry Huizinga WPS1919 The Empirical Effects of Performance Mary Shirley May 1998 P. Sintirn-Aboagye Contracts: Evidence from China Lixin Colin Xu 38526 WPS1920 Education and Earnings in a Peter R. Moock May 1998 M. Christian Transition Economy (Vietnam) Harry Anthony Patrinos 36736 Meera Venkataraman WPS1921 Making Voice Work: The Report Card Samuel Paul May 1998 C. Berriardo on Bangalore's Public Service 31148 WPS1922 Regional Groupings Among Soamiely Andriamananjara May 1998 L. Tabada Microstates Maurice Schiff 36896 WPS1923 When Vintage Technology Makes Giorgio Barba Navaretti May 1998 L. Tabada Sense: Matching Imports to Skills Isidro Soloaga 36896 Wendy Takacs WPS1924 Voucher Privatization with David Ellerman May 1998 M. Murray Investment Funds: An Institutional 36095 Analysis Policy Research Working Paper Series Contact Title Author Date for paper WYPS1925 Half a Century of Development Jean Waelbroeck May 1998 J. Sweeney Economics: A Review Based on 31021 the Handbook of Development Economics WPS1926 Do Budgets Really Matter? Emmanuel Ablo June 1998 K. Rivera Evidence from Public Spending Ritva Reinikka 34141 on Education and Health in Uganda WPS1927 Revenue-productive Income Tax Fareed M. A. Hassan June 1998 A. Panton Structures and Tax Reforms in 85433 Emerging Market Economies: Evidence from Bulgaria WPS19'28 Combining Census and Survey Data Jesko Hentschel June 1998 P. Lanjouw to Study Spatial Dimensions Jean Olson Lanjouw 34529 of Poverty Peter Lanjouw Javier Poggi WPS1929 A Database of World Infrastructure David Canning June 1998 A. Abuzid Stocks, 195G-95 33348 WPS1930 The Main Determinants of Inflation in liker Domac June 1998 F. Lewis Albania Carlos Elbrit 82979 VJPS1931 T he Cost and Performance of Paid Ariel Dinar June 1998 F. Toppin Agricultural Extenion Services: The Gabriel Keynan 30450 Case of Agricultural Technology Transfer in Nicaragua NV2S1932 Air Pollution and Health Effects: Bart D. Ostro June 1998 C Bernardo A Study of Respiratory Illness Gunnar S. Eskeland 31148 Among Children in Santiago, Chile Tarhan Feyzioglu Jose Miguel Sanchez WPS 1933 The 1997 Pension Reform in Mexico Gloria Grandolini June 1998 C. Zappala Luis Cerda 87945 WPS1934 WTO Accession for Countries Constantine Michalopoulos June 1998 L. Tabada in Transition 36896 WPS1935 EIxplaining the Increase in Inequality Branko Milanovic June 1998 G. Evans during the Transition 85734 \NPS1O36 Determinants of Transient and Jyotsna Jalan June 1998 P. Sader Chronic Poverty: Evidence from Martin Ravallion 33902 Rural China VVPS1937 Aid, the Incentive Regime, and Craig Burnside June 1998 E. Khine FPoverty Reduction David Dollar 37471