Programmatic Poverty Assessment Work in Western Balkans Serbia Poverty Assessment

Transcription

1 Background Paper on Trends and Profile of Poverty in Serbia: Programmatic Poverty Assessment Work in Western Balkans Serbia Poverty Assessment Gorana Krstić and Victor Sulla June 2007

2 1. Introduction: 1.1 The Poverty Reduction Strategy (PRS) paper adopted by the Government of Serbia in 2003 outlines a medium-term development framework directed at reducing poverty in the country. It is a national document that contains an analysis of causes, characteristics, and profile of poverty in Serbia, as well as highlights the key government policy initiatives to tackle its adverse consequences. For poverty monitoring over the PRS implementation period ( ) it was planned to make a shift from the Living Standards Measurement Surveys (LSMS), conducted in Serbia in 2002 and 2003 with support from the World Bank, to the official Household Budget Survey (HBS) conducted by the Republic Statistical Office (RSO). The reformed HBS was started in March 2003, with support from the Swedish International Development Cooperation Agency (SIDA), and a number of improvements in survey methodology and data collection procedures have subsequently been introduced over the years. The latest HBS (i.e round) is compliant with Eurostat recommendations for expenditure and income statistics, and provides detailed picture of household income and expenditures in Serbia, along with general data on the demographic structure of households and housing conditions. Data for the HBS are collected throughout the year (every month on a 1/12 of the sample), with parallel data entry, allowing quarterly monitoring of results. The annual sample of HBS is about 4,300 households in total. 1.2 The main objectives of this note are two-fold: (i) using data from the latest (i.e. 2006) HBS, it presents a profile of the poor in Serbia; and (ii) by comparing data from the 2004 and 2006 HBS, it attempts to assess in broad terms the impact of recent economic growth on living conditions in Serbia. In the case of the latter i.e. analysis of changes in living conditions since 2004 it is important to highlight upfront that the changes in HBS survey methodology over the years render the task of constructing fully comparable welfare indices for the various survey rounds a very difficult and challenging undertaking. Given this caveat, the results presented in the latter half of this note should therefore be seen at best as broadly indicative of changes in living standards rather than as precise estimates per se of changes in poverty incidence. A more definitive assessment of the latter i.e. change in the poverty headcount rate in recent years for Serbia will only be possible once new data from the 2007 LSMS and 2007 HBS become available, and can be compared with fully comparable welfare aggregates based on data from earlier rounds of these respective surveys. 1.3 The analysis of the HBS data used in this note, particularly the data for the 2006 HBS, has involved working collaboratively with RSO as well as the Serbia PRS unit, 1 and is part of the ongoing collaboration between the World Bank and the Government of Serbia on analysis of issues related to growth, poverty, and inequality linkages in the context of the Western Balkans Programmatic Poverty Assessment work. Details pertaining to the survey methodology and poverty analysis carried out using the HBS data are described in more detail in a companion paper. Section 2 of this note presents the main findings of the poverty profile for Serbia based on data from the 2006 HBS. Section 3 then presents the results of a more rigorous analysis carried out using these same data to identify some of the main factors closely associated with poverty incidence in the country. Section 4 summarizes the main findings of the tentative inter-temporal analysis of changes in living conditions in Serbia based on data from the 2004 and 2006 HBS rounds. Finally, Section 5 presents some interesting findings regarding the sources of income of households in Serbia, both for the overall population as well as among the poor. 1 In particular, we would like to thank Mr. Vukmirović and his staff at RSO and Ms. Aleksandra Jović and her staff at the PRSP unit in Belgrade. 2 Please see Božanić, Mijakovac, Cvetinović, Ćalasan, and Krstić: Poverty in Serbia in the Year 2006 Preliminary Results, Republic Statistical Office, Belgrade, 2007 for more details. The poverty estimates presented in this note differ slightly from those in the companion paper due to very small differences in the exact procedure followed in estimation of the poverty lines used (thus, while the poverty rate for Serbia is estimated to be 8.8 percent in Serbia in 2006, it is 9.0 percent in the current note i.e. a difference of 0.2 percent at the national level. However, the analysis presented in both notes is essentially the same for all substantive purposes.

3 2. Poverty Profile for Serbia based on the 2006 HBS 1.4 A poverty profile describes who the poor are by indicating the probability of being poor according to various characteristics, such as the place of residence of the household (location and region), the labor market status of the household head and household members, the demographic composition of the household (i.e., gender, age, household size, number of children), and the amount of land the household possesses. This section provides a profile of the poor with respect to those characteristics, as well as the multivariate analysis of poverty. Poverty profile will be presented using the data from the 2006 HBS and will be compared broadly with the estimates based on LSMS. The data are not strictly comparable, however, as the LSMS data provided a benchmark assessment of poverty situation, the comparison to this source is inevitable. First, we summarize very briefly the methodology followed to derive the poverty line used in estimating poverty based on the 2006 HBS. 1.5 The absolute poverty line was constructed, jointly with RSO, using the 2006 HBS data and is estimated at 6,411 Serbian Dinar per RSO equivalent per month. Following established practice in Serbia and elsewhere, 3 the absolute poverty line method was estimated using the cost-of-basic-needs approach. A minimal daily calorie intake of 2,288 calories per adult equivalent per day was used as the reference level, and the cost of the food basket providing this nutritional level was estimated to yield the food poverty line. Clearly, non-food basic needs should be accounted for as well when calculating the poverty line. The cost of non-food basic needs is estimated on the basis of calculating the food share for households with adult equivalent food consumption in the neighborhood of the food poverty line. The calculated food share is 66.6 percent. The overall absolute poverty line is then estimated in 2006 at 76,922 dinars per adult equivalent. The poverty estimates for Serbia corresponding to the application of this methodology using data from the 2006 HBS are presented in Table 1. Table 1: Poverty Estimates for Serbia, 2006 (%) Measure of welfare 2006 Absolute Poverty line (per RSO equivalent adult, in Dinars) 76,922 Headcount Index of Poverty (P0) 9.0 Poverty Gap Index (P1) 2.0 Poverty Severity Index (P2) 0.8 Source: World Bank estimates using 2006 Household Budget Survey data. Applying this poverty line to 2006 HBS data, we turn next to examining the profile of starting first by examining its regional dimensions. Regional Dimension poverty in Serbia, 1.6 Poverty in Serbia is an increasingly rural phenomenon, like in many transition countries. Poverty was significantly higher among the rural than urban population in 2006 (13.9 percent versus 5.2 percent). Poverty is not only more prevalent among rural population, it is also deeper and more severe (Table 2). Although the data from HBS and LSMS data are not fully comparable, it seems that urban-rural differences in poverty have increased considerably since The ratio of rural to urban poverty increased from 1.8 to 2.7 over In 2006, around two thirds of the poor population lived in rural areas. 3 For instance, this method was also used in Serbia when calculating the poverty line used with the 2003 LSMS.

4 Table 2: Poverty Measurem ent by Type of Location in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribu tion of popul ation Poverty gap Severity of poverty Urban Standard error Rural Standard error Total Standard error Source: World Bank estimates using 2006 Household Budget Survey data. 1.7 Serbia is a country with large, persistent and increasing regional disparities in economic development. According to the recently adopted National Strategy for Economic Development of the Republic of Serbia (2006), Serbia appears to be the country with the largest regional economic disparities among European countries, which have even increased in recent period. 4 The process of transition to a market economy further deepened the existing economic disparities across regions due to the collapse of many large socially-owned enterprises, intensified economic restructuring and privatization. The traditionally underdeve loped region of South Serbia was accompanied by new regions with low economic performance (Eastern Serbia and some parts of Central Serbia; regional centers of mining and manufacturing in Western Serbia). 5 Table 3: Poverty by Regions and Type of Location in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribution of population Belgrade Urban Rural Vojvodina Urban Rural Central Serbia* Urban Rural Total * Excluding Belgrade. Source: World Bank estimates using 2006 Household Budget Survey data. 1.8 Regional distribution of poverty is in line with the observed regional economic disparities. Poverty incidence varied considerably across different parts of the country, ranging from 4.7 percent in Belgrade to 11.1 percent in Central Serbia (Table 3). Regional dimension of poverty is analyzed using the data for 4 The ratio of the most developed to the least developed municipality in Serbia measured by the national income per capita increased, from 1:19 in 2000 to 1:26 u Regional disparities between developed northern parts of the country and undeveloped southern parts can be clearly seen from the following economic characteristics of these regions: (a) Belgrade with 21 percent of overall population, employed 22 percent of all workers which contributed to 26 percent of the value added in industry; (half of the highly educated workers are employed in Belgrade); (b) underdeveloped parts of Central Serbia regions (Toplički, Jablanički and Pčinjski okrug, Stari Ras) employed 8 percent of all workers which contributed to only 5 percent of the value added in industry (National Strategy for Economic Development of Republic of Serbia, 2006).

5 only three regions in Serbia (Belgrade, Vojvodina and Central Serbia) and urban/rural desegregation within these regions, as the HBS data are not representative at the lower regional level. 1.9 Belgrade, the capital city of Serbia, where most of the economic opportunities are concentrated, remains better-off than the rest of the country. Central Serbia (without Belgrade) remains the poorest region in Serbia and it was over twice worse off than Belgrade. Vojvodina remained in between these two extremes; it is the region with the incidence of poverty below the national average (8.6 percent versus 9.0 percent respectively) but with large disparities between urban and rural areas. These disparities are evident in all three regions, with rural to urban ratio of poverty incidence ranging from around 2 for Belgrade and Central Serbia to above 3 for Vojvodina. Rural parts of Central Serbia and Vojvodina faced the highest poverty incidence (14.8 percent and 14 percent respectively) The ranking of regions using HBS data for 2006 appears comparable with the estimates based on the LSMS for 2002 (World Bank, 2003). However, the disparities between urban and rural areas within the regions have further deepened, particularly in Vojvodina (the ratio of urban to rural poverty incidence was 1.7 in 2002). In terms of the distribution of the poor across regions, most of the poor are concentrated in Central Serbia (63 percent), which is the largest Serbian region (52 percent of overall population). In two most depressed region of Serbia, rural parts of Central Serbia and rural parts of Vojvodina residing 46 percent and 18 percent of poor population respectively The large disparities in regional poverty incidence could be possibly linked with enterprise restructuring, higher unemployment rates and lower wages in Central Serbia compared to Belgrade. According to the LFS for 2006, unemployment rate (population 15-64) varied across regions in Serbia, from 17 percent in Belgrade to 25 percent in Central Serbia. 6 Analysis of wages at the district level (World Bank, 2006) also shows significant regional differentials - wages in Belgrade were the highest, while in most districts in Central Serbia the lowest. In addition, farm income, which is inferior to wages as a source of livelihood, is much more important source of income for the population in Central Serbia than in Belgrade. Multivariate analysis of poverty will show net regional differences in consumption, once we control for education, demographics and labor market characteristics of a household The large regional disparities may indicate low labor mobility and poor investment climate in many parts of the country. The poorly developed housing market across regions and weak communication systems, which make commuting between some regions extremely difficult and time consuming, may also contribute to low territorial labor mobility. The unemployed and workers are often unwilling to move to regions with better job prospects due to problems in finding affordable housing, but also because of the cost of relocating, the risk of rupturing social support networks, and uncertainty about their own job prospects (World Bank, 2003). Labor market status 1.13 Employment status is strongly correlated with poverty. Households with unemployed household heads faced the highest incidence of poverty among Labor market participants. Their poverty incidence was 15 percent, compared to the average poverty incidence of 9 percent (Table 4). However, the population living in those households represents only about 12 percent of the total poor adults. In contrast, households with employed household heads seem to have the lowest poverty risk, which is not surprising given the large increase in wages (see Table 14 in section 4). Households with self-employed heads are more exposed to poverty risk than households with employed heads and also more exposed to poverty than the average households (11 percent versus 9 percent). It should be noted that the labor market status is defined according to self-declaration of respondents in the HBS, and therefore it does not correspond to the ILO definition of employed, unemployed and inactive. 6 See RSO Report on Labor Force Survey for 2006, No.59, March 2007.

6 1.14 Amongst households with inactive household head, the incidence of poverty is 11 percent, and individuals residing in such households represent almost a half of the total poor. Among non-participants, the highest poverty risk faced households with heads being non-participants other than pensioners (i.e., housewives, students etc.). Households with pensioner heads have poverty incidence at the level that is slightly below the national average, but they represent a sizeable group among the poor (36 percent). Table 4: Poverty by Labor Market Stat us of the Household Head in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribution of population Participants Wage-employed Self-employed Unemployed Non-participants Pensioners Other non-participants Total Source: World Bank estimates using 2006 Household Budget Survey data As the well-being of individuals depends not only on the labor market position of the household head but more broadly on the degree of employment of all household members, it is necessary to consider the employment profile of the overall household. Households are grouped into three categories according to the working-age and Labor market status of all household members: jobless households (active age household which has no working members), non-working households (all members fall into one of the following categories: children aged less than 18; persons aged in education and inactive; persons aged over 65 and not working) and working households (active age household which has at least one employed member) Table 5 shows that jobless households have the highest poverty incidence (16.7 percent), while non-working households fare somewhat better (poverty rate is 14.3 percent). However, the majority of the poor ( 63 percent) in Serbia still live in working households, although their group poverty risk is below the national average. The same pattern was observed using the 2002 LSMS data (World Bank, 2003). More in-de pth analysis of labor market and poverty linkages will be possible using the LSMS data for 2007, as it contains standard Labor market module. Table 5: Poverty by Labor Market Status of Household Members in Serbia, 2006 (%) Poverty incidence Distribution of Distribution of poor population Working household Jobless household Non-active Total Source: World Bank estimates using 2006 Household Budget Survey data Table 6 shows poverty incidence according to the main source of household income. The results are consistent with poverty by labor market status of household head, with households having labor income as main source of income being least exposed to poverty risk and households with other income (income other than from labor and social transfers) being most exposed to poverty risk. Among households with labor income, those employed in the public sector are better off than those working in the private sector, which is not surprising given the fact that wages in the public sector were higher than in the private sector according to the LFS data for 2005, even after controlling for individual characteristics (World Bank, 2006). Households with main source of income from farm and non-farm activities

7 (household business outside agriculture) appear worse off compared to households with other labor income, which is in line with observed low level of income from agriculture compared to other sources of incomes. Only 1 percent of all households reported that their main source of income come from household business (outside agriculture). Table 6: Poverty by the Household s Main Source of Income, 2006 (%) Poverty incidence Distribution of poor Distribution of population Labor income Public sector Private sector F arm and non-farm activities Social transfers (pensions, other transfers) Other income Total Source: World Bank estimates using 2006 Household Budget Survey data However, the next table 7 shows that 9 percent of all households reported that their household is engaged in some business. They are significantly less likely to be in poverty than other households (6.2 percent versus 9.3 percent, respectively) indicating that non-agricultural self-employment can be highrewarded strategy during transition, as in many advanced transition economies of Central Europe (Dutz, M. at. al, 2004). This seems to be true especially for households with registered business, as their poverty incidence is very low (2.7 percent), and considerably lower than for households with non-registered business, which constitutes 4.6 percent of all poor. Table 7: Poverty by Type of Household Business in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribution of population Business of household Registered Not registered Other households with no business Total Source: World Bank estimates using 2006 Household Budge t Survey data. Table 8: Poverty by Occupation of Household Head in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribution of population Managerial Professionals Technicians and associate professionals Clerks Service workers, shop and market sales Skilled agricultural and fishery workers Craft and related trade workers Plant and machine operators and assembly Elementary occupations Total Source: World Bank estimates using 2006 Household Budget Survey data The poverty differences between occupations (Table 8) show very low poverty incidence in occupations that require high education, such as managers and professionals (about 1 percent), while

8 considerably higher incidence of poverty are observed in elementary occupations (21.4 percent) and among agricultural workers (13.1 percent). This suggests a strong link between poverty risk and educational attainment, which will be analyzed in the next paragraph. Education and Demographics 1.20 Highly educated people are not exposed to poverty risk. Differences in educational attainment of heads of households are reflected in considerably different poverty rates (Table 9). Less educated people were more likely to be poor. Households with heads with elementary school or those with no education had a 22 percent poverty rate, which is the highest rate among all education groups. The poverty rate significantly declines as the level of education attainment by a household head increases. Having completed elementary school brings down the probability of being in poverty to 13.8 percent. Only these two groups faced poverty incidence above the national average, however, they made a large share of the poor (67 percent). In contrast, households with highly educated heads (post secondary education and university) had poverty incidence close to zero (1.2 and 1.8 percent respectively) and they made 2.6 percent of all poor. This pattern is also observed in Table 9: Poverty by Education of Household Head in Serbia, 2006 (%) Poverty inci dence Distribution of poor Distribution of population Incomplete elementary or less Elementary Vocational (1-2 years) Secondary (3 yrs. vocational) Secondary (4 yrs. vocational), gymnasium Post secondary University Total Source: World Bank estimates using 2006 Household Budget Survey data Households with small children appear to be more affected by poverty. The incidence of poverty for children below 7 was 54 percent above the national average (Table 10). They constituted 5.8 percent of the overall population and 9 percent of the poor. The next age group, children 7-13, also had poverty incidence significantly above the national average. Poverty incidence decreased with the age of the individual up to the group of elderly. The elderly faced above the average incidence of poverty, but comparison with 2002 suggests that they may have experienced some improvements compared to other age groups. This could be c hecked comparing the results from LSMS over with the new LSMS data for 2007 (using the comparable consumption aggregate and equivalence scales). Table 10: Poverty by Age Groups in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribution of population Children Children Youth Youth Adults Adults Older Total Source: World Bank estimates using 2006 Household Budget Survey data.

9 Improved economic conditions of elderly were observed in all transitional countries over the last five years (Alam et al., 2005). Other age groups had below average incidence of poverty. However, it should be noted that these results are sensitive to assumptions about economies of scale (Lanjouw and Ravallion, 1995) An increase in the number of small children is related to an increase in the poverty incidence. The presence of children below 7 years of age increased poverty risk above the national average. Households with three or more small children experienced the highest incidence of poverty (Table 11). However, this group represents a small fraction of all poor (5.4 percent), as only about 1 percent of population residing in these households. The largest fraction of all poor refers to households without small children, as they r epresent almost two thirds of all poo r. Table 11: Poverty by Demographic Composition of Household in Serbia, 2006 (%) Poverty incidence Distribution of poor Distribution of population Number of children 0-6 yrs or more Household size or more Total Source: World Bank estimates using 2006 Household Budget Survey data Table 11 also shows that only large households with 6 or more members have the incidence of poverty above the national average and the highest incidence of poverty among all demographic groups. They comprised 19 percent of the population and 38 percent of the poor. The important factor in explaining poverty in extended families is dependency ratio. The dependency rate measures the number of household members of non-working age (children and elderly) that have to be supported by the household s working members. Larger households have more children and, thus, lower ratio of income earners than smaller households, which causes lower levels of their consumption Households with three or four members appear to be better off than others. However, most of these households are working households (although most of them have one or two children), which were less exposed to poverty risk as indicated in Table 5. This is also observed using the 2002 LSMS data; three and four memebers households had the lowest poverty incidence (Bogićević at. al, 2003). However, all these results should be treated with caution since the outcomes largely depend on the assumptions made regarding the equivalence scales and economies of scale (Lanjouw and Ravallion, 1995). Ownership of land 1.25 Access to land has not protected rural households from falling into poverty, however, large land holdings considerably reduces probability of being poor in rural areas. Rural households with no land holdings and those with smaller land holdings (less than 1 hectare) experienced the highest incidence of poverty (about 16 percent). These two groups represent 54 percent of all poor in rural areas (Table 12). The risk then declines with the size of land holdings, however, only households with the land holdings

10 over 3 hectares were less likely to be poor compared to the average. However, only one fifth of rural households possess plots of land over 3 hectares. The average size of farm is about two hectares. Table 12: Poverty by Land Size in Rural Areas in Serbia, 2006 (%) Po verty incidence Distribution of poor Distribution of population 0 hectares Less than 1 hectare hectares Over 3 hectares Total Source: World Bank estimates using 2006 Household Budget Survey data In addition to small size of farm land and its fragmentation with an average of seven parcels per farm, there are a number of constrains to increased productivity in agricultural sector and increased farm incomes in rural areas, which can protect many rural population from poverty, such as: obsolete agricultural equipment (average age of it is about 20 years), no irrigation system, no or very little access to financing, undeveloped infrastructure, etc. 3. Multivariate Poverty Profile 1.27 A poverty profile is a useful descriptive method, but it cannot be used to gauge the net association between a household s characteristics and the probability of a given household being in poverty. For this purpose, regression analysis that accounts for the correlation between background characteristics is most useful. For example, a person with low education living in a rural area is likely to have a low-paying casual-wage agricultural job. Regression analysis helps to purge out the multivariate correlation across these factors to find the net effect of each of them. Finally, it is also more efficient to analyze a linear relationship between a log of consumption per adult equivalent and a set of household characteristics, rather than a relationship between a categorical outcome defined as poor/non-poor and a set of determinants Identifying factors that are closely associated with welfare and poverty is an important step in designing economic and social policy aimed at reducing poverty and preventing households from falling into poverty. The examined factors are the same as those examined in the poverty profile and comprise: (i) characteristics of the household including age composition, size, location of the household, business activity, size of agricultural land; as well as: (ii) characteristics of the household head such as age, gender, education and labor market status. These factors are used as explanatory variables in a simple regression model, where consumption per adult equivalent represents dependent variable (Table 13). 7 The following factors were estimated as significantly related to consumption per adult equivalent: 1.29 Household demographics Household size had a negative impact on household consumption in 2006: larger households had lower consumption, being similar in all other characteristics. Small children and elderly had significant negative effect on consumption. The larger the share of children up to seven years old in the household, the lower the consumption of the household relative to the base category (the share of female between 19 and 65 years of age), keeping the household size constant. The same pattern is observed for the elderly. The share of other age groups in the household did not affect consumption. 7 The model is estimated using the standard OLS procedure with robust standard errors.

11 Table 13: Serbia: Determinants of consumption, 2006 (Dependent variable: ln(consumption per adult equivalent) coef sd Household characteristics Log of household size (0.022)*** Share of children (0.093)** Share of children Share of male adults Share of female adults (dropped) Share of Elderly (>=65) (0.038)*** Characteristics of household head Gender and age Log of household head' s age Male Edu cation Incomplete elementary or less (0.030)*** Elementary (drop ped) Vocational (1-2 years) (0.043)*** Secondary (3 yrs. vocational) (0.030)*** Secondary (4 yrs. vocational), gymnasium (0.030)*** Post sec ondary (0.035)*** University (0.037)*** Labor market status Self-employed Unemployed (0.035)*** Inactive (0.025)*** Household business Regi stered (0.040)*** Unreg istered (0.059)* No household business (dropped) Agricultu ral land size 0 ha (drop ped) less than 1 ha (0.026)* 1-3 ha (0.030)*** over 3 ha (0.034)*** Regions Belgrade (0.021)*** Vojvodina Central Serbia (dropped) Urban (0.021)*** Constant (0.198)*** Observations 4558 Adjusted R-squared * significant at 10%; ** significant at 5%; *** significant at 1%

12 1.30 Education and gender Consumption was considerably higher in households whose heads were more educated. Households headed by individuals with post secondary education (collage) had 44 percent higher consumption level than those headed by individuals with elementary education (reference category), controlling for other characteristics. Moreover, households with heads holding university degree on average had consumption level 71 percent above those headed by individuals with elementary education. These results are in line with the observed considerable wage advantage for highly educated workers (collage or more), after controlling for other individual characteristics (World Bank, 2006). Gender of the household head had no significant effect on consumption, being similar in all other characteristics. This feature is also observed in the poverty profile (without any controls). This result is consistent with relatively small gender pay gap observed for Serbian workers in comparison with other transitional countries (Krstić, 2004) Labor market participation Unemployment of the household head significantly reduces consumption (increasing poverty). Individuals living in households with a head who was unemployed reported 17 percent lower consumption than wage-employed heads, controlling for other household characteristics. This feature is observed even if we do not control for any characteristics (see table 4 in the poverty profile). Those living in households with inactive heads reported 6 percent lower consumption than wage-employed heads. Having a household business significantly increases consumption, especially if the business is registered. Individuals living in households who had registered business experienced 32 percent increase in consumption relative to those without household business, controlling for other characteristics. This effect is much smaller if the business is not registered (consumption increases by 10 percent) Size of agricultural land Use of larger land holdings significantly increases household consumption. Individuals living in households with land holdings over 3 hectares were much better-off than others; their consumption was 32 percent higher relative to households with no land holdings Household location Location plays an important role in explaining household welfare in Serbia. The significant location effects on consumption remain after controlling for all other household characteristics. Residents of Belgrade had 6 percent higher consumption compared to the residents of Central Serbia, controlling for other household characteristics. Urban residents had 7 percent higher consumption compared to rural residents. However, the results also suggest that regional disparities in consumption are much smaller than presented in the poverty profile (Table 3).

13 4. Growth and Poverty in Serbia: Preliminary Findings based on HBS 1.34 Economic recovery in Serbia increased GDP growth to 9.3 percent in 2004, 6.3 percent in 2005 and 5.5 in 2006, respectively (Table 14). Average earnings grew even more rapidly, increasing by around 48 percent in real terms. However, overall unemployment rates in Serbia remained quite high over this period and may have dampened the impact that robust growth may otherwise have had on reducing poverty. Table 14: Selected Economic Indicators fo r Serbi a, National accounts Real GDP growth, % GDP per capita, US $ 863 1,413 1,910 2, 542 2,977 3,234 3,642 Gross domestic investme nt Gross domestic savi ngs (% GDP) Prices and wages Consumer price index, end-year, % Growth of net real earnings, % Employment growth (%) Source: World Bank Live Database What can the HBS data series tell us about changes in the average standard-of-livinrecent years? As briefly noted earlier, the HBS survey methodology has improved considerably in recent in Serbia in years. The 2006 HBS is compliant with Eurostat standards, and provides a detailed picture of household income and expenditures in Serbia. However, changes in survey methodology implemented between 2003 and 2006 make it very difficult to derive fully comparable welfare aggregates spanning this period. In particular, when analyzing the data for the 2003 HBS, we found significant inconsistencies in the variable definitions and sampling coverage in relation to the later HBS surveys, so we did not use this survey in the analysis. In addition, since differences in survey estimates across two years are less influenced by white noise as compared to comparisons based on surveys that are only one year apart, we focused in particular on comparing the 2004 and 2006 HBS We start first with carrying out a few consistency checks on data from the 2004 and 2006 HBS, the results of which are reported in Annex I. To summarize briefly, the main findings of these consistency checks are that (i) price changes indicated by the HBS series between 2004 and 2006 correspond well to trends in the overall price level reported by the CPI, (ii) the population covered in the two surveys is similar in terms of age and gender composition, family size, regional breakdowns, etc. i.e. the two surveys are quite consistent in terms of demographic composition, and (iii) growth rates in labor and total income between the two surveys are generally similar to the consumption per capita growth, which increased by about 9 percent in real terms between 2004 and However, as can be seen in Table 15, this overall growth rate in aggregate expenditures hides considerable variation in growth rates of different components of expenditures between 2004 and For instance, while total non-food expenditures increased by about 13 percent, total food expenditures fell by about 4 percent in real terms over this period. Within non-food expenditures, average expenditures in sub-categories like accommodation, alcohol, education, and housing increased by 3 11 percent over this period, while other categories increased at a much faster rate, notably recreation and restaurants which increased by more than 40 percent. While this overall pattern of expenditures is broadly consistent with what one might have expected with increasing incomes and rising standards of living, there are nonetheless some anomalous trends that are harder to explain. For instance, one would indeed expect the

14 share of spending on food items to decline with rising incomes, as is indeed the case between 2004 and However, if average living standards did indeed improve during this period as the overall 9 percent increase in average expenditures per capita in real terms indicates it is difficult to reconcile this with the observed decline in level of food expenditures (these declined from 36,526 dinars per person in 2004 to 35,229 dinars per person in 2006). Table 15: Expenditure per ca pita in Real Te rms by Expenditure Components Component Ratio 2006/2004 Food 36,526 35, Non-Food 52,220 59, Clothing 5,662 6, Housing 12,816 14, Furnishing 3,875 4, Health 3,120 4, Transport 7,585 9, Communication 4,862 3, Recreation 2,970 4, Education 1,509 1, Restaurants 940 1, Accommodation Alcohol 1,244 1, Miscellaneous 4,426 4, Total Consumption per capita (yearly) 85,818 93, Source: World Bank estimates using 2004 and 2006 Household Budget Survey data Clearly, given the somewhat ambiguous nature of findings regarding the comparability of the welfare aggregates derived from the 2004 and 2006 HBS data, the appropriate degree of caution should be exercised when looking at welfare comparisons based on these recent HBS series. Rather than precise point estimates per se of changes in poverty incidence, analysis based on the HBS series should at best be seen as only broadly indicative of changes in living standards over time. To derive poverty estimates for 2004 that are broadly comparable with those presented earlier in this note, the poverty line calculated using the 2006 HBS data (76,922 dinars per equivalent adult in 2006 prices) was adjusted to 2004 prices (i.e. holding its real value constant) to yield a poverty line of 59,202 dinars per equivalent adult for Using this poverty line for 2004 and comparing the derived poverty estimates with those for 2006, we find a significant reduction in poverty in Serbia between 2004 and 2006 according to all three main commonly used po verty indices ( Box 1): for instance, the HBS data series show the headcount p overty rate in Serbia to have declined for 14.6 percent in 2004 to 9.0 percent in 2006 (Table 16). Poverty appears to have fallen quite rapidly in both urban (from 10.4 in 2004 to 5.2 percent in 2006) and rural areas (from 20.2 to 13.9 percent), an d in both Belgrade as we ll as in other urban areas of Serbia. Table 16: Poverty Estim ates for Serb ia: 2004 and 2006 HBS Headcount Rate(P0) Poverty Gap(P1) Squared Poverty Gap(P2) Urban Belgrade Other Urban Rural Total Source: World Bank estimates using 2004 and 2006 Household Budget Survey data.

15 Box 1: Measures of Poverty The headcount index (P0) is a measure of the prevalence of poverty. It denotes the percentage of households who are poor as defined by the poverty line as a proportion of total population. This measure was insensitive to the distribution of the poor below the poverty line. q P 0 = ; n The poverty gap index (P1) is a measure of the depth of poverty, and it denotes the gap between the observed consumption levels of poor households and the poverty line. Assuming perfect targeting of resources (transfers), this poverty gap index indicates the total amount needed to bring all households in poverty up to the poverty line. 1 z = y P1 i ; n z i Q ( ) The poverty severity index (P2) measures the degree of inequality in distribution below the poverty line, giving greater weight to households at the bottom of the consumption distribution. 1 P2 = n i Q ( z ) z y i 2 2 ; where n = total population, q = number of those with consumption y i less than the poverty line z Moreover, these indications of improved living standards in Serbia in recent years do not seem to depend on the choice of poverty line used per se. The cumulative distribution function graph shows the fraction of the population with consumption less than a given level (Figure 1). This graph shows the poverty incidence for different poverty lines. A lower vertical line (i.e. closer to the y axis) corresponds to a lower poverty rate. Given that the curve for 2006 HBS per capita expenditures lies everywhere to the right of the corresponding curve for the 2004 HBS (at least up to around 128,000 dinars on the x-axis i.e. considerably higher than the poverty line used in this analysis), clearly there is less poverty in 2006 than in 2004, at least according to these welfare aggregates. As Figure 1 shows, the poverty in 2006 appears to be lower compared to 2004, regardless of the poverty line used (at least for most of the cumulative distribution shown). Moreover, as the other two graph panels also presented in this figure clearly show, this is true for both the urban and the rural population of Serbia.

16 Figure 1: Cumulative Distribution Functions Serbia, 2004 and 2006 HBS 1 Total 1 Urban Cumulative distribution Cumulative distribution Rural Welfare indicator, ' Welfare indicator, '000 Cumulative distribution Welfare indicator, ' Sources of Income of the Poor The HBS series is relatively stronger in terms of its coverage of household expenditures, as the household income part of the questionnaire is relatively less well-developed. Nonetheless household income aggregates were created and used as complementary information for the household expenditures. The survey collects information on 25 main income sources for each household. An average household income is composed 54 percent of its total income from labor and the rest 46 percent of its income from home-produced goods, particularly food and agricultural products and social transfers, mostly pensions, and private transfers There are slight differences between rural and urban areas in income sources (Table 17). Urban households derive 71.4 percent of their income from wage employment, while rural households derive only 62.7 percent of their income from labor income. Private transfers consist largely of transfers from relatives within Serbia, and reported remittances from abroad consist of a mere percent of household income in both urban and rural areas. Among the poor, the share of labor income is relatively lower and social transfer share is higher compared to the overall population. Social transfers to poor people in urban areas is much higher than in rural areas, however other types of income including agriculture and self employed income is higher in the rural areas. During period labor income grew by 9 percent in real terms, which corresponds well to the 7.4 percent growth observed for total expenditures. However, average social transfers, grew by only 1.4 percent during this period.

17 Table 17: Sources of Income for Urban and Rural Households Total Poor Total Poor Total Labor 68.6% 59.1% 66.6% 60.7% Social Tr ansfers 25.3% 34.6% 26.0% 30.0% Gifts 0.7% 0.7% 1.0% 0.6% Transfer Abroad 2.5% 1.8% 1.7% 1.6% Other 3.0% 3.8% 4. 7% 7.1% Urban Labor 71.4% 58.4% 68.3% 56.9% Social Transfers Gifts 24.5% 0.9% 34.7% 0.8% 26.8% 0.8% 36.3% 0.6% Transfer Abroad 1.4% 2.0% 1. 1% 2.6% Other 1.8% 4.1% 3. 0% 3.5% Rural Labor 62.7% 59.8% 63.6% 62.6% Social Transfers 26.8% 34.5% 24.5% 26.8% Gifts 0.4% 0.6% 1.4% 0.6% Transfer Abroad 4.8% 1.6% 2.7% 1.1% Other 5.4% 3.5% 7.8% 8.9% 1.43 Similarly, average total incomes, which include social transfers and income from self employment, did not grow much between 2004 and 2006 (Table 18). Prima facie, this finding is somewhat at odds with those presented earlier in Section 4 based on expenditure data, which suggested considerable improvement in average living standards during this period. The importance of regular salaries (in the formal sector) increased especially for richer households, indicating that reported earnings from entrepreneurial activities are largely informal sector ear nings that poorer households are forced into when they cannot obtain formal sector jobs paying regular salaries. The poorest quintile earns 52% of labor income (income from salaries). This ratio is 60% in relation to the total income. Table 18 Percentage Gr owth Rates in Income and Expenditures, Labor Income Labor + Social Transfers Labor + Social transfers + Agriculture + Self Employed Total Expenditures Per Capita Total Low 25% High 25% Summary and Concluding Observations 1.44 The poverty profile for Serbia presented in this note has shown relatively large regional differences in poverty rates and a strong link of poverty with employment and education. Despite significant economic growth, there is evidence of locational poverty traps, involving significant concentration of the poor in some Serbian regions which has benefited from growth very little, such as rural parts of Vojvodina and rural parts of Central Serbia. Multivariate analysis of poverty shows smaller

18 regional disparities in consumption levels than those actually observed. This suggests that poor areas are populated predominantly by those with low education, high dependency and other characteristics making them prone to poverty. Among other things this will suggests that people from poor areas face unclear prospects if they decide to migrate to richer growing centers. Thus, a regional emphasis on poverty reduction in the NSPR, to supplement sustained pro-poor growth, is warranted Employment status is strongly correlated with poverty. Individuals living in households whose heads are unemployed face the highest poverty risk, however, majority among the poor are from families where someone works. However, the multivariate poverty profile shows that having a household business significantly increases consumption, especially if the business is registered. This suggests that non- economies of agricultural self-employment is high-rewarded strategy as in many advanced transition Central Europe. However, more in-depth analysis of the labor market and poverty linkages will be possible when the new LSMS data for 2007 will become available The strong correlate of poverty is education. Individuals living in households with highly educated heads (post secondary education and university) had poverty incidence close to zero (1.2 and 1.8 percent respectively) and they made only 2.6 percent of all poor. This indicates that education pays off since the labor market rewards education as considerable wage advantage for highly educated workers (collage or more) is observed. As Serbian education system is relatively restrictive in terms of allowing expansion of there higher tiers, it is the poor who are likely to pay the price of these elitist traditions, and their children are the one who are most likely to be channeled to the lower vocation streams offering rather bleak labor market prospects On comparing data from the 2004 and 2006 HBS to ascertain the impact of recent economic growth on living conditions in Serbia, our findings are somewhat mixed (constructing comparable welfare aggregates has been a challenging undertaking, given various improvements in HBS methodology introduced over the years). On the one hand, poverty estimates based on expenditures per equivalent adult based on the 2004 and 2006 HBS show considerable improvements in living conditions during this period, a finding that is quite consistent with other available macroeconomic data for Serbia covering the same period. On the other hand, income data from the same HBS series show virtually no change in average incomes over this period (even if average labor incomes have increased considerably). Therefore a more definitive assessment of changes in living conditions in Serbia in recent will have to await the availability of new data from the 2007 LSMS and 2007 HBS, which can then be compared with fully comparable welfare aggregates based on data from earlier rounds of these respective surveys.