Riyana Miranti a a University of Canberra, Published online: 17 Mar 2010.

Transcription

1 This article was downloaded by: [Macquarie University] On: 14 July 2013, At: 22:05 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: Registered office: Mortimer House, Mortimer Street, London W1T 3JH, UK Bulletin of Indonesian Economic Studies Publication details, including instructions for authors and subscription information: Poverty in Indonesia : the impact of growth and changes in inequality Riyana Miranti a a University of Canberra, Published online: 17 Mar To cite this article: Riyana Miranti (2010) Poverty in Indonesia : the impact of growth and changes in inequality, Bulletin of Indonesian Economic Studies, 46:1, 79-97, DOI: / To link to this article: PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the Content ) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at

2 Bulletin of Indonesian Economic Studies, Vol. 46, No. 1, 2010: POVERTY IN INDONESIA : THE IMPACT OF GROWTH AND CHANGES IN INEQUALITY Riyana Miranti * University of Canberra This paper examines the growth elasticity of poverty across three development episodes in Indonesia between 1984 and 2002, after controlling for inequality. It relies on estimation of panel data from the National Socio-Economic Survey conducted by the central statistics agency. Contrary to expectations, the growth elasticity of poverty was virtually indistinguishable across the three development episodes a period of far-reaching policy liberalisation ( ); a second period of slower liberalisation ( ); and the period of recovery from the Asian financial crisis ( ). Growth was pro-poor in all three periods, while the impact of growth on poverty was either augmented or offset by changes in inequality, depending on the period. Only during the first liberalisation period did a reduction in inequality serve to augment the impact of growth on poverty. INTRODUCTION The incidence of poverty in Indonesia fell from 29.5% in 1984 to 18.2% in 2002, although the decline was interrupted for a few years by the financial crisis of the late 1990s. The first objective of this paper is to examine the impact of economic growth on poverty, with a special focus on differences during periods of policy liberalisation and crisis recovery. It does not attempt, however, to reach definitive conclusions about why the impact differed across time. Most previous studies measuring the impact of economic growth on poverty, including those on Indonesia, have used a short time-frame and have not examined differences in impact across different time periods (for example, Bidani and Ravallion 1993; Balisacan, Pernia and Asra 2003; Miranti and Resosudarmo 2005). This may be due to data limitations, as a longer time series is necessary for an extended analysis. Friedman (2005) did extend his focus to a 15-year period ( ), but did not decompose this longer time-frame into shorter development episodes. The analysis in this paper is undertaken for the years , and the period is divided into three development episodes, , and The purpose of considering these three sub-periods separately is to shed some empirical light on the view that differing government policies across these sub-periods had different impacts on poverty. * riyana.miranti@natsem.canberra.edu.au. I am grateful to Chris Manning, Hal Hill, Budy Resosudarmo, Ross McLeod and two anonymous referees for their advice on previous drafts. Those who gave advice bear no responsibility for any errors or deficiencies in the final version. ISSN print/issn online/10/ DOI: / Indonesia Project ANU

3 80 Riyana Miranti If the distribution of income is constant, economic growth must reduce poverty. If income distribution becomes more equal, this tends to reduce poverty, and conversely. The second objective of this paper is therefore to determine the extent to which poverty reduction was simply the consequence of growth, and to what extent this effect was augmented or offset by changes in inequality. DEFINITIONS AND DATA Poverty To calculate the growth elasticity of poverty we focus on the head-count ratio or incidence of poverty that is, the proportion of poor people in the total population. People are categorised as poor when their consumption is below a certain threshold, referred to as the poverty line. Henceforth, poverty as used in this paper refers to poverty as measured by the poverty head-count ratio. This paper applies the poverty line methodology currently used by the central statistics agency (BPS), the official institution publishing poverty data for Indonesia (BPS 2003). BPS adopts the basic needs approach, defining the poverty line as the cost of consuming 2,100 calories per person per day, plus a pro-rata allowance for non-food requirements. This approach differs from the analysis of Friedman (2005), who applied the methodology of Bidani and Ravallion (1993) and Ravallion (1994) in constructing his poverty lines, and calculated the cost of non-food needs in terms of the cost of foregone food items. BPS claims that its current methodology is more comparable across regions and consistent across time than previous methods. It has used this method since it revised the 1996 official poverty figure. Three significant methodological improvements have been introduced (BPS 2003). First, the measure includes more commodities as basic needs. Second, consistency across time has been improved by applying the 1999 shares of total consumption (proxied by expenditure) for each of 52 products. The expenditure share of each product in the bundle differs across provinces but is similar over time. Third, comparability across provinces has been improved by indexing the price level in each province to the price level in Jakarta, such that the cost of the basic needs basket of goods is comparable for the reference populations in all provinces. Thus, regional disparities in consumption arise only because of differences in consumption patterns and prices, and not because of differences in income levels. 1 The methodology is considered dynamic in the sense that different consumption tastes and patterns among provinces are taken into account. As a preliminary step, this paper uses the improved BPS methodology to recalculate poverty incidence between 1984 and 1996, the year the official poverty figure was revised. 2 Applying the new methodology results in poverty levels for the period that are different from those officially published by BPS. Table 1 shows the official poverty series and the re-calculated poverty levels used in this paper. 1 However, the BPS poverty line has some shortcomings, especially in relation to regional price differences, which are calculated and discussed in Nashihin (2007). 2 For a full description of the process, see Miranti (2007).

4 Poverty in Indonesia : the impact of growth and changes in inequality 81 TABLE 1 Poverty, (% of population) Official BPS Poverty Series New Poverty Series a revised (see text) a Recalculated by the author for 1984, 1987, 1990 and 1993, using the 2003 poverty line methodology of the central statistics agency (BPS). Source: BPS, National Socio-Economic Survey (Susenas), various years. The poverty level is calculated using a panel of seven consecutive crosssectional National Socio-Economic Surveys (Survei Sosial Ekonomi Nasional, or Susenas) undertaken by BPS from 1984 to 2002, covering 26 provinces. 3 While the core Susenas survey is conducted annually, a Susenas module covering consumption is published every three years. Since 1981, the consumption module sample has covered approximately 65,000 households. Quantity and value data are collected on more than 300 items for representative households in each province. The present study covers the period , using the 1984, 1987, 1990, 1993, 1996, 1999 and 2002 consumption modules (that is, seven waves). Use of provincial-level data provides far more observations of growth and inequality than would relying on national data, and hence results in more reliable econometric estimates. Growth The term economic growth is most commonly understood to refer to a change in the level of gross domestic product (GDP), a measure reported in the national accounts, and its regional counterpart, gross regional domestic product (GRDP). However, changes in the welfare of a country s people may also be represented by other measures. Aggregate household income, estimated from household surveys, is one of these. Deaton (2001: 127) has criticised attempts to use growth data from the national accounts to study the relationship between growth and poverty. He argues that this growth measure has at best a weak relationship with poverty, since the national accounts data and the household surveys used for calculating poverty 3 For consistency, four new provinces Bangka Belitung, Banten, Gorontalo and North Maluku which first appeared in the 2002 Susenas as a result of province fragmentation have been re-combined with the provinces they were separated from, in order to match the former provinces of South Sumatra, West Java, North Sulawesi and Maluku.

5 82 Riyana Miranti TABLE 2 Growth Rates by Growth Measure and Development Episode, (% p.a) First Liberalisation Period Second Liberalisation Period Crisis Period Recovery Period National accounts GDP GDP per capita Manufacturing GDP Household surveys Mean consumption expenditure per capita Source: BPS, Susenas and national accounts, various years; CEIC Asia Database, various years. Data on mean consumption expenditure per capita for the crisis period are from Suryahadi, Sumarto and Pritchett (2003). levels measure different things. There are marked discrepancies in both direction and rate of change between the two types of growth data (table 2). It can therefore be argued that using an income proxy from the national accounts in conjunction with household survey poverty data introduces new errors, in addition to those found in the national accounts (Ravallion and Chen 1997: 363). The two types of errors do not cancel each other out (Ravallion 2003). We would expect the same problem to occur if an income proxy from the regional accounts were used. Adams (2004) found that the growth elasticity of poverty is sensitive to the measure of income used. Most traditional empirical estimations of the growth elasticity of poverty use mean consumption expenditure from household surveys as a proxy for income (Adams 2004). It has been argued that this proxy is a better reflection of welfare than is income calculated from the same source (Ravallion 1995), because it gives a more accurate indication of life-cycle or permanent (that is, long-term) income (Balisacan, Pernia and Asra 2003: 332). Moreover, the data collected on consumption are more accurate than the income data, given that people may have reasons to hide some of their income (Ravallion 2001) and that measured consumption patterns are less variable than measured income patterns (Deaton 1997). This study uses mean consumption from the household surveys as a proxy for income, and uses change in this variable as a proxy for economic growth. It uses the province as the spatial unit for calculating income and changes in income, and calculates monthly mean household consumption expenditure per capita for each province in each Susenas consumption module period. Thus, growth is defined in this study as the percentage change in provincial monthly mean household consumption expenditure per capita (for simplicity, we refer to this below as consumption expenditure per capita ). The term growth elasticity of poverty refers to the percentage reduction in poverty given a 1% increase in mean consumption expenditure per capita (rather than given a 1% change in GDP or GRDP). In this paper, mean consumption expenditure per capita is calculated in real terms, to allow comparison across time. Real mean consumption expenditure per

6 Poverty in Indonesia : the impact of growth and changes in inequality 83 capita is calculated by deflating mean consumption to 1984 prices. The deflator used is the implicit deflator of the weighted average of the urban and rural poverty lines, following Friedman (2001), Tarp et al. (2002) and Grimm and Gunther (2006). The poverty line has been chosen as the deflator in preference to the consumer price index (CPI), because it gives a better representation of the spending patterns of the poor (Adams 2004). Food items comprise only 40% of the CPI, but they represent more than 70% of the items used to calculate the poverty line (BPS 2003). A further consideration is that the CPI is calculated only for 44 urban areas in Indonesia, and does not cover rural areas, even though price movements in urban and rural areas may differ. Use of the CPI as the deflator could therefore result in biased estimates of real consumption. DEVELOPMENT EPISODES Between the 1970s and 2002, Indonesia experienced several distinct episodes of development, including the oil boom from 1972 to around 1981; the rice boom of ; a period of wide-ranging liberalisation throughout the mid- to late 1980s (Woo, Glassburner and Nasution 1994; Hill 2000); a second period of more cautious liberalisation (Hill 1997) and some back-sliding on reform during the first half of the 1990s; the financial and economic crisis of ; and, finally, recovery after the crisis, beginning in There is no clear consensus on when either of the two liberalisation periods started. One strand of the literature refers mainly to microeconomic reform especially trade liberalisation commencing in 1986 (Woo, Glassburner and Nasution 1994: 115) or 1987 (Hill 2000: 17), following a sharp decline in world oil prices in Several liberalisation packages were introduced in 1986 to ease import and export procedures. They included a duty exemption and drawback scheme that allowed export-oriented firms to purchase imported inputs at international prices the initial step towards an export-promoting path of industrialisation. But in addition to trade liberalisation there had been some earlier fiscal reforms (such as tax reforms in 1983 and 1985) and exchange rate reforms (a devaluation in 1983). Soesastro (2006) discusses these, and identifies liberalisation as commencing in Aswicahyono and Feridhanusetyawan (2004: 13) too have argued that microeconomic reform began in the period , although they note that this reform was slower and less effective than the liberalisation of the mid-1980s. 4 The delineation of periods for analysis in this paper is therefore to some extent arbitrary, and has been driven partly by the availability of data at three-year intervals from the Susenas consumption module. The timing of the latter made 1984 a convenient starting point for the analysis, and and appropriate time-spans for the first and second liberalisation periods, while realistically reflects Indonesia s crisis recovery period. Data for 1990, the transitional year linking the first and the second liberalisation periods, is included in both. The following section discusses each of the development episodes. In explaining economic growth during each of the development episodes, two sources of 4 While trade liberalisation would have had a predictable impact on poverty, the impact of other kinds of reform on poverty is less clear.

7 84 Riyana Miranti growth data are discussed: the national accounts and the household surveys. Growth data from the latter source are then used in the estimation below. The first liberalisation period ( ) During the first liberalisation period, GDP grew by 6.3% annually. Manufacturing GDP expanded by 10% p.a. (table 2) almost twice the rate of GDP in agriculture (5.1%; figure not shown). Per capita GDP growth was 4.1%, while per capita expenditure consumption based on household survey data grew more slowly, at 1.1%. The difference in the rate of change between the national accounts (GDP) data and the household survey data is the result of differences in the definition and coverage of GDP and household consumption per capita. This first liberalisation period can be called one of transformation, in which Indonesia experienced rapid industrialisation. There was initially a movement away from heavy reliance on oil and gas exports and towards a more diversified economy based on a strongly export-oriented manufacturing sector. This was followed by a transformation of the structure of employment, away from the primary sector and towards manufacturing and services. Reform in the banking sector began in 1983, when the government removed ceilings on time deposit rates at state banks and lending controls at all banks. This was followed by a further banking deregulation package in 1988, which removed restrictions on the expansion of bank branch networks and the establishment of new banks, except in the case of purely foreign-owned banks (McLeod 1993; Aswicahyono and Feridhanusetyawan 2004). From March 1985 through May 1990, a number of other deregulation packages were launched to liberalise the economy, and especially to promote non-oil exports. The new policies included reform of export incentives and administrative improvements to simplify procedures by reducing the number of licences required to export goods. 5 As a result of the trade reform and deregulation, tariffs and non-tariff barriers (NTBs) declined significantly (Fane 1996; Hill 2000; Aswicahyono and Feridhanusetyawan 2004). Significant growth in exports, structural transformation and poverty reduction took place during this first period (Hill 2000). The sectoral transformation away from agriculture and towards manufacturing and services was mirrored in employment during the mid-1980s, with an expansion of jobs in footloose labour-intensive manufacturing sectors such as clothing, woven fabrics, footwear, electronics, furniture, yarn, toys and sporting goods and glass and glassware (Hill 2000). Previous researchers (Hill 2000; Temple 2003) have argued that Indonesia s growth during the 1980s was not only rapid but particularly favourable to the poor, because trade liberalisation policy meant that Indonesia could make better use of one of its abundant resources low-skilled labour. This was readily available in large quantities in agriculture, and could be drawn out of that sector into more modern activities in manufacturing, construction and other services, where its productivity would be higher and hence it could earn higher incomes. Thus policies at the time were consistent with the reality that the most straightforward way to reduce poverty is to increase the demand for 5 Hill (2000) points out that the strongest trade reform took place in the period, when protection was reduced progressively.

8 Poverty in Indonesia : the impact of growth and changes in inequality 85 TABLE 3 Structural Transformation of Employment Growth episodes Agriculture, Forestry & Fisheries Manufacturing Services Employment Growth (% p.a.) Average Share of Total Employment (%) Employment Growth (% p.a.) Average Share of Total Employment (%) Employment Growth (% p.a.) Average Share of Total Employment (%) First liberalisation period a Second liberalisation period Recovery period a Owing to data limitations, the first liberalisation period starts at 1987 for agriculture, forestry and fisheries and for manufacturing, and at 1989 for services, instead of at Data on employment in other sectors (mining and quarrying; electricity, gas and water; and construction) are not shown in this table. Source: CEIC Asia Database. labour supplied by the poor. Restructuring the economy in line with comparative advantage through trade liberalisation was an obvious way to do this. It has been argued that growth of the labour-intensive manufacturing sector, and of the economy as a whole, reduces poverty. For example, Manning (1998: ch. 5) and Osmani (2004) have argued that declining poverty during the first liberalisation period resulted mainly from the employment expansion created by economic growth, facilitated by slow growth in real wages. Table 3 illustrates this structural transformation of employment. The proportion of jobs in the agriculture, forestry and fisheries sector declined, while the share in manufacturing and the services sector increased. Average annual growth of employment in agriculture, forestry and fisheries during the first liberalisation period was 3.1%, whereas in the manufacturing sector it was more than three times as rapid, at 9.8%. Large quantities of labour were indeed drawn out of agriculture into more modern activities in manufacturing and services. During this period, the proportion of employed persons working in agriculture, forestry and fisheries was around 55.8% on average, while 9.2% of employment was in the manufacturing sector and 30.5% in services. The second liberalisation period ( ) During the period , GDP and GDP per capita grew even more rapidly, by 7.2% and 5.3% annually, while manufacturing output continued to grow at 9.9% (table 2). Mean consumption expenditure per capita grew slightly more rapidly than in the first liberalisation period, by 1.3% annually. Although the magnitude of the two per capita growth rates differs, both measures increased during the

9 86 Riyana Miranti second liberalisation period. Liberalisation through deregulation policies weakened, however (Hill 1997). Aswicahyono and Feridhanusetyawan (2004) have labelled the (slightly different) period as one of deregulation fatigue. Reforms were slower and less comprehensive, and there was some return to more interventionist policy (Fane 1996). For example, a new monopoly on clove trading was introduced in 1991, as were new restrictions on inter-island trade in oranges from West Kalimantan in 1991; the tariff surcharge on imports of propylene and ethylene was increased in The government also established a national car project involving Kia Motors of South Korea and the Indonesian car maker Timor Putra Nasional, under which Timor was exempted from paying luxury taxes (Fane 1996; Aswicahyono and Feridhanusetyawan 2004). During the second liberalisation period, and especially between 1993 and 1996, there was a decline in the rate of job creation, particularly in manufacturing and agriculture, probably indicating that the previous labour surplus had begun to be exhausted. Employment growth in the manufacturing sector was slower during the second liberalisation period, at 5.8%, than it had been in the first, while employment in the agricultural sector contracted by 1.9% per year (table 3). Do such data imply that growth was less pro-poor during this second liberalisation period? In fact, poverty fell by nearly six percentage points in these years, from 23.4% in 1990 to 17.6% in 1996 a slightly larger relative decline than in the first period (table 1). The faster rate of poverty decline was probably due to increases in real wages in all sectors, including construction, textiles and government administration (Manning 1998: ch. 5; Feridhanusetyawan 2002), as the rapid growth of manufactured exports finally began to have an impact on the general price of unskilled labour. Real wages in the manufacturing sector grew by 33% between 1990 and 1996 an annual growth rate of around 5%. On the other hand, as we will see below, inequality worsened during this period, with the Gini coefficient increasing noticeably, from in 1990 to in This seems to indicate that as trade liberalisation progressed, its impact was less heavily concentrated on the poor. The crisis period ( ) The Asian financial crisis hit Indonesia in A political crisis followed that culminated in Soeharto s resignation from the presidency in The economy fell into turmoil, with GDP contracting by more than 13% (and GDP per capita by more than 14%) in 1998 (table 2). Mean consumption expenditure per capita also fell, by 17% (Suryahadi, Sumarto and Pritchett 2003). It was decided not to include the period in the present analysis, because the growth figure for the period as a whole masks enormous underlying volatility: the GDP growth rate fell from 4.7% in 1997 to 13.1% in 1998, before bouncing back to 0.8% in Under such circumstances, and recalling that the Susenas consumption module data are produced only every third year, it would be very difficult to draw any meaningful interpretation of the overall change in the poverty level during this period of turmoil. The recovery period ( ) Following the crisis there was little by way of the kinds of policy reform that had been seen during the 1980s and early 1990s. Indeed, policy in the labour market

10 Poverty in Indonesia : the impact of growth and changes in inequality 87 area which is of crucial importance to poverty began to become much more interventionist. The economy rebounded, with GDP growing at an average rate of 4% per annum, and per capita GDP at 2.7% per annum (table 2), though this growth was much slower than that in the two periods of liberalisation. However, mean per capita consumption expenditure grew markedly in the recovery period, at 3.3% p.a. This growth was more rapid than that in the first and second liberalisation periods, and suggests that, coming from a very low base during the crisis, consumption was quick to recover relative to other sectors. Miranti (2007: 77 8) shows that this recovery of mean consumption expenditure was robust to the choice of deflator. The manufacturing sector recovered slowly to grow by only 4.2% per year during this period (table 2), less than half the rate during the two liberalisation phases. Despite the severe economic contraction during the crisis, unemployment had remained lower than expected (Feridhanusetyawan 2002). Instead, there was a shift of main occupation, with displaced workers moving from shrinking sectors to other sectors especially agriculture and the informal sector. In contrast with the two liberalisation periods, employment growth of 1.9% in agriculture, forestry and fisheries during the recovery phase, while slow, was slightly more rapid than growth of employment in manufacturing, where jobs expanded at just 1.7% annually; services employment grew by only 0.8% during this period (table 3). This adjustment reflected the flexibility of the labour market (Manning 2000; Feridhanusetyawan 2002). Nevertheless, poverty fell by 5.2 percentage points in this recovery period, from 23.4% in 1999 to 18.2% in 2002, a level only slightly higher than that in 1996 (17.6%) (table 1). Suryahadi and Sumarto (2001) categorised people with per capita consumption below the poverty line immediately after the crisis as either chronic or transient poor. The first category refers to those likely to remain poor in the future, and the second to those likely to increase their consumption sufficiently to elevate themselves above the poverty line in the future. Suryahadi and Sumarto found that the head-count ratio of transient poor increased from 12.4% of the population in 1996 to 17.9% in 1999, whereas the head-count ratio of the chronic poor increased by almost a factor of three, from 3.2% to 9.5%, during the same period. This shows that rising numbers of chronic poor contributed most to the increase in poverty, such that their share of the total rose from 20% to 35%. The fact that poverty declined during the recovery period may reflect the impact of renewed GDP growth in raising the consumption of those in transient poverty. Thus it is not surprising that inequality worsened during this period, because the recovery may have had less impact on the chronic poor than on other groups. Changes in poverty and inequality Table 4 shows the trends in poverty and in inequality (as indicated by the Gini coefficient) during the three periods under consideration. During the first liberalisation period, poverty fell by about six percentage points (from 29.5% in 1984 to 23.4% in 1990), while inequality declined very slightly from to Poverty again fell by almost six percentage points a larger relative decline during the second liberalisation period, even though inequality increased noticeably, to in 1996.

11 88 Riyana Miranti TABLE 4 Poverty and Inequality Year Head-count Poverty (%) Gini Coefficient First liberalisation period Second liberalisation period Recovery period Source: BPS, Susenas, various years These changes were reversed during the crisis, but the recovery period saw poverty fall back to 18.2%, only a little higher than its level in 1996, while inequality returned to around its 1984 level (table 4). We turn now to estimate the extent to which changes in poverty can be attributed to economic growth, on the one hand, and changing inequality on the other. The following section provides the empirical methodology for this estimation. EMPIRICAL METHODOLOGY AND APPROACH An absolute poverty measure will be a strictly decreasing function of an increase of mean consumption (growth), given a fixed poverty line and fixed income distribution (or given that the consumption of each individual changes in the same proportion) (Ravallion 1995). This means that when income distribution is held constant, a positive change in mean consumption expenditure per capita (growth) will reduce poverty. This assumption of unchanging income distribution has been made in various studies that measure the impact of economic growth on poverty, including Datt and Ravallion (1992); Kakwani (1993); and Bourguignon (2003). Thus the growth elasticity of poverty (GEP) is defined as the percentage reduction in poverty given a 1% increase in consumption expenditure per capita, with other factors held constant. In reality, however, income distribution, or the level of inequality, is also likely to vary over time. In this paper, the degree of inequality of the income distribution is measured by the Gini coefficient, which has a value between zero and one. A value of zero means perfect equality, such that everyone in the

12 Poverty in Indonesia : the impact of growth and changes in inequality 89 population has the same level of income. A value of one indicates perfect inequality, where one person accounts for all income. More generally, the smaller the Gini coefficient, the more equal the distribution of income. An increase in the Gini coefficient is likely to contribute positively to absolute poverty, other things being equal. The inequality elasticity of poverty (IEP) is defined as the percentage change in poverty given a 1% increase in the Gini coefficient, with other factors held constant. To estimate the growth and inequality elasticities of poverty, I follow a basic model suggested by Ravallion and Chen (1997): e= 6 it, 0 1 it, 2 it, e= 1 e t ln P = γ + γ ln MEAN + γ ln GINI + β d + δ where P it, is poverty in province i at time t (%); + ε it, (1) MEAN it, is mean consumption expenditure per capita (Rp/month, 1984 prices) for province i at time t, as a proxy for income; GINI it, is the Gini coefficient for province i at time t, as a proxy for inequality. t is the year index (t = 1984, 1987, 1990, 1993, 1996, 1999, 2002); d t are the year dummies for the years when the Susenas was conducted: d87 = 1 if t = 1987, and 0 otherwise; d90 = 1 if t = 1990, and 0 otherwise; d93 = 1 if t = 1993, and 0 otherwise; d96 = 1 if t = 1996, and 0 otherwise; d99 = 1 if t = 1999, and 0 otherwise; d02 = 1 if t = 2002, and 0 otherwise. The base period is δ i is the province fixed effect (unobserved heterogeneity). ε it, is a white-noise error term that includes errors in the poverty measure. As equation (1) is in logarithmic format for both dependent and independent variables, the coefficient of ln MEAN refers to a 1% change in monthly mean consumption per capita (as a proxy for income) and the coefficient of ln GINI refers to a 1% change in inequality. While Ravallion and Chen (1997) and Adams (2004) use first differences estimation, this paper uses fixed effects, for the following reasons. First, fixed-effects estimation allows the use of all the information available, even though these are unbalanced panel data. They are unbalanced because there are several missing observations for poverty in 1990 in nine provinces (BPS did not publish the data for those provinces in that year), and because of the absence of consumption data for Aceh, Maluku and Papua in 2002 (when a Susenas module could not be undertaken in these provinces because of significant social and political unrest). Second, the fixed-effects approach is more efficient than the first-differences approach, because in any province random errors are usually assumed to be serially independent that is, not serially correlated with each other across time periods (Wooldridge 2003). Equation 1 assumes that the impact of growth on poverty is similar across periods, but responds to the possibility of omitted variable bias by incorporating province and year dummies. The latter also capture, in general, the impact of

13 90 Riyana Miranti macroeconomic conditions in each Susenas consumption module period. Comparison of the coefficients of the year dummies with that for the base period of 1984 will indicate whether poverty was higher on average (if there is a significant positive sign) or lower (if there is a significant negative sign) than for the base period, after the analysis has controlled for changes in consumption and in inequality. However, we also aim to discover whether there are differences in the growth and inequality elasticities of poverty across development episodes. Equation 2 therefore incorporates a set of interaction terms intended to determine whether the impact of growth and inequality on poverty differs across development episodes. The six year dummies are still included to avoid omitted variable bias and to capture, in general, the impact of macroeconomic conditions in each Susenas consumption module period. In addition, there are interactions between the second liberalisation period (EPISODE 2 ) and the recovery period (EPISODE 3 ) with the growth variable (ln MEAN). Similarly, in calculating the IEP across development episodes, there are interactions between the Gini coefficient variable (ln GINI) and both the second liberalisation period (EPISODE 2 ) and the recovery period (EPI- SODE 3 ). The base period is the first liberalisation period, Thus we now have: ln Pit, = ρ0 + ρ1ln MEANit, + ρ2 ln GINIit, + ρ3episode2, t* ln MEANit, + ρ EPISODE *ln MEAN + ρ EPISODE * GINI + 4 3, t i, t 5 2, t ln e= 6 ρ EPISODE *lngini + β d + δ + ε 6 3, t i, t e t i i, t e= 1 where: EPISODE jt,, j = 2, 3, are the dummies for development episodes; 6 EPISODE 2, t = 1 if t = 1990, 1993 or 1996, and 0 otherwise (the second liberalisation period); and EPISODE 3, t = 1 if t = 1999 or 2002, and 0 otherwise (the recovery period). The reference development episode is (the first liberalisation period). To test the structural stability of the regression model (that is, to check whether there is a structural change in the overall relationship between growth and poverty and between inequality and poverty), I performed an F-test of joint significance of the interaction terms. it, (2) EMPIRICAL RESULTS Full period We begin by assuming that the relationships between growth and poverty and inequality and poverty remained unchanged throughout the whole period. The middle column of table 5 shows that, after we had controlled for inequality, the GEP (ln MEAN) was 2.43 during the period This means that a 10% increase in consumption per capita reduced poverty proportionately by 24.3%. (By way of illustration, suppose that a particular province had a real mean 6 The dummies for the development episodes are used to capture the macroeconomic conditions during the specific development episodes.

14 Poverty in Indonesia : the impact of growth and changes in inequality 91 TABLE 5 Regression Results a Explanatory Variable For Full Period (Equation 1) For Individual Development Episodes (Equation 2) b ln MEAN 2.43*** 2.37 *** ( 21.68) ( 21.62) ln GINI 0.92*** 0.78 *** (5.85) (5.11) EPISODE 2 *ln MEAN 0.09 ** ( 2.01) EPISODE 3 *ln MEAN 0.12 * ( 1.86) EPISODE 2 *ln GINI 0.32 ** (2.01) EPISODE 3 *ln GINI 0.52 *** (2.56) d d d * 0.07 d *** 0.36 *** d ** 0.43 d *** 0.53 Constant 29.94*** *** R-sq within a ***, ** and * denote significance at the 1%, 5% and 10% levels, respectively. The figures in parentheses are t ratios. The number of observations is 189. b For equation (2), the first liberalisation period, , is the reference period. Therefore the coefficients of ln MEAN and ln GINI in the last column refer to the first period of liberalisation. consumption level of Rp 100,000 per month and a head-count poverty ratio of 20%. At a later time, when consumption had increased by 10% to Rp 110,000, poverty would have decreased by 24.3% of 20%, to 15.1%, other things being equal.) The IEP (ln GINI) was 0.92, meaning that a 10% increase in inequality (as measured by the Gini coefficient) increased poverty proportionately by 9.2%. The GEP in table 5 is similar to that found by Friedman (2005) for Indonesia over the period It is also comparable with that found in studies of other countries that also used the head-count poverty ratio measure as the dependent variable. For Bangladesh, Wodon (2002) found growth elasticities for in the range 1.6 to 2.6, depending on which of various poverty lines was used; this was slightly higher than the GEP for Thailand of 2.2 in (Deolalikar 2002). At 0.92, the IEP in table 5 is lower than those estimated by Friedman (2005), which ranged from 1.3 to 1.9. It is also much smaller than those found in other countries. For example, Thailand (Deolalikar 2002) recorded an IEP of 3.0, which is around three times higher.

15 92 Riyana Miranti TABLE 6 Summary of Growth and Gini Elasticities of Poverty, Development Episode Growth Elasticity of Poverty a Inequality (Gini) Elasticity of Poverty First liberalisation period Second liberalisation period Recovery period Entire period a See text for an explanation of growth as used in this paper. Source: Author s calculations. The estimations were carried out by applying the fixed-effects approach to control for provincial heterogeneity, an approach similar to that of applying ordinary least squares (OLS) estimation, with 25 provincial dummies as explanatory variables. The results show that, with the province of Papua as the base category, all province dummies except that for East Kalimantan are significant at the 1% level. If other factors are held constant, including inequality, the other 24 provinces on average had head-count poverty ratios that fell short of that estimated for Papua, by percentages ranging from 26% (Riau and the Jakarta Special Region) to 82% (South Sulawesi). Poverty across development episodes What happened to the GEP across periods? The last column of table 5 shows that both elasticity measures differed across the three development episodes. Table 6 provides a summary (in which the elasticities for each period are obtained by adding the coefficients on the interaction terms to those for the first liberalisation period that is, the reference development episode). The GEP was 2.37 during the first liberalisation period. The coefficients of the interaction terms EPISODE 2 *ln MEAN and EPISODE 3 *ln MEAN are 0.09 and 0.12, respectively (table 5), both being significantly different from zero at the 1% level. This means the GEP for the second liberalisation period was 2.46, increasing slightly in magnitude to 2.49 in the recovery period. This is not unexpected. While poverty was lower during the second and third periods than during the first, simple arithmetic indicates that the GEP will tend to increase as poverty declines. From a different perspective, with the manufacturing sector now quite large, a given amount of growth in manufacturing would create far more job opportunities for people who would otherwise still be in agriculture than it would have done when manufacturing was relatively small. Table 6 also shows the IEP across the three periods. The elasticity was 0.78 during the first liberalisation period, meaning that a 10% increase in the Gini coefficient would increase poverty by 7.8%, other things being equal. The impact of inequality on poverty was higher during the second liberalisation period and the recovery period, however, when the IEPs were 1.10 and 1.30, respectively. These results lend little support to the hypothesis that the GEP differed significantly across different development episodes. It turns out that growth was pro-poor regardless of the development episode. However, the IEP during the

16 Poverty in Indonesia : the impact of growth and changes in inequality 93 first liberalisation period was significantly different from that during the second liberalisation and recovery periods. When the different development episodes are taken into account, it is interesting to find that the only year dummy coefficient that was significant at the 1% level was that for This is in contrast to the full period estimation, in which the 2002 dummy coefficient was also significant at 1%. This indicates that growth of mean consumption expenditure per capita and changes in inequality have explained almost all the change in poverty, and that macroeconomic conditions played a role only in (However, all the coefficients on the province dummies, except that for East Kalimantan not shown here are still significant at the 1% level.) As with the estimates for the full period, so in the individual development episodes, other things being equal, the other 24 provinces on average had estimated head-count poverty ratios lower than that in Papua, in this case by percentages ranging from 24% (Riau) to 81% (South Sulawesi). Quantifying the growth and inequality effects on poverty Having measured the magnitude of the growth and inequality elasticities of poverty across different development episodes, and knowing the amount of growth and degree of change in the Gini coefficients that actually occurred, it is possible to calculate the average implied change in poverty resulting from growth (changes in consumption expenditure) and changes in inequality. For example, the coefficient γ 1 in the main poverty equation (equation 1) represents the impact of growth on poverty when other variables are held constant. The magnitude of this partial impact of growth on poverty can be estimated as follows: P t ln P t P = ln MEAN 1 t t MEAN t MEAN t = 1 P t = ( 1 * MEAN t * P t ) MEAN t (3) In the same way, the partial impact of a change in the Gini coefficient on poverty can also be estimated: P ln P t t P = ln GINI 2 t t GINI t GINI t = 2 P t = ( 2 * GINI t * P t ) GINI t (4) Table 7 shows the respective contributions of growth (as defined here) and change in inequality to the change in national poverty. When we apply the estimated parameters from the regressions above (using the growth and inequality elasticities of poverty shown in table 6), the contribution of growth to the decline in poverty can be seen to be largest during the recovery period, while its lowest contribution was during the second liberalisation period. By contrast, changes in inequality resulted in an offsetting impact on poverty, particularly during the second liberalisation and recovery periods.

17 94 Riyana Miranti TABLE 7 Contribution of Growth and Inequality to Change in Poverty Development Episode Growth a ( MEAN) Inequality Change ( GINI) Consumption at Beginning of Period (Rp/month, 1984 prices) Inequality at Beginning of Period (Gini coefficient) Contribution to Poverty Change Growth a (%) Inequality Change (%) Total Change (%) First liberalisation period Second liberalisation period Recovery period Entire period 14, , , , , , , , a See text for an explanation of growth as used in this paper. Source: Author s calculations. The negative impact of change in inequality on poverty during the first liberalisation period seems credible, given that inequality declined slightly during this period (table 4). The structural transformation away from agriculture and towards labour-intensive manufacturing and services created many new job opportunities, and this impact was probably more significant for unskilled labour which is found near the bottom of the income distribution thus reducing inequality. The offsetting impact on poverty of changes in inequality during the second liberalisation period may reflect the fact that liberalisation had begun to benefit the population more widely, rather than mainly just the poor. The Gini coefficient increased rather noticeably during this period (table 4). Further, as a study by Akita and Alisjahbana (2002) reveals, the rapid GDP growth just before the 1997 economic crisis was characterised by increasing within-province inequality, particularly in Riau, and in Jakarta and West and East Java, where the manufacturing sector was mainly concentrated. The offsetting impact of increased inequality on poverty during the recovery period suggests that the recovery might have affected those in transient poverty to a greater extent than those in chronic poverty. A thorough examination of the factors that might explain the different signs of the contribution of changes in inequality to poverty changes requires further research, however. CONCLUSIONS This paper has used seven waves of the Susenas consumption module to track the dynamics of poverty change across three development episodes. It began with the presumption that different policies followed during each of these episodes could be expected to have differential impacts on poverty during the respective

18 Poverty in Indonesia : the impact of growth and changes in inequality 95 periods, especially through the employment effect of labour-intensive growth. Various analysts have argued that, viewed from the demand side, the growth of export-oriented, labour-intensive manufacturing can be seen as one factor among several that help to explain Indonesia s good record of poverty alleviation since the early 1980s. (At the same time, on the supply side, improvements in education and health standards have also benefited the poor.) However, our results show that the GEP was remarkably stable across the three development episodes, moving within the narrow range 2.37 to The presumption of substantial differences in the GEP between the sub-periods is therefore not supported by the data. By contrast, the inequality elasticity of poverty ranged much more widely between 0.78 and 1.30 across the development episodes, with the change between the first and second periods of liberalisation especially noticeable. The results also show that change in inequality re-inforced the impact of growth on poverty only during the first liberalisation period, when inequality was reduced. In the other two periods, worsening inequality tended to offset the declines in poverty resulting from growth. Further research will be necessary for a full understanding of these phenomena, and will need to control for other explanatory variables that are expected to influence the GEP in particular, the relevant range of government policies. The practical reality, however, is that it is difficult perhaps impossible to find quantitative measures of the nature of government policies that could be used in analysing the relationship between these policies and changes in inequality, economic growth and poverty. REFERENCES Adams, R.H. Jr (2004) Economic growth, inequality and poverty: estimating the growth elasticity of poverty, World Development 32 (12): 1,989 2,014. Akita, T. and Alisjahbana, A.S. (2002) Regional income inequality in Indonesia and the initial impact of the economic crisis, Bulletin of Indonesian Economic Studies 38 (2): Aswicahyono, H. and Feridhanusetyawan, T. (2004) The evolution and upgrading of Indonesia s industry: how can Indonesia maintain its export competitiveness?, CSIS Working Note 073, Centre for Strategic and International Studies (CSIS), Jakarta. Balisacan, A.M., Pernia, E.M. and Asra, A. (2003) Revisiting growth and poverty reduction in Indonesia: what do subnational data show, Bulletin of Indonesian Economic Studies 39 (3): Bidani, B. and Ravallion, M. (1993) A regional poverty profile for Indonesia, Bulletin of Indonesian Economic Studies 29 (3): Bourguignon, F. (2003) The growth elasticity of poverty reduction: explaining heterogeneity across countries and time periods, in Growth and Inequality, eds T. Eichler and S. Turnovsky, MIT Press, Cambridge MA. BPS (Badan Pusat Statistik, Central Statistics Agency) (2003) The Methodology and Poverty Profile for 2002 [Metodologi dan Profil Kemiskinan Tahun 2002], BPS, Jakarta. Datt, G. and Ravallion, M. (1992) Growth and redistribution components of changes in poverty measures: a decomposition with applications to Brazil and India in the 1980s, Journal of Development Economics 38: Deaton, A. (1997) The Analysis of Household Surveys: A Microeconometric Approach to Development Policy, Johns Hopkins University Press, Baltimore MD. Deaton, A. (2001) Counting the world s poor: problems and possible solutions, World Bank Research Observer 16 (2):