Research Paper No. 2008/26 Asset-Based Poverty in Rural Tajikistan Who Climbs out and Who Falls in? Oleksiy Ivaschenko and Cem Mete* March 2008 Abstract Tajikistan’s rural sector has witnessed substantial development since the country began to emerge from civil conflict in 1999. Gross agricultural output increased 64 per cent from 1999 to 2003, and there were significant developments in the agricultural reform agenda. This paper uses the panel component of two surveys conducted in Tajikistan at one-year interval (2003 and 2004) to explore the major determinants of the transition out of/into poverty of rural households. Poverty status is measured in the asset space, thus indicating structural rather than transitory poverty movements. The empirical analysis reveals several interesting findings that are also important from a policy perspective: first, cotton farming seems to have no positive impact on poverty levels, nor on mobility out of poverty. Second, the rate of increase in the share of private farming at the district level had little impact on poverty levels and poverty mobility. …/. Keywords: welfare, poverty, Tajikistan JEL classification: I3 Copyright © UNU-WIDER 2008 * ECSHD, World Bank, Washington, DC: emails: oivaschenko@worldbank.org (O. Ivaschenko); cmete@worldbank.org (C. Mete) This is a revised version of a paper originally prepared for the UNU-WIDER conference on Fragile States—Fragile Groups, directed by Mark McGillivray and Wim Naudé. The conference was jointly organized by UNU-WIDER and UN-DESA, with a financial contribution from the Finnish Ministry for Foreign Affairs. UNU-WIDER gratefully acknowledges the contributions to its project on Fragility and Development from the Australian Agency for International Development (AusAID), the Finnish Ministry for Foreign Affairs, and the UK Department for International Development—DFID. Programme contributions are also received from the governments of Denmark (Royal Ministry of Foreign Affairs), Norway (Royal Ministry of Foreign Affairs) and Sweden (Swedish International Development Cooperation Agency— Sida). ISSN 1810-2611 ISBN 978-92-9230-072-2 Third, there is strong evidence of geographic poverty mobility traps in Tajikistan. Higher levels of poverty in a district appear to reduce significantly the chance of a household shedding poverty. Living in a region with overall slow economic growth is also found to undermine the odds of exiting poverty and to increase the risk of falling into poverty. Finally, several key household-level factors, such as the share of adults, education level, health status and participation in wage employment, also emerge as significant predictors of poverty mobility. Authors notes The findings, interpretations, and conclusions expressed herein are those of the author(s) and do not necessarily reflect the views of the International Bank for Reconstruction and Development/the World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. The World Institute for Development Economics Research (WIDER) was established by the United Nations University (UNU) as its first research and training centre and started work in Helsinki, Finland in 1985. The Institute undertakes applied research and policy analysis on structural changes affecting the developing and transitional economies, provides a forum for the advocacy of policies leading to robust, equitable and environmentally sustainable growth, and promotes capacity strengthening and training in the field of economic and social policy making. Work is carried out by staff researchers and visiting scholars in Helsinki and through networks of collaborating scholars and institutions around the world. www.wider.unu.edu publications@wider.unu.edu UNU World Institute for Development Economics Research (UNU-WIDER) Katajanokanlaituri 6 B, 00160 Helsinki, Finland Typescript prepared by Liisa Roponen at UNU-WIDER The views expressed in this publication are those of the author(s). Publication does not imply endorsement by the Institute or the United Nations University, nor by the programme/project sponsors, of any of the views expressed. 1 Introduction This paper exploits an asset-based approach to study the (asset-based) poverty dynamics of Tajikistan rural households. We use a panel of rural households that have been observed during two time periods: June-July 2003 and July-November 2004. Analysing the dynamics of rural poverty in Tajikistan during this time period is particularly interesting in view of the drastic changes that have occurred in the country over the last several years. Emerging in 1999 from civil war and a prolonged period of economic collapse,1 the country’s economic performance has been impressive from the year 2000, with sustained real GDP annual growth rates of 7 to 9 per cent.2 Economic growth has been accompanied with substantial reduction in poverty, dropping from 81 per cent of the population living below the poverty line (US$2.15 per day) in 1999 to 64 per cent in 2003 (World Bank 2006). Although poverty headcount fell during this period by 19 percentage points in the rural areas compared to 14 percentage points in urban centres, it remains higher in the rural regions: 65 per cent versus 59 per cent. As 73 per cent of the population live in the countryside, poverty in Tajikistan continues to be an overwhelmingly rural phenomenon. Economic growth and the resultant poverty reduction are explained by three major factors: (i) conflict cessation, which allowed economic activity to resume and markets to develop; (ii) initial impact of the macroeconomic stability and agricultural reforms in the non-cotton sector that enabled farmers to diversify production and increase productivity; and (iii) large increase in migrant workers exiting Tajikistan for Russia and other countries. However, there have been concerns that once the initial benefits of these ‘special’ factors dry out, Tajikistan’s poverty reduction trends may not be sustainable (World Bank 2006). In view of the sound economic growth rates, markedly reduced but still very high rural poverty, and concerns over the sustainability of the country’s poverty reduction trend, it is important from a policy perspective to understand the key factors at the micro (household/community) level that explain the transition of rural households in and out of poverty. This is the main objective of this paper. The paper contributes to the literature on welfare dynamics in general, and to the studies of poverty in Tajikistan in particular on several fronts: i) utilizing an assets-based approach to better capture the permanent (as opposed to transitory) component of welfare changes for rural households; ii) investigating explicitly the importance of community/local factors versus household/individual level characteristics to explain movements in and out of (asset-based) poverty. It is worth noting that a study of the general factors affecting poverty transition in Tajikistan has been undertaken by Angel-Urdinola, Mete and Cnobloch (2008). Our 1 Looking at cotton output across rayons (the smallest administrative unit), we observe that between 1991 and 1999, there was an average output decline of 62 per cent. Cotton output has increased since 1999 by an average of 91 per cent, but still remains at about 66 per cent of its 1991 level. 2 Despite solid growth rates, Tajikistan’s per capita GDP in 2004 was still merely US$225, making it the poorest country in the Europe and Central Asia region. 1 study expands on this work by focusing specifically at the determinants of welfare dynamics in rural areas. More specifically, we combine household survey data with district (rayon) level data to take into account the poverty impact of community-level factors such as the share of private (dekhan) farms, per hectare of land under cotton cultivation, level of debt and distance to market or district centre.3 The paper is structured as follows. Section 2 discusses the main developments in Tajikistan’s rural sector since 1999, as well as the correlates of rural poverty at the district (rayon) level. Section 3 provides the theoretical and empirical framework for the analysis of the (asset-based) poverty mobility at the household level. It also describes the data and the constructed asset index. Section 4 presents the empirical results, and section 5 concludes with a summary of the main findings and their policy implications. 2 Tajikistan’s rural sector developments since 1999, and correlates of rural poverty at the district (rayon) level 2.1 Rural sector developments since 1999 Tajikistan’s rural sector has witnessed substantial changes since the country emerged from civil conflict in 1999. These include agricultural reform, and specifically the rapidly changing structure of land ownership; significant output growth due to increased yields; and unfavourable developments in the price of cotton, the dominant cash crop of Tajikistan. In terms of the ownership structure, the country’s agricultural sector had been fairly unreformed until the late-1990s, but experienced considerable transformation thereafter. In 2000, the agricultural sector was still dominated by the old state-farms inherited from the soviet system.4 These farms accounted for more than 60 per cent of the arable land (Figure 1) but contributed only about 30 per cent of the total agricultural sector output because of low efficiency (Figure 2). The ownership structure changed radically during 1999-2004, as old state-farms were dismantled and private ownership dekhan farms were created.5 As a result, the share of land cultivated by the state-farms declined to approximately 30 per cent, while that cultivated by the newly created dekhans increased to almost 50 per cent (Figure 1). A fourfold increase in the output of the dekhans during 1999-2003 raised their contribution to sector output from 10 to 24 per cent. The share of 3 Angel-Urdinola, Mete and Cnobloch (2008) differentiate between urban and rural areas by including an urban/rural dummy in their regression that uses combined (urban and rural) panel sample, and thus they make no attempt to analyse welfare dynamics determinants that would be specific to the rural areas. 4 State-farms encompass ownership forms of both the sovhoz (soviet farms) and kolhoz (collective farms), which are effectively the same. 5 A thorough overview of agricultural reforms in Tajikistan is provided in World Bank (2005). The major findings of this study are: (i) the process of land restructuring has been rather inequitable; (ii) the reform of state-farms in especially cotton-producing areas has resulted in numerous distortions; many state cotton farms were dismantled into a number of smaller units, each with a farm manager and 150-200 workers, with workers having little decisionmaking power and being paid mostly in kind; (iii) cotton production under current conditions is generally not profitable to the farmers. 2 land cultivated under household plots remained stable, at about 20 per cent, but contributing about half of the agricultural output (Figure 2). After a period of prolonged decline, Tajikistan’s agriculture sector enjoyed noticeable growth after 1999: over the period 1999-2003, gross output increased by 64 per cent, with most of this expansion occurring during 2001-03. The crop sector, which accounts for 81 per cent of output, grew by 65 per cent during 1999-2003, while livestock, Figure 1 Change in land ownership structure: 2000 to 2004 60 40 20 0 2000 2004 dekhan farms household plots state farms Source: Authors’ estimates. Figure 2 Composition of gross agriculture output 60 50 40 30 20 10 0 1996 1997 1998 1999 2000 2001 2002 2003 Other agricultuYrael aenrsterprises Dekhkan farms Household plots Source: Ukaeva (2005). 3 Figure 3 Trends in the cotton output, yield and value: 1991 to 2003 2.0 1.5 1.0 0.5 0 1991 1999 2001 2003 Yield, tons/ha Output, mil. tons Value of output, mil. $ (x10) Source: Computed by the authors using official data on cotton production. accounting for 19 per cent of sector output, expanded by 61 per cent.6 The agricultural production developments of this period can be well illustrated with data on cotton production. Cotton traditionally has been a major agriculture commodity, and continues to account for about two-thirds of total crop output value (Ukaeva 2005). The cotton sector has experienced substantial output fluctuations during periods of civil conflict and economic transition. Between 1991-99, cotton output declined 62 per cent, from 820,000 tons to 313,000 tons, but increased 73 per cent between 1999 and 2003, still accounting for only 65 per cent of the 1991 level. The increase in output is mostly a reflection of improved yields (1.1 tons per hectare in 1999 to 1.8 tons per hectare in 2003, Figure 3) as well as an increase in cultivation area. The cotton sector in Tajikistan has been severely hit by declining global prices. Despite output increasing by more than two-thirds between 1999 and 2003, the declining global prices reduced the real value of output by 7 per cent during the same period (Figure 3).7 Adverse developments in international cotton prices, coupled with the farmers’ ill- 6 See Ukaeva (2005) for a detailed discussion based on the decomposition of agricultural growth between 1999 and 2003. 7 Due to civil conflict and low cotton production, Tajikistan missed the opportunity to benefit from the historically high cotton prices in the mid-1990s. International cotton prices declined from about UD$0.90 per pound in 1995/96 (the highest level over the last 30 years) to about US$0.45 pound in 2003 (one of the lowest levels over the last 30 years). Since then, prices have bounced back somewhat to about US$0.60 per pound. International prices are projected to remain at about the same level for the next few years, or at least that they are not likely to go up due to such factors as new technologies (genetically-modified cotton), more extensive use of existing technologies, new areas allocated to cotton production (i.e., increased role of China in the global production of cotton), and government policies (such as direct subsides to cotton farmers). For a more elaborate discussion of the issues, see Becerra (2004). 4 advised pricing arrangements with investors, have resulted in dubious ‘debts’ that the producers are struggling to repay.8 These developments are expected to have an impact on the standards of living of Tajikistan’s rural population. 2.2 The correlates of rural poverty at the rayon level How are the developments of the rural sector, as outlined above, correlated with poverty? To gain some insights into this issue, we look at the correlation of selected key variables of the rural section at the rayon level with the poverty headcount at a similar level, as obtained from the poverty mapping conducted in Tajikistan (Baschieri and Falkingham 2005). Some of the variables will be used later to explain poverty mobility at the household level. A number of interesting findings emerge from simple scatter plots of the district-level data (Figures 4 and 5). A U-curve relationship exists between the share of land under cultivation in mountainous terrain and poverty headcount, while the share of pastoral land is not correlated with poverty. Overall, about 60 per cent of Tajikistan is covered by mountainous terrain, with significant differences across the rayons. The data suggest that both types of territories, whether encompassing an insignificant or a significant percentage of mountainous terrain, are likely to be very poor, with a poverty headcount of about 80 per cent (Panel A, Figure 4). The share of pastoral land does not seem to be a factor (Panel C, Figure 4). A higher share of irrigated farming land is associated with somewhat lower levels of poverty. However, the level of irrigation is a very weak correlate of poverty. The data indicate that even well-irrigated areas are likely to have huge variations in the level of poverty, ranging from a high of 80 per cent to a low of 40 per cent (Panel B, Figure 4). A larger portion of land under dekhan cultivation is correlated with lower poverty levels. However, the increase in dekhan farming land between 2000-04 shows no correlation with poverty. The level of dekhan farming in 2000 (prior to its substantial increase in the structure of land ownership) seems to be negatively correlated with poverty headcount (Panel D, Figure 4). However, additional dekhan cultivation has not been reflected in poverty levels (Panel E, Figure 4). This finding is consistent with other evidence which suggests that the increasing ratio of dekhan cultivation has not been accompanied by improved productivity on these farms (World Bank 2005). Moreover, many of these dekhan farms are in cotton production, and have thus been affected by the sector’s adverse development. 8 Budgetary pressures in 1997 led the government to sign a partnership with the Swiss cotton trading company, P. Reinhart, which, based on cotton deliveries backed by a government guarantee, was to provide needed financing. In 1998, the government guarantee was replaced with a ‘commercial’ financing scheme whereby Reinhart worked with a number of local agents (referred to as financiers, futurists or investors). This framework became the basis of cotton production and marketing. Unwise pricing arrangements squeezed the profit margins of the cotton farmers, putting many in a debt trap. These debts, currently estimated at US$280 million, have paralysed the cotton sector, as indebted farmers are unable to obtain credit elsewhere. Indebted farmers are also reluctant to privatize and invest in their land, adding a further impediment to the growth of agriculture sector. For further discussion, see World Bank (2005). 5 Figure 4 Correlations between rayon (district) poverty headcount* and land variables 100 90 80 70 60 50 40 30 Panel (A): Poverty vs. mountainous terrain 100 90 80 70 60 50 40 30 Panel (B): Poverty vs. % of farming land which is irrigated 20 10 0 10 20 30 40 50 60 70 80 90 100 % of mountainous terrain 20 30 40 50 60 70 80 90 100 % of farming land which is irrigated (2003) 100 Panel (C): Poverty vs. % of pastoral land 100 Panel (D): Poverty vs. % of farming under dekhan farms 90 90 80 80 70 70 60 60 50 50 40 40 30 20 30 40 50 60 70 80 90 100 % of pastoral land (2003) 30 0 10 20 30 40 50 60 % of farming under dekhan (private) farms (2000) 100 Panel (E): Poverty vs. change in area under dekhan farms 100 Panel (F): Poverty vs. % of arable land under household plots 90 90 80 80 70 70 60 60 50 50 40 40 30 0 10 20 30 40 50 60 Change in land area under dekhan farms, % point (00 to 04) 30 10 20 30 40 50 60 70 80 % of arable land under household plots (2000) Note: * Per cent of the population living on less than US$2.15 per day. Source: Authors’ calculations based on TLLS data. 6 Figure 5 Correlations between rayon (district) poverty headcount* and cotton production 100 Panel (A): Poverty vs. % of arable land under cotton 100 Panel (B): Poverty vs. mean wage on cotton farms 90 90 80 80 70 70 60 60 50 50 40 40 30 0 10 20 30 40 50 60 70 80 % of arable land under cotton, 2003 30 10 15 20 25 30 35 40 45 50 55 mean monthly wage on cotton farms, Somoni (2003) 100 Panel (C): Poverty vs. cotton debt (per capita) 100 Panel (D): Poverty vs. cotton output (2003 as % of 1991) 90 90 80 80 70 70 60 60 50 50 40 40 30 0 50 100 150 200 250 300 cotton debt (per capita), $ 30 30 40 50 60 70 80 90 100 cotton output, tones (2003 as % of 1991) 100 Panel (E): Poverty vs. % change in cotton output (1999 to 2003) 100 Panel (F): Poverty vs. % change in value of cotton (1999 t 90 90 80 80 70 70 60 60 50 50 40 40 30 0 50 100 150 200 250 % change in cotton output (1999 to 2003) 30 -40 -20 0 20 40 60 80 100 % change in $ value of cotton output (1999 to 2003) Note: * Per cent of the population living on less than US$2.15 per day. Source: Authors’ calculations based on TLLS data. 7 A larger share of arable land under household plots is weakly associated with lower poverty levels. The share of this type of land has remained quite stable over time at about 20 per cent. It would appear that districts with a somewhat higher than average share of land under household plots exhibit lower levels of poverty (Panel F, Figure 4). The importance of this type of farming to overall agricultural production is, however, unlikely to change. The districts with more extensive cotton cultivation are likely to have higher poverty levels (Panel A, Figure 5). As can be expected, there is a negative correlation between a district’s average cotton-farm wages and poverty levels (Panel B, Figure 5). Although, this relationship is at least partly driven by the fact that in poorer districts, cotton-farm workers are paid lower wages. We also find that the average wage arrears per person are higher in the poorer areas than elsewhere.9 These findings suggest that in poorer areas cotton-farm labourers receive smaller wages, and they are less likely to be paid on time. We find a very weak correlation between the level of the cotton-farm debt (per capita) and poverty at the district level. The extent of decline in output during the 1990s is strongly (positively) associated with poverty. Based on cotton output data, we find that districts with greater output gaps between 1991 (the peak output year before economic collapse) and 2003 are much more likely to be poorer. The deviations in cotton output and its value between 1999-2003 are not correlated with the levels of poverty at the district level. The above examination of the key correlates of poverty at the district level provides a solid basis for analysing the (asset-based) poverty mobility at the household level in the next section. 3 Theoretical and empirical framework for an asset-based analysis of poverty mobility 3.1 Using an asset-based poverty line to identify poverty transitions The literature on poverty dynamics has increasingly recognized the importance of adopting an asset-based approach to study changes in wellbeing, especially in response to a wide range of different (climatic, health, political and other) shocks.10 Differentiating between stochastic and structural poverty transitions implies the availability of information on assets and expected levels of wellbeing. To illustrate the importance of the assets-based approach in capturing welfare-status changes, we use the conceptual framework advocated by Carter and May (2001) and Carter and Barrett (2006). This framework is presented in Figure 6. In essence, in any timeperiod a household can be regarded as structurally poor if household consumption falls below the consumption poverty line u* and its stock of 9 The graph is not presented here, but available from the authors on request. 10 See February 2006 special issue of the Journal of Development Studies, 42 (2) for the set of papers presenting the conceptual framework and empirical evidence for an asset-based approach. 8 Figure 6 Poverty transitions: consumption poverty line vs. asset poverty line Utility Asset poverty line Consumption poverty line Assets Source: Carter and Barrett (2006). assets falls below the asset poverty line A.11 Such a state is described by point B in Figure 6. A household can be regarded as stochastically poor if it holds assets above level A, yet its level of consumption is below the poverty line u* (described by point E above). A household that has over time moved from below to above the consumption poverty line u* could be regarded as having made a stochastic transition out of poverty if its assets are still mapped below the asset poverty line A. This case is represented in by the shift from point B to point C, which may occur because of increased crops in a given year due to favourable weather conditions. As a result, the livelihood function shifted upwards from u (A) to u’ (A), reflecting the increased returns on existing assets. The stochastic transition into poverty is represented here by the shift from point D to point E, whereby household consumption drops below the poverty line, as returns to existing assets temporarily diminish, but the level of asset holdings stays above the asset poverty line. This is exemplified by the household that has experienced a temporary consumption decline because of a negative shock (e.g., drought), but is expected to bounce back to a level of consumption above the poverty line. A household that has shifted over time across the consumption poverty line u* could be regarded as having made a structural transition out of poverty, if it has also accumulated sufficient additional assets to move above the asset poverty line A (represented here by the shift from point B to point D). Conversely, the shift from point D to point B would represent the structural transition into poverty. As Figure 6 indicates, there could be multiple options for poverty transition paths, but the main point here is that changes in 11 The asset poverty line in Figure 1 is simply the level of assets corresponding to the level of wellbeing equal to the consumption poverty line. 9 consumption that are not accompanied by changes in the assets base can be regarded as stochastic rather than structural transitions. 3.2 Data This paper uses panel components from two surveys: the 2003 Tajikistan Living Standard Survey (TLLS) and the 2004 Energy Household Survey (EHS). The 2003 TLLS provides a nationally representative sample of households stratified by oblast and rural/urban settlements based on a selection of households recorded in the 2000 census. The survey was conducted during June-July 2003. The sample size is 4,156 households representing 26,141 individuals. The 2004 EHS survey was conducted between July and November of 2004. The sample, also representative of the overall population, includes 2,600 households and 15,339 individuals. The panel component consists of 1,396 households representing 8,368 individuals; 589 of the households are rural.12 The 2004 HES used the same sample frame (list of clusters) as the 2003 TLLS. A comparison of the distribution of the basic variables from the panel sample against the 2003 cross- section indicates that the panel sample is fairly representative of the overall population, both at rural and urban levels. Both surveys collected information on such household attributes as demographics, education and health, income and expenditures, assets, and consumption. The analysis of poverty dynamics here uses the panel component of the two surveys, and is based on an asset index. Construction of the asset index is described below. In addition to utilizing household-level data, the empirical analysis at the micro (household) level exploits a few key district-level variables to capture agricultural reform and various policy changes (discussed earlier). The analysis also uses community survey data from the 2003 TLLS, to allow us to identify whether cotton was grown in a particular community, whether rainfall during the survey year was better/worse relative to the previous year, as well as certain other important community-level characteristics that are likely to be associated with a household’s mobility out of or into poverty. 3.3 Constructing the asset index and asset-based poverty line In order to construct an asset index, we rely on principal component analysis (Lawley and Maxwell 1971).13 The principal component constitutes a linear index capturing most of the information (variance) common to all the variables. Denoted by Aij the observation for household i and asset j (for example, whether or not a household has a 12 The majority of households in the panel are urban because the 2004 EHS over-sampled these areas. 13 For a comprehensive discussion of the pros and cons of an assets-based welfare analysis, see Filmer and Pritchett (1998), Sahn and Stifel (2000), and Sahn and Stifel (2003). An asset index retains certain properties necessary for proper welfare analysis, such as transparency in construction and ranking individuals credibly in terms of welfare. As argued by Filmer and Pritchett (1998), Carter and May (2001) and Carter and Barrett (2006), an asset index is likely to be a better indicator of the long-run household wealth than per capita household consumption. However, a significant limitation of an assets index is that it treats ownership of assets as giving similar utility without allowing for differences in unobserved quality. 10 television). Principal component analysis finds a small number of n factors, denoted by the letter f, which can be used to reconstruct the original variables (in this case the original information on assets) as linear functions of the q factors, so that : Aij = fi1 1j + fi2 2j + ... + fiq qj + ij (1) In (1), Aij is known since it is one of the values describing whether or not household i has asset j. The term fik represents the observation for household i of the value of factor k which needs to be estimated. The term kj is the coefficient indicating the dependence of the observed asset variable j upon the factor k, this coefficient being also estimated. The residual, ij, is the error term. In other words, factor analysis produces an index representing (through the vector of common factors F) the data generating process underlying the actual observations Aij. This is done by finding the one dimension of the space in which the original observations are represented with the largest variance, from j = 1, ..., p to k = 1, ..., n with n