34487 The Effects of the Indonesian Economic Crisis on Agricultural Households: Evidence from the National Farmers Household Panel Survey (PATANAS) A report prepared by The Center for Agro-Socioeconomic Research Bogor, Indonesia and The World Bank Washington, DC, USA July 31, 2000 The Effects of the Indonesian Economic Crisis on Agricultural Households: Evidence from the National Farmers Household Panel Survey (PATANAS) Prepared by Daniel O. Gilligan, Hanan Jacoby, Jaime Quizon This report is part of a joint study by The World Bank and CASER on the effects of the economic crisis on agricultural households in Indonesia. The World Bank team would like to thank the entire CASER/ASEM staff for their diligence in collecting and cleaning the data. In particular, we would like to express our appreciation to Made Oka, Sjaiful Bahri, Sumaryanto, Chaerul Saleh, Muchiddin Rahmat, Bambang Sayaka, Benny Rahman, Reny Kustiari, Suprapto Djojopuspito, and Sri Heri. We would also like to thank Christina Neagu for her diligent assistance in cleaning the data. This research was financed by Asia-Europe Meeting (ASEM) Trust Fund Grant No. 021688. The findings and interpretations expressed here are not necessarily those of the ASEM, World Bank or it Executive Directors. i TABLE OF CONTENTS I. INTRODUCTION....................................................................................................................................................1 MACROECONOMIC INDICATORS OF CRISIS EFFECTSON AGRICULTURE..................................................................4 THE PATANAS SURVEY DATA.....................................................................................................................................9 II. IDENTIFYING VULNERABLE HOUSEHOLDS .....................................................................................13 III. THE EFFECTS OFTHE CRISIS ON RURAL HOUSEHOLDS .........................................................23 EVIDENCEON SHOCKS TO HOUSEHOLD INCOME AND SOURCES............................................................................23 PRICE AND WAGE CHANGES DURING THE CRISIS....................................................................................................28 THE EFFECTS OF EL NIÑO..........................................................................................................................................33 THE EFFECT OF THE CRISIS ON AGRICULTURAL PRODUCTION AND INPUT USE...................................................35 Income From Crops..............................................................................................................................................35 Crop Choice...........................................................................................................................................................41 Cost of Production................................................................................................................................................43 Fertilizer Use.........................................................................................................................................................44 On-Farm Labor Demand.....................................................................................................................................48 IV. OTHER HOUSEHOLD STRATEGIES FOR ADAPTING TO CRISIS EFFECTS .......................51 MIGRATION..................................................................................................................................................................51 V. THE EFFECTIVENESS OF GOVERNMENT PROGRAMS TO PROTECT THE POOR..........53 PADAT KARYA WORK PROGRAMS AND OTHER GOVERNMENT TRANSFERS............................................................53 SUBSIDIZED RICE PROGRAM......................................................................................................................................57 VI. CONCLUSIONS AND POLICY IMPLICATIONS ..................................................................................59 REFERENCES...............................................................................................................................................................62 ii Executive Summary This report examines the effects of the 1997 Indonesian economic crisis on the socio- economic status of rural households, and of agricultural households in particular, using the National Farmers Household Panel Survey (PATANAS). The PATANAS survey captures the structure of agricultural production and all sources of income for nearly 1500 rural households in 35 villages from six provinces during the 1994/95 and 1998/99 crop years. This research considers the effects of the crisis on-farm and off-farm incomes, agricultural production, migration, and employment, with a particular emphasis on crisis effects on poor households. The main findings of the report are as follows: · The effects of the crisis on rural households were heterogeneous, but the data reveals significant income gains across many broad categorizations of households from 1995 to 1999. · Farmers generally benefited from the crisis. As expected, crop portfolios largely determined gains in crop income; tree crop farmers, located off java, did especially well, as did rice farmers. Growers of crops for domestic markets did worse. · Surprisingly, non-farm income also rose substantially, though not for the poor. These gains were driven by business income on Java and by labor earnings off Java. Undoubtedly, some, but not all, of these gains were realized prior to the crisis (from 1995 to 1997). · The income boom enjoyed by many households in the outer islands generally did not emerge on Java as per capita incomes there stagnated. Landless Javanese households, in particular, were among the worst off in relative terms. · In general, poorer households experienced smaller income gains than richer households. · Migration increased after the onset of the crisis. The share of arrivals to rural areas from cities increased. Also, people moved greater distances once the crisis began, with a much larger share of movers traveling outside their sub-district. · Landless households and agricultural wage earners performed worse than land- owning, farming households. Wage earners did not appear to share in the benefits of the terms of trade boom experienced by producers. · Average intensity of fertilizer use declined for most producers, with the exception of rice farmers. · Household labor use on farm increased sharply during the crisis while hired labor demand per hectare was unchanged. Within the household, female household iii members increased hours worked on-farm much more in percentage terms than did male members. iv I. Introduction The economic crisis that has plagued Indonesia since 1997 is its worst crisis in thirty years and has had considerable negative effects on the welfare and livelihood of millions of Indonesians. The crisis began in August 1997 with a run on the Indonesian rupiah after similar currency crises in Thailand, Malaysia and the Philippines. By January of 1998, the rupiah had lost 75% of its value, falling from Rp. 2,599/$US to Rp. 10,375/$US. Throughout the following year, the exchange rate fluctuated widely from Rp. 7,500/$US to Rp. 14,900/$US in response to economic and political events. The initial currency crisis was fueled by capital flight that eventually led to a serious banking crisis. As interest rates jumped to six times their pre-crisis levels and uncertainty climbed, banks refused to make even routine loans to businesses to cover shipment of goods to customers. Production slowed dramatically and in 1998 the economy contracted by 13.4%. Transmission of higher tradables prices into the economy lead to substantial inflation, which reached 78% in 1998 according to the Central Bureau of Statistics (BPS). These dramatic economic events precipitated a political crisis as well. Street demonstrations following rice and kerosene price increases in November 1997 focused attention on corrupt practices of President Suharto, his family and associates. The ensuing political uncertainty further weakened the rupiah and boosted inflation, contributing to a downward spiral that both deepened and prolonged the recession and led to President Suharto's resignation in May 1998 after more than thirty years of rule. As industrial production ground to a halt, national unemployment figures rose and the poverty rate increased. Initial predictions based primarily on national accounts and industrial production data suggested steep increases in unemployment and a near tripling of poverty rates. However, initial results based on micro-data sources including the 100 Villages survey, the qualitative national Kecamatan survey and the Indonesian Family Life Survey provide a more complex picture of crisis impacts (see, for example, Poppele, Sumarto, and Pritchett (May, 1999) and Frankenberg, Thomas, and Beegle (March. 1999)). While these studies found substantial unemployment and worsening conditions in the cities, they concluded that the experience in rural areas was more heterogeneous and generally less severe. The purpose of this research is to examine more closely the effects of the crisis on the socio-economic status of households in rural areas, and of agricultural households in particular, using the National Farmers Household Panel Survey (PATANAS). The PATANAS survey is the most detailed socio-economic survey of agricultural households conducted in Indonesia. As such it provides a unique opportunity to assess how rural households were affected by the crisis and how their responses differed based on cropping patterns, landholdings, and modern input use, for example. The PATANAS survey captures the structure of agricultural production and all sources of income for the same 1494 rural households in 35 villages from six provinces during the 1994/95 and 1998/99 crop years. These farm income modules, collected in 1995 and 1999, include detailed seasonal, plot-level data on crop output, labor and non-labor 1 inputs, on-farm household labor supply, land characteristics, and livestock ownership and production. In addition, supporting data were collected in 1994 and 1996-1998 on varying topics that included land ownership and contracting, labor relations, asset ownership and consumption expenditure. All of these topics were included in a comprehensive survey instrument used in the most recent survey round conducted in April-May 1999. The households surveyed are in the provinces of Lampung, Central Java, East Java, Nusa Tengara Barat (NTB), North Sulawesi, and South Sulawesi. While the sample is not statistically representative of these provinces, villages were selected to capture the primary crops and agroclimatic zones within each province. This makes the PATANAS survey well suited to identifying how the effects of the economic crisis varied over the diverse cropping regions of these provinces. The timing of the PATANAS survey rounds is well-suited to capturing the effects of crisis. The baseline survey data, collected in October 1995 on the previous 12 months, covers a period of relatively "normal" pre-crisis rural economic activity. It also pre-dates the drought caused by El Niño in 1997. The period covered by the most recent round of PATANAS begins in the second quarter of 1998, just after prices began to rise significantly throughout Indonesia. These 12 months from April 1998 to March 1999 represent a period during which some of the greatest impacts of the crisis should have been felt. There are several avenues through which the economic crisis could effect the socio- economic status of rural households. The sharp exchange rate depreciation shifted terms of trade in favor of tradables relative to non-tradables and fueled a general increase in prices. Households that primarily produce crops whose prices are linked to world prices should have experienced an income boom while those whose livelihood is based on production of non-tradable goods and services may have suffered real income losses. The recession created a drop in domestic demand for many agricultural commodities and boosted unemployment, particularly in urban areas. The demand contraction will reduce incomes most for producers of beef and luxury crops. If reports made during the crisis of substantial migration from urban to rural areas are true, there may be a decline in rural wage rates and changes in customary labor arrangements. The tightening of capital markets also will affect farmers' ability to borrow, reducing use of fertilizer and other purchased inputs. Rural households' asset holdings may also have been affected as the credit crunch drove down prices and as households suffering adverse income shocks sell off assets in order to maintain consumption. We will pursue each of these avenues to answer a number of important questions about crisis effects on rural households. In particular, we will consider whether the poor disproportionately felt the negative effects of the crisis. We will also identify regional differences in the extent and nature of the crisis. In addition, we will treat as hypotheses to be tested a number of preliminary results from other studies on the effects of the crisis on rural households as well as some of our own hypotheses. Among the standing hypotheses we address are the following: 2 · In rural areas, the poorest households were worst hit by the crisis. Thomas et al, (RAND, March, 1999) report that households at the top and bottom of the income distribution suffered the greatest declines in per capita expenditure in a sample of urban and rural households. Their results show that the distribution of crisis effects on rural household socio-economic status depends critically on the choice of price deflator used to calculate real income (or expenditure). We test for differential effects of the crisis on poor households identified by per capita landholdings. · Households on Java were affected more severely by the crisis than households in other regions (Poppele, Sumarto and Pritchett, SMERU/World Bank, May, 1999). This result follows presumably from the closer linkages of rural Javanese households to contracting urban industrial centers than rural households in the other provinces. · While male labor supply was relatively unchanged, female labor supply increased after the start of the crisis. (Feridhanusetyawan, CSIS, August, 1999 and Frankenberg, Thomas and Beegle, RAND, March, 1999). While these earlier studies considered labor market participation by gender, we will test for gender- based differences in changes in on-farm labor supply. In addition, we will test several other hypotheses about crisis effects on agricultural households: · Urban to rural migration increased after the onset of the crisis. · Because rice prices rose while real wages fell, landless households and those earning most of their income from wages were hurt by the crisis. Although a time series comparison of consumption is not possible with the PATANAS data, we will determine whether growth in per capita income for households in these groups declined or was lower than for the sample as a whole. · Because agriculture offers an array of tradable outputs but has incomplete linkages to industrial sectors, it acted as a form of "safe haven" from the effects of the economic crisis. Non-farm income sources, which are more closely linked to depressed industrial centers and urban labor markets, will suffer a recession. As a result, households with a smaller share of pre-crisis agricultural income suffered worse effects from the crisis. · Within agriculture, producers of tradable commodities experienced an income boom as a result of the crisis. Demand for agricultural inputs is subject to competing effects from the crisis. For example, because the government heavily controlled the increase of fertilizer prices until November 1998, there was room for greater demand for fertilizer from tradable crop producers. However, the credit crunch brought on by climbing interest rates should have dampened fertilizer demand. Given the fact that 1998/99 was also a year of recovery from 3 drought, we expect that farmers were relatively cash starved, so that demand for fertilizer and other purchased inputs declined. Using the baseline survey from 1995, we will identify households that were vulnerable to macroeconomic shocks based on household characteristics, asset holdings, and crop composition. We will then consider how increased prices and urban unemployment affected farming households. Because the depreciation of the rupiah led to a sharp increase in tradables prices, the effect of the crisis on farmers' terms of trade and the welfare of farm workers depends on crops grown, use of modern inputs, and transmission of price changes to rural areas. For example, while landless households may be most vulnerable to rising food prices and uncertain employment opportunities, farm laborers in cocoa- and coffee-growing regions likely benefited from the substantial rise in the rupiah price of these commodities following exchange rate depreciation. We will also assess households' mechanisms for coping with crisis effects through changes in cropping patterns and changes in household labor supply across income-generating activities. We will compare the effectiveness of these coping strategies to the benefits of government programs designed to mitigate crisis effects on rural households. Macroeconomic Indicators of Crisis Effects on Agriculture The major forces contributing to the economic crisis are borne out in the macroeconomic indicators. Figure 1 shows the monthly market exchange rate of the rupiah to the US dollar from the period of the first PATANAS survey in late 1995 until November 1999. Indonesia's central bank, Bank Indonesia, allowed the rupiah to float in August 1997, one month after Thailand's unsuccessful attempt to defend the Thai baht. The rupiah depreciated steadily from its July level of Rp. 2,599/$US until January 1998 when it fell dramatically to Rp. 10,375/$US. The exchange rate depreciated sharply again following President Suharto's resignation in May 1998. The rupiah recovered somewhat and last year fluctuated between Rp. 9000/$US and Rp. 6500/$US or only 30-40% of its pre-crisis value. In general, the exchange rate depreciation increases the price of tradable products and factors relative to non-tradables. This means an increase in farmgate prices for commodities such as cocoa, coffee, tea, and to a lesser extent rice whose domestic prices are determined in part by world market prices. However, rural farmers also face higher prices for imported inputs such as fertilizers and engineered seed varieties. Prices of most commodities rose sharply in early 1998 as the exchange rate depreciation fed increases in tradables prices and as the government removed subsidies on necessities such as rice and kerosene as part of an IMF-sponsored reform program. Many non- tradable commodities experienced price increases due to higher transport and processing costs. In addition, speculation and social unrest interrupted trade in many regions. On Java, traders occasionally kept trucks off the roads out of fear of looting. Inflation continued to accelerate until the last quarter of 1998; by December, the government's urban-based CPI posted an annual inflation rate of 77.6%. 4 Figure 1: Monthly Market Exchange Rate 16000 14000 12000 10000 8000 Rupiah/$US 6000 4000 2000 0 Oct- Feb- May- Aug- Dec- Mar- Jun- Sep- Jan- Apr- Jul- Nov- Feb- May- Aug- Dec- 95 96 96 96 96 97 97 97 98 98 98 98 99 99 99 99 Source: International Financial Statistics, IMF, various years. Figure 2 shows urban and rural consumer price indices and the rural producer price index for the six provinces included in the PATANAS survey. In each case, the urban CPI series is for the capital city and is taken from the data used by BPS to calculate the national CPI. The rural CPI and PPI series are regional averages taken from the Farmers Terms of Trade data collected by BPS.1 By mid 1999, prices had risen to roughly two- and-a-half times their pre-crisis levels. For the provinces on Java, as well as NTB and North Sulawesi, rural price increases were considerably greater than in the capital cities. This reflects the relatively greater increase in food prices in these provinces, since food has a larger share of the consumption basket in rural areas. From the onset of the crisis until mid-1999, consumer prices rose more than producer prices in Lampung, North Sulawesi, Central Java and East Java (see Figure 2). This amounts to a terms of trade shift against agricultural households in general. However, as shown below, these aggregate price indices mask substantial heterogeneity in the effects of changing prices across commodities. Inflation remained high in all six provinces throughout the period covered by the 1999 survey. This raises some methodological issues that will be addressed below concerning how to deflate income and expenditure figures between 1995 and 1999. 1We would like to thank Kai Kaiser for providing us with the Farmers Terms of Trade data. 5 Figure 2: Provincial Rural and Urban Consumer Price Indices Bandar Lampung, Lampung Semarang, C. Java 400 400 350 350 300 300 100 100 250 = 250 = 200 200 1994 150 1994150 100 100 50 50 Jan- Jul- Feb- Sep- Apr- Nov- Jun- Jan- Aug- Mar- Oct- Jan- Jul- Feb- Sep- Apr- Nov- Jun- Jan- Aug- Mar- Oct- 94 94 95 95 96 96 97 98 98 99 99 94 94 95 95 96 96 97 98 98 99 99 urban CPI rural PPI rural CPI (FTT IB) urban CPI rural PPI rural CPI (FTT IB) Surabaya, E. Java Mataram, NTB 400 400 350 350 300 300 100 100 = 250 = 250 200 200 1994150 1994150 100 100 50 50 Jan- Jul- Feb- Sep- Apr- Nov- Jun- Jan- Aug- Mar- Oct- Jan- Jul- Feb- Sep- Apr- Nov- Jun- Jan- Aug- Mar- Oct- 94 94 95 95 96 96 97 98 98 99 99 94 94 95 95 96 96 97 98 98 99 99 urban CPI rural PPI rural CPI (FTT IB) urban CPI rural PPI rural CPI (FTT IB) Manado, N. Sulawesi Ujung Pandang, S. Sulawesi 400 400 350 350 300 300 100 100 = 250 = 250 200 200 1994150 1994150 100 100 50 50 Jan- Jul- Feb- Sep- Apr- Nov- Jun- Jan- Aug- Mar- Oct- Jan- Jul- Feb- Sep- Apr- Nov- Jun- Jan- Aug- Mar- Oct- 94 94 95 95 96 96 97 98 98 99 99 94 94 95 95 96 96 97 98 98 99 99 urban CPI rural PPI rural CPI (FTT IB) urban CPI rural PPI rural CPI (FTT IB) Source: Urban CPI data is from the CPI series published by the Central Statistical Bureau, BPS. The rural CPI and PPI figures are from the Farmers Terms of Trade (FTT) survey collected by BPS. We would like to thank Kai Kaiser for providing us with the FTT data. 6 The economic crisis led to a dramatic tightening of credit markets. Bank Indonesia raised interest rates substantially in a bid to attract capital back into the country and shore up the exchange rate. Figure 3 shows the interest rate paid by Bank Indonesia on 28 day SBI certificates, one of the government's primary macroeconomic policy instruments. The figure shows that interest rates were raised from about 11% to roughly 22% at the beginning of the crisis, and then were increased sharply to 70% in mid-1998. Once inflation began to cool, the government reduced interest rates to near pre-crisis levels in mid-1999. Figure 3: Bank Indonesia Discount Rate on 28 Day Certificates 80 70 60 % 50 40 Annual30 20 10 0 Oct- Feb- May- Aug- Dec- Mar- Jun- Sep- Jan- Apr- Jul- Nov- Feb- May- Aug- Dec- 95 96 96 96 96 97 97 97 98 98 98 98 99 99 99 99 Source: Bank Indonesia Monthly Report, September 1999, and International Financial Statistics, IMF, various years. Higher interest rates raised the cost of debt to Indonesian firms. In addition, the rupiah depreciation substantially increased the cost of debt repayment for firms with dollar- denominated debt. This exposed the weakness of a number of Indonesian banks, with massive loan defaults leading to a major restructuring of the banking sector which is still under way. The result of the credit crunch was a sharp reduction in industrial activity in 1998 as many firms could not even obtain short-term loans to cover shipment of goods to retailers. The depth of the crisis is made clear by the figures on national income. Table 1 shows real gross domestic product (GDP) and GDP growth rates by sector from 1994-1999. Production began to slow in the second half of 1997, which is reflected in the relatively low growth rate of 4.75% that year. But the negative impact of the crisis was greatest in 1998, when GDP fell by 13.4%. Quarterly data for 1999 show that GDP continued to fall (by 8.0%) in the first quarter, but grew 3.1% in the second quarter, the first growth since the end of 1997. 7 Table 1: Gross Domestic Product of Indonesia by Source, 1994-99, Constant Prices Gross Domestic Product GDP Growth Rate (%) (Constant 1993 Billion Rupiah) Item 1994 1995 1996 1997 1998 1999* 1994 1995 1996 1997 1998 1999* 1. Agriculture 59291 61885 63742 64289 64987 67929 0.56 4.38 3.00 0.86 1.09 4.53 (a) Food 31407 32951 33647 32752 33310 36336 -2.14 4.92 2.11 -2.66 1.70 9.08 (b) Estate crops 9471 9912 10330 10483 10786 10964 5.07 4.66 4.22 1.48 2.89 1.65 (c) Livestock 6451 6789 7132 7483 6953 6897 4.01 5.24 5.05 4.92 -7.08 -0.80 (d) Forestry 6300 6303 6384 6960 7056 6808 0.53 0.05 1.29 9.02 1.38 -3.51 (e) Fisheries 5659 5928 6248 6610 6880 6919 5.11 4.75 5.40 5.79 4.08 0.56 2. Mining and quarrying 33261 35502 37568 38385 37353 36759 5.60 6.74 5.82 2.17 -2.69 -1.59 (a) Oil and Natural Gas 23719 23719 24062 23919 23417 22236 2.59 0.00 1.45 -0.59 -2.10 -5.04 (b) Mining 4506 6097 7096 7645 9659 10240 13.62 35.31 16.39 7.74 26.34 6.02 (c) Quarrying 5035 5684 6408 6820 4276 4280 14.20 12.89 12.74 6.43 -37.30 0.09 3. Industry 82649 91637 102259 108828 94847 95040 12.36 10.87 11.59 6.42 -12.85 0.20 (a) Oil and Gas 10268 9782 10863 10650 10817 11351 4.85 -4.73 11.05 -1.96 1.57 4.93 (b) Non-oil 72380 81854 91395 98178 84030 83687 13.52 13.09 11.66 7.42 -14.41 -0.41 4. Electricity, Gas & Water 3702 4291 4840 5498 5581 5923 12.52 15.91 12.79 13.60 1.51 6.12 (a) Electric 3042 3519 3984 4464 4584 18655 12.25 15.68 13.21 12.05 2.69 306.95 (b) Gas 133 181 215 269 225 208 25.47 36.09 18.78 25.12 -16.36 -7.56 (c) Water 526 591 640 764 772 800 11.44 12.36 8.29 19.38 1.05 3.63 5. Construction 25857 29197 32923 35040 21035 21025 14.86 12.92 12.76 6.43 -39.97 -0.05 6. Trade 59504 64230 69372 73503 60253 58493 7.61 7.94 8.01 5.95 -18.03 -2.92 7. Transport 25188 27328 29701 32169 26975 26124 8.34 8.50 8.68 8.31 -16.15 -3.15 8. Financial 30901 34313 37400 38730 28278 25551 10.18 11.04 9.00 3.56 -26.99 -9.64 9. Other services 34285 35405 36610 37649 36739 37651 2.77 3.27 3.40 2.84 -2.42 2.48 Gross Domestic Product 354640 383792 414418 434095 376051 374507 7.54% 8.22% 7.98% 4.75% -13.37% -0.41% * Annual estimate based on QI-QIII, 1999 Source: Central Bureau of Statistics (BPS) 8 Almost without exception, each sector of the economy experienced negative growth rates in 1998. Among the sectors contracting most sharply, construction output fell 40%, financial services fell 27%, and trade fell 18%. Non-oil industrial production, responsible for 22.6% of GDP in 1997, fell 14.4% in 1998. Agriculture, on the other hand, enjoyed growth of 1.1% in 1998. This is well-below the growth rates of 3-4% agriculture experienced in 1995-96, but it represents a small recovery from a growth rate of only 0.9% in 1997 during the drought caused by El Niño. Within agriculture, food crop production fell 2.7% in 1997 because of the drought. The sector recovered somewhat in 1998 with a growth rate of 1.7%. This growth rate is still low and probably reflects both lingering drought effects and production problems associated with the crisis. However, the net effects of the crisis for food crop output appear to be positive, since annual growth of the sector in 1999 based on the first 3 quarters was due to be 9.1%. We will show below that this is largely due to an improvement in farmers' terms of trade as a result of the depreciation of the rupiah. Estate crops experienced a similar dip in growth of production (though still positive) in 1997, but did not show such strong recovery in 1999. Again, part of the answer to this quandary comes from less beneficial changes in terms of trade changes as falling world prices ate up some of the gains from exchange rate depreciation. Some of the strongest evidence of crisis effects on rural households in Table 1 can be found in the sharp fall in livestock output in 1998 (7.1%). It is likely that there are two sources for this decline: (i) a reduction in demand for meat and other animal products as households substituted in favor of cheaper sources of calories and (ii) sales of livestock assets as a means of maintaining consumption in the face of declining incomes. Using the 1995/1999 PATANAS data, we will consider the role of livestock as an instrument of savings for rural households below. The transmission of the crisis into the labor market was evident more in an increase in labor market participation than in rising unemployment. This is due in part to the narrow definition of unemployment used and the prevalence of underemployment rather than unemployment in rural areas. Citing data from the national SAKERNAS labor force survey, Feridhanusetyawan (TDRI, 1999) found that the growth rate of the rural labor force increased from 0.5 % from 1990-1996 to 2.9% from 1997-1998. It is unclear whether this growth in rural areas was due more to out-migration from the cities or to an increase in new entrants to the labor market. Much of the increase in the national labor force came from acceleration of growth in the female labor force, from 2.3% per year from 1990-96 to 4.8% between 1997-98. The increase in the unemployment rate in SAKERNAS was small, from 4.7% in 1997 to 5.5% in 1998. However, the national socio-economic survey, SUSENAS, showed a larger increase in unemployment, from 5.0% in 1997 to 6.8% in 1998. The PATANAS Survey Data The PATANAS rural household surveys of 1994-1999 consist of a village census in 1994 and a re-census in 1998 plus household surveys on various topics conducted annually in 1995-1997 and in 1999. The survey design and data collection were done by the Center for 9 Agricultural Socio-Economic Research (CASER) in Bogor, Indonesia.2 The most recent survey round from April-May 1999 was undertaken as a collaboration between CASER and the World Bank in order to study the effects of the economic crisis on rural households. This survey round uses the most comprehensive questionnaire in terms of coverage of topics on household socio-economic status. Table 2 presents the number of households in each survey round, the timing of the survey, and topics covered. Table 2: Sample Composition from PATANAS Census and Survey Rounds Survey Round # 1994 1995 1996 1997 1998 1999 villgs Census Survey Survey Survey Census Survey Period in the field Oct 1994 Oct 1995 Aug/Sept Sept 1997 Aug 1998 Apr/May 1996 1999 Topics HH & income, land & migration, HH & income, farm farm farm labor consump- farm production, character- production, relations tion character- employment, istics employ- istics labor supply, ment consump- tion, savings No. of households Lampung 6 988 301 299 -- 953 279 Central Java 7 1535 356 353 143 1321 326 East Java 6 1387 296 301 149 -- 282 W. Nusa Tenggara 5 827 255 250 -- 751 233 North Sulawesi 6 895 254 250 -- 790 221 South Sulawesi 5 953 296 279 138 882 266 Total 35 6585 1758 1733 430 4697 1607 The households in the PATANAS sample were first identified in the census undertaken in 35 rural villages in 6 provinces in October 1994. CASER researchers chose the provinces of Lampung, Central Java, East Java, West Nusa Tenggara (NTB), North Sulawesi, and South Sulawesi to represent the diverse agroclimates of Indonesia.3 Based on secondary 2CASER first conducted an annual agricultural household survey under the acronym PATANAS from 1984-1988. 3In fact, the 1994 census and the first two rounds of the survey also included 14 villages in the provinces of South Kalimantan and Aceh, plus six additional villages in the other six provinces, for a total of 55 villages in 8 provinces. Because of funding constraints in later survey rounds, two provinces had to be dropped from the sample. South Kalimantan and Aceh were chosen because of unreliable data from the first survey in South Kalimantan and because the political situation made survey work difficult in Aceh. The other six villages dropped were fishing villages. They were considered to be outside the survey's primary focus on agricultural households. Since the data on these villages do not belong to the current panel, they will not be discussed here. 10 information4, villages were chosen within each province to capture the key crops (including livestock and fishing), topographies and cropping patterns in these diverse agroclimates. The census was conducted as a "block census" over a sub-region of each village because funding and time constraints precluded a complete village census. Villages in Indonesia consist of a series of kampung or sub-regions differentiated by physical borders such as a river or road. In the first stage, data were collected on each kampung concerning primary sources of economic activity, crops grown, and topographical characteristics. Researchers then identified an area (or "block") made up of one or more kampung that would provide a census population of close to 200 households per village. The kampung chosen for the census were those considered to be qualitatively representative of the village population based on commodities grown and share of agriculturally-based households. A benefit of this block census approach for socio-economic research is that the households surveyed in each village are likely to have common institutional arrangements in land, labor and credit markets due to their proximity. In the initial census, CASER teams collected information on socio-economic, demographic and farm characteristics for 6585 households in the 6 provinces in our panel. In the first survey, conducted in October 1995, a sample of approximately 50 households was chosen from the roughly 200 census households per village, for a total of 1758 households. The sample of households selected from the census was stratified by landholdings (of 0 ha., 0-0.25 ha., 0.25-0.49 ha., 0.5-1.0 ha., and >1.0 ha.) and farming status (primary income source from agriculture or outside agriculture). The questionnaire included detailed questions on farm production, the sources of household income, and employment during the past 12 months. The 1995 survey serves as the baseline for our analysis of the effects of the economic crisis on rural households. The second round of the PATANAS survey was collected in 1996. The same households interviewed in the 1995 survey were asked about land ownership and titling arrangements as well as the terms of labor contracts and employment over the past year. The 1997 PATANAS survey was collected for a small sub-sample of the original sample because of funding constraints. CASER teams only revisited three provinces (Central Java, East Java and South Sulawesi) and conducted interviews in three villages per province. This survey included an extensive module on rural migration and labor mobility and a simple consumption expenditure module. However, the data from the 1997 survey will not be used in the forthcoming analysis because the sample included only 430 households, the consumption data is rather limited and comparable migration data was not collected in 1999 in the interest of time. In 1998, another census was conducted of PATANAS villages in order to gather current data on household characteristics, asset ownership, and farm characteristics. Along with these annual PATANAS surveys, CASER managed a bi-weekly survey of wages and prices in each of the 35 villages in the Census. This regular monitoring activity was conducted in cooperation with CASER-trained and -appointed representatives in each 4CASER consulted local agricultural extension offices (BPP), who acted as CASER's local partners in the PATANAS surveys. 11 village. It focused on collecting information on wages of key farm and non-farm activities (176 variables) and prices of the main agricultural outputs and inputs (119 variables) from local sources. This activity began in August 1994 and stopped in June 1998 due to CASER project funding problems. CASER resumed collection of the wage and price survey temporarily during April-May 1999 when the latest PATANAS field survey was conducted. The consistency of this price series varies across villages and over time owing to differences between villages in crops grown and factors employed and to failures of local agents to collect the data regularly. Still, this data is sufficiently complete to allow the construction of a time series for wages and prices of a number of important commodities. The 1999 survey was conducted jointly by CASER and the World Bank with funding from the Asia-Europe Meeting (ASEM). The questionnaire was designed to be fairly comprehensive, capturing in some manner nearly all of the topics addressed in the previous three survey rounds. The survey instrument included sections on household member characteristics, land ownership and use, and agricultural production and input use including a detailed accounting of labor contracts, employment, individual household member labor supply, consumption expenditure, savings and credit. In designing the questionnaire, care was taken to elicit responses that were comparable to those from previous rounds. New topics covered in the 1999 questionnaire included individual labor supply data and more complete questions on the terms of labor contracts both on- and off-farm. The 1999 PATANAS survey also included a village module through which village officials provided information on village infrastructure, health, education, and banking. Of the 1758 households interviewed in the 1995 survey, 1607 of these were re- interviewed in 1999 (an attrition rate of only 8.6 % in four years). However, after a careful analysis of the data for both the 1995 and 1999 survey rounds, the final 1995-99 panel contains 1494 households that could be reliably matched between the two years (an effective attrition rate of 15%). At the same time that the formal survey was being conducted, a two-person team of Indonesian sociologists visited a number of the 1998/99 PATANAS sample villages and neighboring "control" villages for the purpose of examining how rural households are coping with the ongoing economic crisis. The sociologists focused on factors that may not have been fully accounted for by the 1999 PATANAS questionnaire. These include crisis-induced outcomes at the household and village levels such as changes in: land tenure arrangements, borrowing and lending behaviors, rural household attitudes toward asset management and risk, inter- and intra-household relationships, incidence of illicit activities, formal and informal institutional arrangements, communal activities, maintenance/upgrading of village infrastructure and social services, and so on. This informal data gathering effort, based on village immersion, provides important supporting information for the quantitative results from the main survey. 12 II. Identifying Vulnerable Households Vulnerable households are those that are susceptible to a greater than average change in socio-economic status due to the economic crisis. In this section, we identify households that appear vulnerable based on their characteristics in 1995, before the crisis, and then measure the change in their socio-economic status after the crisis. The consideration of ex ante vulnerability of households and their ex post crisis impacts will focus on households that experience a greater than average change in socio-economic status.5 It is also appropriate, however, to give attention to the vulnerability of the poor, since a negative change in their socio-economic status can threaten their subsistence requirements and because they are often least able to recover from such shocks. Thus, the focus here is on which households experienced the greatest crisis impacts, but with additional consideration for crisis effects on poorer households. In their analysis of vulnerability to macroeconomic shocks, Glewwe and Hall (1995) recognize three stages to the effects of an economic crisis: the household experiences an economic shock, it adapts, and the government undertakes expenditures to protect negatively affected households. Each of these stages of crisis effects will be considered below. The primary measure of household socio-economic status used throughout this paper is a measure of household income or income from crops since these are the best indicators for which complete data is available both before and after the onset of the economic crisis.6 While household expenditure is often considered to be a preferred measure of household living standards for micro data sets, complete household expenditure data from before the crisis was only collected for a small sub-sample of PATANAS households in 1997. 5Glewwe and Hall (1995) define vulnerable households as those that experience a greater than average change in socio-economic status (ex post) due to a crisis. We use the term vulnerability to mean ex ante susceptibility to such a change in socio-economic status. Using our definition, a useful exercise for policy purposes is to identify those households that are considered vulnerable and determine whether these households indeed experienced greater than average changes in socio-economic status. 6A welfare interpretation to the observed changes in household income is intentionally avoided because these shocks do not capture the full effect of the crisis on utility of household members. The purpose of this paper is to identify the effects of the various components of the crisis on households. Since a complete welfare analysis of the effects of such a far-reaching economic crisis is not feasible, a reliance on summary measures of household socio-economic status was deemed more appropriate. 13 Table 3: Household Summary Statistics by 1995 Per Capita Income Quintiles 1995 1999 1995 PC Income Quintiles 1 2 3 4 5 All 1 2 3 4 5 All Median HH Characteristics Household size (mean) 4.78 4.64 4.39 4.26 3.84 4.38 4.87 4.81 4.52 4.59 4.21 4.60 Household head age 41 41 43 45 47 43 46 45 48 48 50 47 Household head education 3 3 3 4 4 4 4 4 4 5 6 5 Land owned per capita, Ha 0.08 0.10 0.11 0.13 0.16 0.10 0.09 0.12 0.11 0.14 0.18 0.12 Land planted per capita, Ha 0.08 0.10 0.12 0.13 0.15 0.11 0.14 0.14 0.13 0.14 0.22 0.15 Intensity of land planted 0.10 0.14 0.15 0.19 0.25 0.15 0.18 0.20 0.19 0.24 0.33 0.22 per capita, Ha Quintile Summary Statistics Number of households 299 299 299 299 298 1494 299 299 299 299 298 1494 Share of female headed HHs 4.0 5.4 4.7 6.0 10.4 6.1 5.8 5.6 4.5 6.3 9.6 6.4 Share households owning no 20.1 23.4 21.1 17.1 20.1 20.4 16.1 15.7 17.7 13.4 17.1 16.0 land, % 14 Households that may be vulnerable to either negative or positive shocks to income from the economic crisis can be identified based on household and farm characteristics, ownership of land and other assets, sources of income, and commodity portfolio.7 For each of these indicators, we review summary statistics in Tables 3-6 to identify the potential vulnerability of households. We then undertake a formal test of the effects of the crisis on household socio-economic status based on these measures. The test of the correlation of socio-economic indicators to change in household socio-economic status is made by regressing change in household income against the relevant household characteristic plus a constant term to remove average change in socio-economic status. The t-statistics from these regressions identify whether the particular characteristic is associated with an improvement (positive t-statistic) or decline (negative t-statistic) in socio-economic status. This provides an ex post measure of crisis effects that indicates whether those households identified as potentially vulnerable were, in fact, adversely affected.8 In addition, we use this approach to test many of the hypotheses outlined in the previous section. Table 7 presents t-statistics for a variety of household and farm characteristics. Household socio-economic status is measured by percentage change in household income per capita in column 1 and by change in logarithm of household income per capita in column 2. The latter measure gives greater weight to improvements in income of the poor. Table 3 presents summary statistics on household characteristics for the entire sample from 1995 and from 1999 (for comparison) by 1995 per capita income quintiles. Average household size, head age, head education, farm size, planted area, and cropping intensity are presented for the entire sample and by 1995 per capita income quintile. Household size includes household members that reside at the location of the interview. Land planted per capita is total area planted during the year (normalized by household size), ignoring the number of times crops were planted on this land. The intensity of planted area adds up planted area for each time that crops were planted on any plot. In general, household heads of poorer households are younger and less educated than their richer counterparts. One can interpret their relative lack of experience (or tenure) and education as determinants of their poverty. However, age of the primary decision-maker may be associated with greater vulnerability to income shocks if older people are less able to adapt. More household head education suggests greater skills, but also a greater likelihood in working off-farm in a crisis-effected sector. Notice that median farm area under the farmer's control (land planted) increased after the start of the crisis. Cropping intensity also rose, though this is driven, in part, by increased land planted. 7Although the concept of vulnerability encompasses both negative and positive larger than average changes in socio-economic status, households that endured large negative shocks are of greater interest. Therefore, we will often refer to these households as "more vulnerable", implicitly assigning the negative connotation to vulnerability. References to vulnerability to positive shocks will be explicit. 8Glewwe and Hall (1995) interpret t-statistics from similar regressions of individual household characteristics on change in per capita consumption as identifying whether that characteristic is correlated with vulnerability. We interpret the dependent variable as simply a measure of change in socio-economic status following the crisis (ex post), since we consider vulnerability to be a condition of the household ex ante. 15 Tests of the correlation of these variables with actual change in socio-economic status during the crisis are presented in t-statistics in Table 7. Results show that larger households experienced greater improvements in socio-economic status, probably because of greater income earning opportunities. In the log regression, households with older household heads had lower growth in per capita income. Roughly 20% of the sample was landless in 1995. Over the entire sample, median farm size per capita in 1995 was 0.1 hectare, which translates into a median household holding of roughly one half hectare. Not surprisingly, poorer households have smaller holdings. Average farm size in Table 3 increases monotonically with 1995 per capita income quintile and average holdings for the top quintile are twice as large as for the bottom quintile. The poor are also somewhat more likely to be landless in 1995 than wealthier households. Regression results from Table 7 do not show any aggregate effect of size of landholdings on income growth. However, the hypothesis that landless households were more adversely affected by the crisis is supported in the log specification. Despite the lack of statistical correlation between size of landholdings and change in socio-economic status, landholdings may be important in some regions and merit further consideration. Of course, the importance of land ownership and size of holdings as an indicator of wealth or coping ability varies by province. The population pressure on land on Java is far greater than in the outer islands, for example, with Java having among the highest rural population density in Asia. Table 4 shows the share of landless and median landholdings by province for the 1995 PATANAS sample. In order to focus on households whose income-generating activities are generally agriculturally based, 189 households that were engaged in business and had no crop income in at least one of the two survey rounds were removed from our sample. We will refer to this sub-sample as the "non-business" sample. On Java, 32.3% of all households in the sample owned no land in 1995. In other regions, this share was much lower. Median farm size per capita for landholding households was only 0.08 hectares on Java, while farm sizes are twice as big in South Sulawesi and roughly three times bigger in Lampung, NTB and North Sulawesi. However, these landholdings comparisons ignore regional differences in land fertility due to differences in quality and irrigation usage. Land fertility varies considerably across provinces in the PATANAS sample. In order to account for these fertility differences, we use median 1995 rice yields by province to re-scale farm size into effective fertility-adjusted landholdings. Median rice yields are shown in Column 7 of Table 4. Using these figures, a fertility index is created by dividing province-specific rice yields by rice yields in Central Java, the province with the highest yields. Household data on per capita landholdings are then multiplied by the appropriate province fertility weight to obtain fertility-adjusted effective landholdings. For example, since rice yields in NTB are only 68% of yields in Central Java, a farmer in NTB owning one hectare of land per capita would have the same effective landholdings as a farmer in Central Java with 0.68 hectares per capita. After adjusting landholdings for land fertility, much of the difference in median farm size between Java and the outer islands is removed. 16 Table 4: Regional Distribution of Land Ownership and Landholdings, Non- Business Sample Non-business Landless Distribution of Landholdings Per Capita Households Households for Households Owning Land, Ha Province N N Share of N Median Median Median HHs with Area Per Rice Yield, Effective Area no land (%) Capita, Ha Ton/Ha Per Capita Lampung 260 22 8.5 238 0.251 1.690 0.093 C. Java 249 70 28.1 179 0.079 4.545 0.079 E. Java 169 65 38.5 104 0.075 4.472 0.073 NTB 198 15 7.6 183 0.210 3.100 0.143 N. Sulawesi 173 13 7.5 159 0.225 2.240 0.111 S. Sulawesi 256 2 0.8 254 0.169 2.810 0.105 Total 1,305 187 14.3 1,117 0.164 3.000 0.099 Also of note from Table 3 are the figures on share of households headed by a female. While female headship is relatively rare in Indonesia, it is an indicator of potential vulnerability because these households typically have fewer prime-age income earners and because, on average, men earn higher wages than women in Indonesia. Thus, it is remarkable that more female headed households are found in the top two per capita income quintiles in 1995 than in the bottom three quintiles. However, the probability that a household has a female head increased by 1999 for households that were in the bottom 1995 per capita income quintile and decreased for households in the top 1995 quintile. Poorest households were more likely to loose a male head and the richest were more likely to gain one. Regression results from Table 7 are consistent with these figures; in the log specification, female-headship is negatively correlated with per capita income growth. The bivariate regressions in Table 7 also provide strong support for the hypothesis that households in Central Java and East Java were more vulnerable to adverse shocks. In the coming sections, we consider the sources of adverse shocks to household income on Java. Vulnerability to crisis effects depends critically on the sectoral sources of household incomes. In order to analyze how sources of income vary across households with different welfare levels, the sample is disaggregated by per capita landholdings quintiles (adjusted for land fertility) in 1995. We use per capita landholdings to differentiate welfare levels of households because land area is more accurately measured than income in micro data and because land represents the most significant household asset in the sample. As such, it is the best indicator of household wealth. 17 Table 5 presents mean income and share of income by source for fertility-adjusted per capita landholdings quintiles in the "non-business" sample in 1995.9 Agriculture is by far the most important source of income for PATANAS households with 63.6% of 1995 income coming from crop production or agricultural labor on average. The share of income from agricultural production is monotonically increasing in landholdings quintiles and the labor share of income is monotonically decreasing in landholdings quintiles. Although the share of income from agricultural production is small (13.4%) for the lowest landholdings quintile, the second quintile has as much of 45.2% of income from farming. This suggests that while farm income has a smaller share for the poor, its contribution to household income is not negligible.10 Tests of the hypothesis that agriculture acted as a safe haven during the crisis yielded mixed results. In Table 7, the 1089 households whose head was primarily engaged in crop production or agricultural labor experienced a smaller percentage change in per capita income than other households (P-value = 0.071), but enjoyed a relatively larger change in the logarithm of per capita income (P-value = 0.007). One interpretation of these results is that poorer agricultural households were less vulnerable than richer ones, since the logarithm measure of income change places heavier emphasis on the poor. As a second test of this hypothesis, the share of income earned from crop production or agricultural labor in 1995 was used as a broader measure of reliance on agriculture. Here, agriculture was associated with higher income growth regardless of the measure of income change used. One set of potentially vulnerable groups includes those who earn their income as laborers or through a business in a sector that experienced a deep contraction during the crisis. In 1995, 10.1% of household heads identified their main work activity as agricultural labor and 8.2% claimed non-agricultural laborer. In general, laborers will be vulnerable to declining income due to falling real wages and low job security. However, agricultural workers may have benefited from increased demand for labor and possibly higher wages in regions producing booming tradables crops such as coffee. The test of our initial hypothesis that wage laborers were more adversely affected by the crisis is supported for agricultural workers (in the log-linear specification), but not for non-agricultural workers in Table 7. There were 5.5% of household heads that claimed their primary work activity was in a non-agriculturally related business. While those with businesses in industry or construction may be vulnerable, Table 7 shows no significant relationship between business ownership and growth in income. 9Means are used as a measure of central tendency rather than medians in the presentation of income data by source so that the components of income sum to the total and shares are well defined. The first landholdings quintile in the "non-business" sample includes 188 landless households. Among these are hired farm managers and salaried professionals, which accounts for higher median and mean per capita income in the first landholdings quintile than in the second quintile. 10Sources of "other income" include gathering materials, equipment rental, land rental, pensions, etc. 18 Table 5: Average Income and Share of Income by Source for 1995 Effective Per Capita Landholdings Quintiles, Non-business Sample Quintile 1 2 3 4 5 All Average Landholdings PC 0.00 0.06 0.14 0.25 0.66 0.22 N 261 261 261 261 261 1305 Median PC Income 274.8 241.2 279.7 380.6 493.0 320.0 Income Source Mean Per Capita Income in 1995 by Source, Rp. '000 1. Agriculture 52.0 173.0 255.1 367.9 520.5 273.7 Crops 47.7 164.3 254.0 367.9 525.6 271.9 Livestock/Poultry 4.3 8.8 1.1 -0.1 -4.4 1.9 Fishponds 0.0 0.0 0.0 0.0 -0.7 -0.1 2. Labor income 206.3 79.0 62.4 60.6 16.3 84.9 Agricultural 114.9 43.0 36.2 34.3 6.7 47.0 Non-agricultural 91.3 36.0 26.3 26.3 9.6 37.9 3. Business income 34.6 48.1 45.8 78.1 80.0 57.3 Industry 1.5 11.5 3.8 6.8 18.4 8.4 Trade 20.9 26.2 28.0 51.6 38.2 33.0 Services & others 12.2 10.4 14.0 19.6 23.4 15.9 4. Transfers 2.2 4.1 2.4 4.8 14.8 5.7 5. Other income 93.3 78.7 39.8 57.7 140.9 82.1 TOTAL 388.3 382.9 405.6 569.1 772.5 503.7 Income Source Share of 1995 Mean Per Capita Income by Source, % 1. Agriculture 13.4 45.2 62.9 64.6 67.4 54.3 Crops 12.3 42.9 62.6 64.7 68.0 54.0 Livestock/Poultry 1.1 2.3 0.3 0.0 -0.6 0.4 Fishponds 0.0 0.0 0.0 0.0 -0.1 0.0 2. Labor income 53.1 20.6 15.4 10.6 2.1 16.9 Agricultural 29.6 11.2 8.9 6.0 0.9 9.3 Non-agricultural 23.5 9.4 6.5 4.6 1.2 7.5 3. Business income 8.9 12.6 11.3 13.7 10.4 11.4 Industry 0.4 3.0 0.9 1.2 2.4 1.7 Trade 5.4 6.8 6.9 9.1 4.9 6.5 Services & others 3.1 2.7 3.4 3.4 3.0 3.2 4. Transfers 0.6 1.1 0.6 0.9 1.9 1.1 5. Other income 24.0 20.6 9.8 10.1 18.2 16.3 TOTAL 100.0 100.0 100.0 100.0 100.0 100.0 19 Table 6: Distribution of Farmers by Major Commodity Group--Number and Share of Farmers of Each Commodity Group by Province and by Landholdings Quintile, 1995 Rice Dryland Tobacco & Tree Others* Total Crops Sugarcane Crops Province Lampung N 70 57 38 81 12 258 % 27.13 22.09 14.73 31.40 4.65 100.00 C. Java 34 52 78 13 38 215 15.81 24.19 36.28 6.05 17.67 100.00 E. Java 26 33 5 19 46 129 20.16 25.58 3.88 14.73 35.66 100.00 NTB 117 36 4 19 15 191 61.26 18.85 2.09 9.95 7.85 100.00 N. Sulawesi 37 46 0 57 27 167 22.16 27.54 0.00 34.13 16.17 100.00 S. Sulawesi 78 72 1 100 5 256 30.47 28.12 0.39 39.06 1.95 100.00 1995 Effective PC Landholdings Quintiles 1 36 29 11 26 72 174 20.69 16.67 6.32 14.94 41.38 100.00 2 87 72 33 46 22 260 33.46 27.69 12.69 17.69 8.46 100.00 3 78 75 30 64 14 261 29.89 28.74 11.49 24.52 5.36 100.00 4 78 70 35 64 14 261 29.89 26.82 13.41 24.52 5.36 100.00 5 83 50 17 89 21 260 31.92 19.23 6.54 34.23 8.08 100.00 Total 362 296 126 289 143 1216 29.77 24.34 10.36 23.77 11.76 100.00 * The "Others" category includes those who did not have more than 50% of their revenue from within one commodity group in 1995 plus any farmer who did not grow crops in 1995, but did grow crops in 1999. Given the predominance of farming households, one of the most important determinants of crisis-induced changes in household income is the choice of crops grown. A useful characterization for this purpose is to distinguish farming households by major commodity grown. We have separated the 1113 households with some crop production in the 1995 non- business PATANAS sample into those that primarily grow rice, dryland crops, tobacco and sugarcane, or tree crops. Households were assigned to a commodity group if more than 50% of the value of production in 1995 derived from commodities in that group. The distribution of farmers in each major commodity group in 1995 by province and by landholdings quintile is presented in Table 6. 20 These commodity groupings are informative. Rice is the primary staple food crop in Indonesia and is grown by more farmers than any other crop. Rice farmers could have lower vulnerability to crisis effects because of rising rice prices during this period. Dryland crops include all other grains and vegetables. The dryland crops with the greatest total value of production in 1995 are (in order) potatoes, maize, shallots, cabbage, cassava, garlic, and soybeans. They are grouped together in part because there is a high degree of substitutability in production between many of these crops. Some of these crops show limited responsiveness to world prices, which makes farmers of those crops relatively more vulnerable. Dryland crop farmers are mostly found in the middle three landholdings quintiles. Tobacco and sugar cane are grouped because they represent the two largest cash crops that are effectively non-tradable in Indonesia. As such, the terms of trade are expected to shift against these crops following the exchange rate depreciation. The majority of the tree crops are tradable and many experienced a large jump in price after the onset of the crisis. The tree crops with the greatest value of production in 1995 were coffee, cocao, pepper, coconut, elephant grass, vanilla and cloves. Tree crop farmers are concentrated in the higher landholdings quintiles. Table 7 lists the number and share of farmers in each commodity group by province in 1995. Table 7: T-statistics on Correlation of Household and Farm Characteristics with Income Vulnerability Vulnerability Household and Farm % Change in PC Change in Ln of PC Characteristics Household Income Household Income N obs* t stat N obs* t stat Household size (mean) 1444 1.807 1432 3.756 HH head age 1441 -0.161 1429 -1.684 HH head education 1428 -0.782 1416 -0.835 Land owned per capita, Ha 1444 -0.774 1432 0.009 Landless household 1444 -0.896 1432 -2.892 Female household head 1444 -0.617 1432 -2.258 Household resides on Java 1444 -1.958 1432 -6.689 HH head main job: crop farmer or ag. laborer 1444 -1.807 1432 2.682 1995 income share from crops or ag. labor 1444 10.617 1432 8.443 HH head is agricultural laborer 1444 -1.204 1432 -3.318 HH head is non-agricultural laborer 1444 1.597 1432 -1.251 HH head owns non-agric. business 1444 0.262 1432 -0.899 >50% of crop revenue from rice 1042 -1.002 1032 1.020 >50% of crop revenue from dryland crops 1042 0.751 1032 -1.849 >50% of crop rev. from tobacco & sugarcane 1042 -1.141 1032 -5.255 >50% of crop revenue from tree crops 1042 1.233 1032 4.847 *Households with negative income were omitted from regressions because percentage change in income levels is not well defined and the logarithmis undefined. Additional observations were dropped if data for the regressor was missing. 21 Results from Table 7 show that the change in income for rice farmers was not significantly different than the average over the entire sample. This failure to separate crisis effects of rice- growing households from other households may be due to the large number of rice farmers in the sample and the diversity of their other income sources. Not surprisingly, there is evidence for the hypothesis that producers of tradable (non-tradable) crops experienced an income boom and that non-tradable producers had lower income growth. In the log specification, farmers of dryland crops and tobacco & sugar experienced larger than average declines in per capita income and tree crop farming is associated with larger than average income growth. 22 III. The Effects of the Crisis on Rural Households The foregoing analysis identified female headed households, agricultural laborers and tobacco farmers among the groups that experienced lower than average growth of per capita income as a result of the economic crisis. This does not explain why households with these characteristics were at risk or how large were their relative losses. In this section, we review the extent of the economic crisis, the manner and magnitude of the crisis' effects on household welfare, and coping strategies used by households to take advantage of crisis-induced opportunities and avoid crisis-related losses. We begin this exercise by presenting summary statistics on changes in income between the 1995 and 1999 survey rounds. This enables identification of gainers and losers from the crisis and of the magnitude of the shocks. Evidence on Shocks to Household Income and Sources Table 8 presents the change in median real per capita household income from 1995 to 1999, by 1995 effective per capita landholdings quintiles. Income in both years is at 1994/95 prices for the months covered by the 1995 survey (October-September). The deflator used is the provincial rural consumer price index (IB Index) from the Farmers' Terms of Trade data collected by the Central Bureau of Statistics. These are the rural CPI series presented in Figure 2.11 The income changes by landholdings quintiles in Table 8 are also presented with per capita income re- weighted to account for the fact that the size of the sample in each province does not reflect the true provincial share of the rural population. Households on Java are under-represented in the PATANAS sample, for example. Using data from the 1990 Population Census, the observations on per capita income are weighted by the actual province share in the rural population divided by the sample share. The corresponding income measure is more representative of the rural population of the six provinces. Table 8 provides some interesting results, subject to a few caveats concerning measurement issues. The first result is that the PATANAS sample is made up of reasonably poor households. With an average exchange rate of Rp. 2220/$US for October 1994 ­ September 1995, median per capita income during that period was $US 151 per year. Table 8 also shows that average income rose between 1995 and 1999 for all landholdings quintiles. However, the distribution of the income gains generally favors households with larger landholdings. We cannot conclude directly from this that the poorest households experienced the smallest growth in income since median incomes of the lowest landholdings quintile are higher than for quintiles 2 and 3. Comparing individual quintiles results in an inconsistency in welfare ranking between farm size and median income. However, grouping the bottom two or three quintiles removes this inconsistency and shows that the income gains of relatively poor households are much smaller than for the two largest landholdings quintiles. 11The FTT IB index was chosen as a deflator because it more reliably captures changes in rural prices than the national CPI series, for example, which is based on a sample from urban areas, where price increases were generally smaller (see Figure 2). The FTT IB Index provides separate inflation measures for each province, but does not control for variation in prices within provinces (which may be present in provinces with remote villages, such as South Sulawesi and NTB). 23 Table 8: Change in Median Real Per Capita Income by 1995 Effective Per Capita Landholdings Quintiles, Rp `000 Unweighted Census weighted 1995 Effective PC N 1995 1999 % Change 1995 1999 % Change Landholdings Quintiles12 1995-99 1995-99 1 305 312 381 22.3 341 357 4.5 2 293 278 440 58.6 304 447 47.2 3 299 281 479 70.5 306 455 48.7 4 299 360 595 65.5 379 579 52.6 5 298 482 759 57.5 577 667 15.6 All households 1494 332 505 52.3 360 439 21.9 In the weighted income figures, the gains in socio-economic status for the smallest and largest landholdings quintiles are much smaller than in the unweighted sample. Still the income gains of the bottom two landholdings quintiles are well below those of the top two quintiles. We conclude from these figures that absolute income gains for the rural poor in the six provinces were small, and that in relative terms the poor faired worse than the rich during the crisis. A measurement problem associated with the income figures concerns the method of deflating income between the survey rounds. The income figures presented in Table 8 were deflated by the average annual price change from 1995 to 1999. This was the most consistent method available for controlling for inflation since non-farm income was not provided by season. However, prices grew by 68% on average during the 12 months covered by the 1999 survey (April 1998 ­ March 1999). Since the largest portion of agricultural revenues are earned at the end of the rainy season in March, deflating by average annual prices overstates real incomes in 1999.13 This suggests that the growth of income between the two periods may be somewhat overstated for the PATANAS sample. In addition, it is useful to keep in mind that the baseline data from 1995 were collected more than one-and-a-half years before the beginning of the economic crisis. Taking into account the lower growth rates just mentioned and assuming per capita household income grew at the rate of GDP (roughly 8% per year) from late 1995 to mid 1997, median per capita income may have fallen for the lowest landholdings quintile since the start of the crisis. Table 9 highlights the wide disparity in experiences during the crisis for households on Java versus those living in the outer islands. The table shows a small increase in median per capita incomes for Central and East Java. Meanwhile, households in Lampung, NTB, and Sulawesi 12The first landholdings quintile for the full sample of 1494 households is slightly larger than the other quintiles because it contains all households that owned no land in 1995. 13This measurement issue are not likely to affect the qualitative result that poorer households had lower income growth than richer households, since both have comparable shares of combined crop income and agricultural labor income. 24 enjoyed substantial income growth. The median income in NTB, the poorest province in the PATANAS sample, grew enough to close most of the income gap between NTB and East Java. Landless households on Java, which make up 73% of all landless households in the sample, fared particularly badly. Median incomes for this group grew only 4% from 1995 to 1999. Those that were still landless in 1999 had no income growth at all. Table 9: Change in Median Real Per Capita Income by Province, Rp `000 Province N 1995 1999 % Change 1995-99 Lampung 266 286 524 83.2 Cent. Java 301 381 434 13.9 East Java 243 378 396 4.8 NTB 213 197 366 85.8 N. Sulawesi 213 520 882 69.6 S. Sulawesi 258 313 593 89.5 All 1494 332 505 52.3 Table 10 presents the change in median per capita farm income and non-farm income by landholdings quintile for the entire PATANAS sample. Farm income is composed of the value of crop production minus variable costs and land rent plus net profit from livestock and fishponds. Non-farm income derives from labor income, business profits, transfers, and other sources. Farm income increased during the crisis for all landholdings quintiles. In the census- weighted data, the increase in farm income is less pronounced, which reflects the under- representation of Java in the sample. The growth in farm income for the first quintile--which consists of all households that own no land in 1995--indicates the initiation of agricultural production. Out of 305 landless households in the 1995 survey, 115 had acquired some land by the time of the 1999 questionnaire. The land-owning poor in quintile 2 also had a large percentage increase in income from agriculture. This suggests that agricultural production was an important source of income security for the poor during the crisis. Surprisingly, non-farm income grew substantially from 1995 to 1999. While some of this income growth may have occurred before the crisis began, its magnitude is still remarkable given that the sectors that experienced sharp declines in urban areas should have strongest links to the non-farm rural sector. The only landholdings quintile that did not enjoy growth in non-farm income was the lowest quintile. Census weighted data even show a slight fall in non-farm income for the lowest-quintile. This reinforces the importance of the initiation of agricultural production for income growth among those that were landless in 1995. 25 Table 10: Change in Median Per Capita Farm Income and Non-Farm Income, 1995-99 1995 Effective PC Unweighted Census Weighted Landholdings Quintiles 1995 1999 % Change 1995 1999 % Change Per Capita Farm Income 1 0.0 10.5 -- 0.0 6.3 -- 2 64.7 142.5 120.2 48.2 124.5 158.3 3 173.0 214.0 23.7 175.9 177.7 1.0 4 216.3 334.3 54.6 218.4 324.0 48.4 5 324.4 459.2 41.5 410.9 421.1 2.5 All Households 111.5 190.4 70.7 49.5 119.6 141.6 Per Capita Non-Farm Income 1 283.3 278.6 -1.7 311.0 276.7 -11.0 2 120.2 191.1 59.0 166.8 188.3 12.9 3 35.8 155.8 334.7 21.4 132.9 520.4 4 52.5 109.4 108.4 41.7 89.4 114.7 5 10.4 116.8 1027.6 0.0 79.6 -- All Households 102.5 182.6 78.1 160.0 193.4 20.9 The sources of this growth in non-farm income are worth consideration. Table 11 presents the change in mean real per capita non-farm income from labor, business and all other sources. Average labor income nearly doubled, while business income grew more moderately. After census weighting, average business income grew faster (73%) than labor income (48%) because of under-representation of Java. Note that the labor income growth is driven by non-agricultural sources, as agricultural labor income stagnated. In terms of regional distribution of non-farm income sources, business income was the greatest source of non-farm income growth on Java, while off Java labor income growth was more significant. This is consistent with a labor market recession centered in Java and a reallocation of labor to business activities there. Table 11: Change in Mean Sources of Real Per Capita Non-Farm Income Unweighted Census Weighted 1995 1999 % Change 1995 1999 % Change Labor 87.2 170.6 95.6 115.4 170.7 48.0 Business 120.9 175.4 45.1 130.3 225.1 72.7 All other 87.9 60.0 -31.8 80.8 58.0 -28.2 Java Off Java 1995 1999 % Change 1995 1999 % Change Labor 130.0 171.1 31.7 62.8 170.3 171.4 Business 146.4 247.3 69.0 106.3 134.2 26.3 All other 84.6 55.1 -34.8 89.8 62.8 -30.1 26 The combined effect of these changes in farm and non-farm income was a general, but small, decline in the importance of farm income. The share of income from farming in 1995 was generally increasing by landholdings quintile: 0% for Q1, 44% for Q2, 92% for Q3, 82% for Q4, and 99% for Q5. However, after the crisis, farm income shares fell for all but the lowest landholdings quintile. The 1999 farm income shares were 4%, 42%, 56%, 75% and 81%, respectively. The increase in non-farm income share is not so significant insofar as baseline levels were very low. Of course, the preceding summary statistics mask some of the effects of the crisis on individual households. Per capita household income fell from 1995 to 1999 for nearly 35% of all households and for 43% of landless households. Central and East Java had the highest share of households with declining income at roughly 45%, while Lampung and South Sulawesi had the lowest share with only 25%. These results contradict the hypothesis that contraction of the economy in urban areas and slumping urban labor markets would cause a depression of rural non-farm income and leave agriculture as the only source of income maintenance in rural areas. Agriculture remained an important source of income growth during the crisis. However, non-farm income sources grew significantly among those with traditionally little non-farm income. Only the lowest landholdings quintile did not benefit from non-farm income growth. Finally, it is useful to identify the individual effects of various household and farm characteristics on household socio-economic status while controlling for other contributing factors. For this purpose, a multivariate regression explaining change in per capita income was estimated. Regression results are presented in Table 12. Because the denominator in the share of income from crops--base base year household income--also appears (in log linear form) in the construction of the dependent variable, we use an instrumental variables estimator. The instruments for initial income share from crops are land area planted, a dummy variable for whether the household head is a farmer, and dummy variables for primary commodity group as rice, dryland crops, tobacco and sugar, or tree crops. The most important result for our purposes is that initial income share for crops is a significant determinant of the growth of household income. This result will be explored in greater detail below, as we consider changes in production patterns and input use during the crisis. Table 12 also identifies business-owning households as having higher income growth, and household head age is negatively correlated with income growth. A preliminary step in determining the causes of the changes in household income identified here and in the previous section is examination of changes in prices and wages during the crisis. We now turn to this issue. 27 Table 12: Instrumental Variables Estimates of Income Growth Dependent Variable: Change in Log Per Capita Income, 1995-99 Coef. t-stat Per capita land area owned -0.207 -1.315 Household size -0.016 -0.502 HH head education 0.014 0.794 HH head age -0.006 -1.764 * 1 if female household head -0.110 -0.585 Predicted share of income from crops° 1.014 3.828 ** HH head is agricultural laborer 0.245 1.087 HH head is non-agricultural laborer 0.304 1.424 HH head is a business owner 0.428 1.773 * Constant 0.222 0.890 N 1391 ** Significant at 5% level. * Significant at 10% level. ° Instruments for initial crop share are land area planted, dummy if HH head is a farmer, and dummies for primary commodity group (rice, dryland crop, tobacco and sugar, tree crop). Price and Wage Changes During the Crisis The effect of the crisis on agricultural households was determined in large part by higher producer prices for some crops and for tradable inputs like fertilizer. Poorer non-agricultural households suffered other effects, but were hurt by higher prices for staple foods and by a decline in real wages. Figure 4 presents real consumer and producer prices for rice by province taken from the Farmers Terms of Trade data collected by BPS. The rice consumer price index is deflated by the rural provincial CPI and the rice producer price index by the rural provincial PPI from Figure 2. Data from the Wage and Price Survey that was collected in conjunction with the PATANAS household survey show that, in nominal terms, the consumer price of local rice rose from an average of Rp. 900 per kilogram in June, 1997 to Rp. 2,000 per kilogram by June,1998. In real terms, average rice prices from the FTT survey increased between 7% and 22% from the second half of 1997 to the second half of 1998, with the greatest increases in South Sulawesi, Central Java, and NTB. 28 Figure 4: Real Consumer and Producer Prices for Rice by Province Rice Consumer Price Deflated by Rural CPI, Rice Consumer Price Deflated by Rural CPI, 3 Mo. Moving Avg. 3 Mo. Moving Avg. 200 200 180 180 160 160 140 140 120 120 1994=100 1994=100 100 100 80 80 60 60 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Jun-99 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Jun-99 Lampung C. Java E. Java NTB N. Sulawesi S. Sulawesi Rice Producer Price Deflated by Rural PPI, Rice Producer Price Deflated by Rural PPI, 3 Mo. Moving Avg. 3 Mo. Moving Avg. 200 200 180 180 160 160 140 140 120 1994=100 120 1994=100 100 100 80 80 60 60 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Jun-99 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Lampung C. Java E. Java NTB N. Sulawesi S. Sulawesi 29 In November 1998, the government implemented a large package of reforms designed to liberalize food marketing (see Tabor, Dillon, and Sawit, 1999). The reforms included substantial restrictions on the operations of the BULOG marketing board, which had until then closely managed prices of rice and a few other commodities. Also included in the reforms was the elimination of a large fertilizer subsidy and additional expenditures for increasing subsidized credit programs to farmers (called KUT) and for funding public works programs and a subsidized rice program intended for poor households. Because these reforms were motivated by substantial price increases in mid-1998, we will consider them part of the government's response to the crisis and will include them as a component of household crisis vulnerability. The liberalization of rice marketing did not have a significant effect on rice prices until late 1999, after the most recent round of the PATANAS survey was completed. As a result, the effect of the policy change on household incomes through rice prices cannot be determined with the PATANAS data. Producer price data from the PATANAS Wage and Price Survey show nominal rice prices rising from Rp. 405 per kilogram to Rp. 970 per kilogram from June 1997 to June 1998. Using the real rice producer price index calculated from FTT data, farmgate prices were unchanged in South Sulawesi, rose 21% in Lamung, and rose 6-12% in the other four provinces from the second half of 1997 until the second half of 1998. A central component of the Indonesian government's agricultural policy throughout the 1990s was a large subsidy on fertilizers. The government maintained the fertilizer subsidies until the food marketing reforms of November 1998 when the subsidy was eliminated. Once the subsidies were removed there was an immediate and large increase in the cost of fertilizer. Figure 5 presents nominal price indices for urea and TSP, the two most commonly used fertilizers in the 1995 PATANAS data. These indices were constructed as the simple average of the provincial series from the FTT surveys. From the third quarter of 1997 until the same period in 1998, urea and TSP prices rose only 21.5% and 18.1%, respectively. As a result, many farmers enjoyed a significant improvement in terms of trade as output price increases for rice and many tradable crops outstripped the rise in fertilizer prices. However, once fertilizer markets were liberalized, prices for urea and TSP rose 104.0% and 90.6%, respectively, in the six months from Q3, 1998 until Q1, 1999. These price increases raised fertilizer costs after the start of the 1998-99 rainy season, the largest planting season during the 1999 survey round. The effect on farmer's fertilizer use will depend critically on the timing of their fertilizer purchases. Figure 6 presents the real wage index from the FTT survey for hoeing and planting activities in agriculture. Wages adjusted slowly to price increases for food and fuel, leading to a considerable decline in real wages. The decline in real wage was at least partially responsible for the lower income growth for agricultural workers identified in the regressions above. 30 Figure 5: Fertilizer Prices 500 400 300 1994=100200 100 0 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Jun-99 Urea TSP Source: Components of the Farmers Terms of Trade input price (It) series from BPS. 31 Figure 6: Real Agricultural Wages Hoeing Wage Index Deflated by Rural CPI 160 140 120 100 1994=100 80 60 40 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Jun-99 Dec-99 Lampung C. Java E. Java NTB N. Sulawesi S. Sulawesi Planting Wage Index Deflated by Rural CPI 160 140 120 100 1994=100 80 60 40 Jan-94 Jun-94 Dec-94 Jun-95 Dec-95 Jun-96 Dec-96 Jun-97 Dec-97 Jun-98 Dec-98 Jun-99 Dec-99 Lampung C. Java E. Java NTB N. Sulawesi S. Sulawesi Source: Farmers Terms of Trade Index data, BPS. 32 The Effects of El Niño One of the difficulties in isolating the effects of the economic crisis on agriculture is separating the effects of the changing economic conditions from those of the drought brought on by El Niño in most regions of Indonesia. It appears, however, that in most areas the worst period of drought occurred in 1997, before most of the effects of the economic crisis would have been felt. Figure 7 presents monthly figures on total rainfall over the previous 12 months for a major city in each province in the PATANAS sample.14 In the first frame, one can see that in Lampung, Central Java and East Java the 12 month period of lowest rainfall was the one from February 1997 - January 1998. In NTB, North Sulawesi and South Sulawesi, these troughs seem to occur one to two months later. Nonetheless, in all 6 provinces the drought would have substantially reduced production during the rainy season from October 1997 - March 1998. The latest round of the PATANAS survey covered the period from April 1998 - March 1999. Since levels of rainfall were relatively normal during this period, agricultural productionwould not have been directly affected. However, at the beginning of this period households would have lower stores of grain than in an average year, suggesting that they were cash constrained entering production during the dry season from April until September. This also suggests that households that were negatively affected by the crisis may have suffered two sequential crises: the drought followed immediately by the economic crisis. 14We would like to thank Derek Holmes for providing us with the rainfall data. The data are from the meteorological office (BMG) in the capital city in each province, with the exception of Central Java. The figures for Central Java are from the BMG in Yogyakarta. Rainfall amounts and timing can vary considerably within provinces in Indonesia. As a result, these figures provide only a rough guide as to the actual amount of rainfall that occurred in PATANAS villages. 33 Figure 7: Total Rainfall for Previous Twelve Months 5000 4000 3000 Millimeters 2000 1000 0 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 Lampung C. Java E. Java 5000 4000 3000 Millimeters2000 1000 0 Jan-92 Jan-93 Jan-94 Jan-95 Jan-96 Jan-97 Jan-98 Jan-99 NTB N. Sulawesi S. Sulawesi Source: Meterological office (BMG) in a major city in each province. 34 The Effect of the Crisis on Agricultural Production and Input Use The sources of shocks to household income for many of the cohorts identified above can be traced to vulnerability of agricultural incomes to relative price changes and to changing labor market conditions. The large changes in output and factor prices observed are likely to have diverse effects on the profits of farming households. The size of these effects depends on the farmers' existing technology and ability to respond through changes in production patterns, crop choice and input use. In this section, we consider the magnitude of the effect of crisis-related price changes on agricultural incomes and the forms of adjustment undertaken by farm households. For these purposes, we focus on the 1216 households in the "non-business" sample that had crop income in at least one year, since most of these households are primarily engaged in agriculture. Income From Crops Panel A in Table 13 shows the change in crop profit of non-business farming households by landholdings quintile.15 Crop profit is the value of production minus cost. Prices for the non-marketed portion of crop output are imputed from the household's revenue from actual sales, then from a village average price, then from a province average price for the relevant season. Median profits from crop production grew in per capita terms for all landholdings quintiles except the middle one. Crop profits per hectare--or rents from land--only increased for the largest and smallest landholdings quintiles (see column 7). In general, profits grew substantially for the lowest landholdings quintile, presumably because a number of those who did not grow crops in the base year began operating in 1999. Indeed, in Table 14, which shows the number of farming households in each year and median area planted by province, the number of households engaged in farming in the lowest landholdings quintile jumped substantially from 1995 to 1999. In addition, the smallest increases in area planted came in the lowest and highest landholdings quintiles. One of the reasons that profits per hectare remained strong in these households is because area planted was not sharply increased. Panel B in Table 13 presents the crop profit figures re-weighted to reflect the actual rural population shares by province. In this formulation, smaller farms had the largest increase in median crop profit per capita. The enhanced profitability of larger landholdings quintiles from the unweighted sample in Panel A is eliminated by the population correction. The results from Panels A and B suggest that crop production provided a significant and growing source of income for poor households during the crisis. Results from Table 8 show that the crop income growth for poor households was not sufficient to protect them entirely from the effects of the crisis, since they experienced less than average income growth over the period. However, it appears that, for these households, growth in income from crop production mitigated the effects of declining income from labor and other sources. 15The sample in Table 13 includes the 1216 households in the non-business sample with crop income in at least one of the two years (1995 or 1999). 35 Table 13: Change in Crop Profit per Capita and Crop Profit per Hectare Planted for Non-Business Sample, by Landholdings and by Primary Commodity Grown Median Profit % Chg in Median Profit % Chg in Per Capita Median Per Hectare Median (Rp '000) Profit/Cap (Rp '000) Profit/Ha N 1995 1999 1995 1999 A. 1995 Effective PC Unweighted Landholdings Quintiles 1 174 14 63 367.4% 167 756 353.4% 2 260 100 128 27.9% 1492 1198 -19.7% 3 261 181 170 -5.8% 1398 1232 -11.9% 4 261 238 266 11.8% 1176 1007 -14.4% 5 260 334 349 4.7% 716 845 18.1% All Households 1216 155 196 26.1% 1053 1028 -2.3% B. 1995 Effective PC Census Landholdings Quintiles Weighted 1 0 49 -- 0 1006 -- 2 97 122 26.1% 2224 1865 -16.2% 3 185 138 -25.5% 1644 1488 -9.5% 4 251 232 -7.6% 1493 1166 -21.8% 5 401 339 -15.5% 1266 1156 -8.7% All Households C. Primary Commodity Group Rice 362 153 201 31.2% 1036 884 -14.7% Dryland 296 208 210 1.1% 1401 1152 -17.8% Tobacco & Sugar 126 215 184 -14.4% 1812 1191 -34.3% Trees 289 189 274 44.5% 1085 1137 4.8% D. Province Lampung 258 150 260 73.1% 703 804 14.2% C. Java 215 200 165 -17.8% 3029 2244 -25.9% E. Java 129 43 73 68.4% 765 1123 46.8% NTB 191 129 120 -7.0% 949 570 -39.9% N. Sulawesi 167 115 205 77.6% 538 813 51.2% S. Sulawesi 256 245 336 37.0% 1461 1322 -9.5% Panel C in Table 13 shows change in median returns to farming by primary commodity grown. A household is considered to be a farmer of a particular commodity group if 36 more than 50% of its crop revenue came from that commodity group in 1995.16 Here the effects of the change in relative prices are quite clear. Rice farmers and tree crop farmers experienced the largest gain in crop profits per capita, while dryland crop farmers had almost no income growth and tobacco and sugar farmers suffered losses. The last column of Panel C in Table 13 shows that median farm returns per hectare declined for each group with the exception of tree crop farmers. A more detailed examination of the performance of these groups is presented in kernel density estimates in Figure 8.17 The distribution of profit per hectare shifts back, with a modest fattening of the upper tail, for both rice and dryland farmers. Tobacco and sugar farmers, who had among the highest returns per hectare in 1995, experienced dramatic declines in crop income as the sharply skewed distribution in favor of higher incomes became skewed instead toward the bottom of the distribution in 1999. The distribution of profits for tree crop farmers shows a significant shift into the upper tail. Cumulative distributions show that 16.6% of tree crop farmers had real returns per hectare over Rp. 10 million in 1998/99 (roughly $4,500 in 1994/95 US dollars), while only 3.6% of rice farmers and 8.1% of tobacco and sugar farmers made such profits. Table 14: Number of Farming Households and Area Planted, 1995 & 1999 Number of Farming Net Median Area % Change in Households (with non- Change Planted, Ha Median Area zero area planted) Planted Either 1995 1999 1995-99 1995 1999 year A. 1995 Effective PC Landholdings Quintiles 1 174 108 165 57 0.32 0.38 18.8% 2 260 254 256 2 0.26 0.50 92.3% 3 261 257 258 1 0.52 0.75 44.2% 4 261 259 258 -1 1.00 1.25 25.0% 5 260 254 254 0 1.54 1.82 18.2% All 1216 1132 1191 59 0.65 0.87 34.2% Panel D in Table 13 shows that crop profits per capita declined in Central Java and NTB, but increased substantially in Lampung and North and South Sulawesi, as is consistent with the regional cropping pattern. However, crop profits per hectare declined in South 16 The total number of households listed by primary commodity group is 1073. This is less than in the non- business farming sample because some farms were sufficiently diversified that they did not earn more than 50% of revenues in 1995 from any one commodity group. 17 Kernel densities are a form of non-parametric density estimation in which the density, f(x), of a variable, x, is constructed by looking in a band around each point in x and counting the number of observations in the sample that are "near" the point (within the band). For a brief introduction into kernel density estimation, see Deaton (1997). 37 Sulawesi as median area planted increased there from 0.75 hectares to 1.1 hectares from 1995 to 1999. Regional differences in the change in the distribution of crop income can be further illustrated through nonparametric density estimation. Figure 9 presents kernel density estimates for the logarithm of crop income per hectare on Java and off Java for 1995 and 1999.18 Figure 9 provides convincing evidence that returns to farming were more adversely effected on Java. In 1995, Javanese households generally have higher crop profits per hectare than households off Java. This reflects more intensive cultivation techniques on Java as a result of relative land scarcity there. However, there is a dramatic shift back in the distribution of crop profits per hectare on Java following the crisis that is not experienced on the outer islands. The mode returns to farming on Java fell by 40% between the periods. Despite the broad decline in farm incomes on Java, the distribution also flattened out, with considerable movement into the upper tail; some Javanese households became much richer after the start of the crisis. Off Java, there was very little change in the distribution at lower income levels, but a considerable fattening of the upper tail, presumably a result of the boom in the price of tradable crops. 18The density estimates in Figure 8 are based on the 1216 households in the non-business sample that have non-zero crop income in at least one year. Households with negative or zero profit are excluded from the density estimation of log crop profit per hectare. These households comprise 12.7% of the sub-sample in 1995 and 7.4% in 1999. 38 Figure 8: Kernel Density Estimates of Distribution of Ln Crop Profit Per Hectare by Primary Commodity Group, 1995 & 1999 Rice, 1995 Rice, 1999 Dryland, 1995 Dryland, 1999 .5 .5 .4 .4 .3 .3 .2 .2 .1 .1 0 0 2 4 6 8 10 12 14 2 4 6 8 10 12 14 Kernel Density of Ln Crop Profit per Ha Kernel Density of Ln Crop Profit per Ha Rice Farmers, 1995 & 1999 DrylandCropFarmers,1995&1999 Tobacco & Sugar, 1995 Tobacco & Sugar, 1999 Tree, 1995 Tree, 1999 .5 .5 .4 .4 .3 .3 .2 .2 .1 .1 0 0 2 4 6 8 10 12 14 2 4 6 8 10 12 14 Kernel Density of Ln Crop Profit per Ha Kernel Density of Ln Crop Profit per Ha Tobacco&SugarFarmers,1995&1999 TreeCropFarmers,1995&1999 Figure 9: Kernel Density Estimates of Distribution of Ln Crop Profit Per Hectare on Java and off Java, 1995 & 1999 Java, 1995 Off-Java, 1995 Java, 1999 Off-Java, 1999 .6 .4 .2 0 2 4 6 8 10 12 14 Kernel Density of Ln Crop Profit per Ha Java & Off-Java, 1995 & 1999 39 Thus far we have identified the extent of changes in crop profits following the onset of the crisis by landholding quintiles, by primary commodities grown, and by province. In Table 15, the analysis is broadened to examine changes in crop profit as a function of household and farm characteristics using multivariate regression. Dependent variables are, alternately, change in log crop profit per capita and change in log crop profit per hectare from 1995-1999 for the non-business sample of farmers. All regressors are figures from 1995 unless stated otherwise. Household characteristics used as explanatory variables include number of household members belonging to several age/gender cohorts, log of household size, female headship, and education of household head. Of these regressors, the number of male household members over age 65 is negatively related to growth in crop profit in both regressions. This most likely reflects the lower productivity of older laborers and possibly their reduced ability to adapt to rapidly changing incentives. Larger households experienced higher growth in crop returns per capita; an increase in initial household size from four to five people is associated with a 7.5% higher income growth rate.19 This probably represents the effect of a larger labor endowment on output. Table 15: Determinants of Change in Crop Profit, 1995-1999 Dependent Variable: Change in Log Per Change in Log Capita Crop Crop Profit Per Profit, 1995-99 Hectare, 1995-99 Coef. t-stat Coef. t-stat No. male HH members age 13-24 -0.014 -0.202 -0.042 -0.589 No. female HH members age 13-24 0.001 0.013 -0.033 -0.466 No. male HH members age 25-64 -0.063 -0.526 -0.079 -0.660 No. female HH members age 25-64 0.037 0.321 0.017 0.148 No. male HH members age 65 and up -0.631 -3.349 ** -0.433 -2.293 ** No. female HH members age 65 and up 0.234 1.013 0.187 0.807 Ln of household size 0.337 1.805 * 0.024 0.130 1 if female household head -0.025 -0.090 0.338 1.199 HH head education -0.010 -0.379 0.014 0.515 Ln of land owned per capita -0.072 -1.495 0.120 2.487 ** Share of planted area in corn -0.618 -2.850 ** -0.481 -2.211 ** Share of planted area in (non-corn) dryland crops -0.519 -2.643 ** -0.469 -2.382 ** Share of planted area in tobacco & sugarcane -0.592 -1.686 * -0.586 -1.665 * Share of planted area in tree crops 0.642 3.688 ** 0.560 3.204 ** 1 if HH owns plow in 1994 0.157 1.062 0.071 0.481 1 if HH owns pesticide sprayer in 1994 0.182 1.580 0.142 1.231 1 if HH owns water pump in 1994 0.083 0.205 0.086 0.212 Constant -0.245 -0.909 0.348 1.284 N 937 937 Adj. R2 0.072 0.062 ** Significant at 5% level. * Significant at 10% level. 19Average household size in 1995 was 4.56 people. 40 Among the farm characteristics used as regressors, log of land area owned is positively related to returns per hectare. This variable may reflect the benefits of land ownership as a source of collateral, in addition to its direct productive role. The coefficient of 0.12 implies that, at the average farm size per capita in 1995 of 0.24 hectares, a one standard- deviation increase in farm size resulted in a 3.9 percentage point increase in crop profits per hectare. The share of area planted under various commodity groups in 1995 indicate the farmer's reliance on a particular type of crop, since there are costs to shifting cultivation between these commodity groups in the short run. The share of area planted in rice is omitted as an explanatory variable in order to avoid singularity of the regressor matrix. This means that the coefficients on the included area shares represent the effect of the share of area under the given crop relative to rice. Results show that area share planted to corn, dryland crops and tobacco and sugarcane reduce crop returns relative to the share of area planted in rice. This implies, for example, that shifting 10% of area planted from rice to corn leads to a decline in the (log) growth rate of crop income per capita of 6.18%. A similar shift in area share from rice to tree crops is associated with a 6.42% increase in the growth rate of crop income per capita and a 5.6% increase in the growth rate of crop income per hectare. Finally, ownership of farm assets such as a plow, pesticide sprayer or water pump were not significant determinants of crop income growth during this period. Crop Choice Given the large shifts in terms of trade following the start of the crisis, we have seen that performance of agricultural households during this period depends to some extent on household composition and critically on farm size and crop choice. A key component to households' sensitivity to crisis-related shocks is their response to shifting terms of trade and market conditions through changes in choice of crops grown and factor utilization. Table 16 provides summary statistics on measures of crop composition. Columns 1-3 present the percentage of farming households growing commodities in each of four commodity groups in 1995 and 1999. Columns 4-6 show changes in mean area planted under each commodity for all farming households. This measure of planted area includes multiple plantings per year on the same plot. There was a modest increase in the number of households growing rice, but a large jump in mean rice area planted in response to higher producer prices. Both the number of households and area under dryland crop production grew considerably. The increase in number of dryland crop farmers comes, in part, from a 32% rise in the number of households growing corn. Some of these households may have shifted to corn production as a more reliable source of income during the 1997-98 drought. A large number of households withdrew from tobacco and sugarcane production, leading to a decline in average area planted in these crops. Also, there was a 12.8% increase in households growing tree crops. These results are generally consistent with the observed changes in relative output prices. 41 Table 16: Change in Crop Composition and Revenue, 1995 and 1999 Commodity Percentage of Farm HHs Mean Area Median Real Revenue per Ha Group Growing Commodity Planted, (Hectares) from Commodity Group (Rp. `000) 1995 1999 chg, % 1995 1999 chg, % 1995 1999 chg, % Rice 46.4 48.5 4.7 0.364 0.487 33.7 1055 1205 14.1 Dryland Crops 47.1 55.2 17.1 0.412 0.757 83.6 840 739 -12.0 Tob. & Sugar. 14.2 8.7 -38.6 0.082 0.065 -21.0 1624 1188 -26.9 Tree Crops 50.6 57.1 12.8 0.471 0.560 19.0 878 722 -17.7 The effect of changing output prices and input use on revenue per hectare planted is captured in columns 7-9 of Table 16. The measure of area planted used here represents cropping intensity, so it includes the area for each planting. For example, if a 0.5 hectare plot is planted three times in a year and a 0.2 hectare plot is planted once, the cropping intensity measure of area planted is 1.7 hectares. This differs from the measure of planted area used to measure crop profit per hectare in Table 13. In Table 13, area planted is plot size, regardless of number of plantings. This measure would be 0.7 hectares in the example.20 Using cropping intensity in Table 16 removes the income gains that derive purely from more intensive use of the land in order to highlight output price increases and intensity of use of other factors. Only rice experienced an increase in revenue per hectare planted. All other commodity groups had declining revenue per hectare, with the smallest drop hitting tree crop farmers. This suggests a reduction in non-land factor intensity, which we will investigate below. Finally, the growth in revenue per hectare for rice warrants further investigation. Table 17 shows median rice yields by landholdings quintiles and for the entire non-business sample. Rice yields dropped for each quintile except the highest, and the decline in yields was greatest among the two bottom landholdings quintiles. Based on these results, the increase in median crop revenue per hectare must be due in part to increasing rice prices, since output per hectare is falling. In addition, the larger decline in median yields for the poor suggest that these households were constrained in taking advantage of higher rice prices. 20The land planted variable used in Table 13 is the first land planted variable summarized for per capita income quintiles in Table 3. The cropping intensity variable used here in Table 16 is the second area planted variable in Table 3. 42 Table 17: Change in Median Rice Yields by Effective PC Landholdings Quintile 1995 Effective PC 1995 1999 % Change Landholdings Quintiles N Median N Median 1995-99 1 39 3448 68 2879 -16.5 2 117 3750 123 3200 -14.7 3 108 3045 114 2800 -8.1 4 130 2993 129 2800 -6.5 5 128 2371 146 2500 5.4 All 522 2973 580 2857 -3.9 Cost of Production Table 18 presents average costs of production and cost shares by year for the sample of non-business farming households that have non-zero costs in the relevant year.21 Because of the diversity of crops grown and the extensive use of inter-cropping, it is not possible to separate factor demand by crop. The most striking result from these figures is the considerable substitution of labor for other factors of production during the crisis; average non-labor cost per hectare declined 28%, while labor cost per hectare rose nearly 150%. The extent of substitution of labor for other factors is even greater if the increase in household labor hours (discussed below) is considered. The share of non-labor cost (excluding the implicit cost of household labor time) fell from 67.8% in 1995 to 57.5% in 1999, while hired labor cost share rose from 32.2% to 42.5%. The greatest reductions in non-labor cost came in a 60% drop in seed cost per hectare and a 29% decline in fertilizer per hectare. An exception to declining non-labor input use is in herbicide expenditures, which nearly doubled. Growth in labor costs fed an increase in total cost of production, but aggregate cost per hectare declined. The composition of on-farm labor use also shifted somewhat toward hired labor away from household labor, with hired labor share of hours worked rising from 26% to 29%. 21Costs of production in Table 18 are deflated using the provincial input price index from the BPS Farmers Terms of Trade data. All expenditures are in Oct. 1994 ­ Sept 1995 prices. Calculation of real 1999 expenditures is sensitive to the choice of deflator. The sub-index of consumer prices from the FTT index rose 25% more on average than the sub-index of input prices. We believe that the input price index is a better deflator for fertilizer prices. However, the reader should be aware that the choice of deflator has a significant effect on calculation of changes in expenditures between 1995 and 1999. For example, mean real fertilizer costs increased 14.8% when deflated by the input price index, but fell 10% when the CPI is used as the deflator. Of course, annual cost shares are not affected by the choice of deflator. 43 Table 18: Cost of Production Mean Cost of Mean Cost per Mean Cost Cost Structure Production Hectare Share, (%) (Rp '000) (Rp '000) 1995 1999 1995 1999 1995 1999 Non-Labor Cost 474.7 537.3 939.7 674.4 67.8 57.5 Seeds 121.9 75.3 306.7 123.7 17.1 7.4 Fertilizer 204.4 234.6 394.8 281.1 37.4 26.8 Pesticide 42.1 83.1 121.4 122.5 4.6 4.4 Herbicide 3.3 16.1 4.6 8.9 1.3 2.6 Tax 7.9 9.6 23.8 14.1 0.5 7.3 Other non-labor cost 95.2 118.5 88.4 124.1 6.8 8.9 Hired labor cost 219.4 493.3 254.1 628.7 32.2 42.5 (Hours worked by hired labor) (485.9) (571.5) (656.7) (694.8) (37.5) (33.2) (Hours worked by HH labor) (639.2) (1067.4) (1369.0) (1481.6) (62.5) (66.8) Total cost 694.1 1030.7 1193.8 1303.1 Number of Households 1067 1107 Fertilizer Use Changing patterns of fertilizer use deserve additional attention given the large increase in fertilizer prices in 1998 and the large share of fertilizer in the budget of most Indonesian farmers. Nominal fertilizer prices grew at a rate of roughly 20% per year for the first year of the economic crisis (from QIII of 1997 to QIII of 1998), only slightly faster than trend for the previous three years. However, in November 1998, substantial fertilizer subsidies were eliminated as part of the fertilizer market liberalization required under an agreement with the IMF signed earlier in the year that provided $40 billion in capital to bail out the failing Indonesian economy. Fertilizer subsidies had created massive distortions that resulted in the development of a black market for smuggling of subsidized fertilizer. Following the elimination of subsidies, fertilizer prices rose dramatically, with prices of urea and TSP increasing 93% and 80%, respectively, in just two months (see Figure 5). Since output prices for many commodities rose faster than the price of fertilizer at the beginning of the crisis, fertilizer demand should have remained strong until the elimination of subsidies, unless credit constraints limited access to purchased inputs. The data on cost of production from Table 18 show an increase in mean fertilizer expenditure from 1995 to 1999, but a decline in average fertilizer expenditure per hectare. Table 19 presents change in fertilizer expenditure per hectare by landholdings quintile and by primary commodity grown for the sample of non-business farming households with crop 44 production in at least one year.22 Median fertilizer expenditures increased for the poorest landholdings quintile, primarily due to initiation of crop production by households that did not produce any crops in the base year. All other landholdings quintiles had a reduction in median fertilizer expenditures per hectare except for the richest landholding quintile. This trend probably reflects the constrained cash position of poorer households following the drought and during the crisis. Table 19: Change in Fertilizer Expenditure per Hectare, 1995-1999 No. of HHs Using Median Fertilizer Mean Fertilizer HHs Fertilizer Expend. per Ha, Rp `000 Expend. per Ha, Rp `000 1995 1999 1995 1999 Chg, % 1995 1999 Chg, % 1995 Effective PC Landholdings Quintile 1 151 67 108 0.0 87.5 -- 314.9 227.1 -27.9 2 252 207 200 152.6 130.6 -14.4 513.6 345.3 -32.8 3 258 222 202 146.6 127.6 -13.0 337.4 315.0 -6.6 4 253 221 203 125.0 100.8 -19.4 335.1 271.6 -18.9 5 251 190 191 49.3 63.1 27.9 196.8 152.1 -22.7 Primary Commodity Grown Rice 360 327 313 104.8 98.3 -6.2 141.3 181.3 28.3 Dryland Crops 294 255 257 170.1 155.8 -8.4 370.7 344.8 -7.0 Tobacco & Sugarcane 126 123 107 781.5 401.9 -48.6 1436.5 711.6 -50.5 Tree Crops 267 161 145 19.8 15.5 -22.0 137.6 113.0 -17.9 All Farm Households 1165 907 904 111.7 101.2 -9.4 341.8 265.6 -22.3 For the entire sample, median real fertilizer expenditure per hectare fell 9.4%. This decline was smallest for households earning more than 50% of their revenue from rice or dryland crops in 1995. In fact, mean fertilizer expenditures per hectare increased 28% for rice farmers.23 The rise in average fertilizer expenditures for rice farmers is not surprising; fertilizer demand should have grown as rice producer prices (in the Farmers Terms of Trade data from BPS) roughly doubled during the period of limited fertilizer price increases from QIII 1997 to QIII 1998. However, given the large increase in fertilizer prices at the end of 1998, it is not possible to determine from the annual expenditure data how demand for fertilizer was affected in quantity terms. Did the intensity of fertilizer demand per hectare increase after the start of the crisis or did credit constraints and low cash reserves following the drought hinder growing demand? And is 22It is not possible to separate fertilizer expenditure by commodity, so we identify changes in expenditure for each household by the commodity group that was responsible for more than 50% of revenue in 1995. Figures in Table 19 are real fertilizer expenditures, deflated by the FTT provincial input price index into 1994/95 prices. 23Mean fertilizer expenditure per hectare differs in Table 18 and the bottom of Table 19 because the former summarizes expenditures in each year for households with strictly positive production costs in the given year. Table 16 presents summary statistics for households with non-zero costs of production in either year. 45 the general decline in annual fertilizer expenditures due primarily to the elimination of fertilizer subsidies in November 1998? In order to answer these questions, we consider the change in demand for quantities of urea, TSP, and KCL by season. These three types of fertilizer are the largest components of fertilizer demand; they represented 43.4%, 24.4%, and 5.9% of average fertilizer expenditure in 1995, respectively. Table 20 presents average fertilizer demand by season for each type of fertilizer.24 The timing of the rainy season during the year varies by region, but generally falls between October and March. There are two dry seasons: "Dry I" immediately follows the rainy season and lasts until June, and "Dry II" is the period of lowest production, from July­September. The fertilizer liberalization occurred after the dry seasons of 1998/99 and several weeks into the rainy season. Any farmer that had not purchased fertilizer prior to the elimination of subsidies faced significantly higher prices during the rainy season. With regard to the first question, Table 20 shows an increase in intensity of urea application and a decline in intensity of TSP and KCL use on seasonal crops for all three seasons from 1995-1999. Since the change in the prices of these types of fertilizer were roughly equivalent (see Figure 5), this pattern of expenditure probably did not result from substitution between fertilizers by farmers who grew the same commodities in both periods. Instead, the increased use of urea and declining demand for TSP and KCL is probably due to relatively higher prices for commodities for which urea is the typical fertilizer treatment, such as rice. Similarly, this expenditure patterns signifies a decline in the terms of trade for growers of commodities that rely on TSP and KCL. Also of note in Table 20 is the substantial increase in fertilizer demand for annual crops, many of which benefited from favorable terms of trade shocks during the crisis. The second question concerns whether the change in fertilizer expenditure can be attributed to direct crisis-related effects or the elimination of fertilizer subsidies--an indirect outcome of the events following the start of the crisis. Table 18 shows a limited effect of the fertilizer market liberalization on rainy season fertilizer demand. The drop in demand for TSP and KCL was greatest in the rainy season, providing some support for a demand reducing effect of liberalization. However, there was a substantial decline in TSP demand per hectare during the dry season--before the removal of subsidies-- suggesting that crisis-related changes in output prices may be responsible for the declining factor demand. Evidence for the timing of the change in urea demand is mixed. Average rainy season urea demand per hectare from all crops increased from 1995 to 1999, and by more than the increase in the first dry season. However, average urea demand from households whose primary commodity is rice (not shown in Table 20) grew 9.5% between the first dry seasons of 1995 and 1999, but rose only 2.2% between the rainy seasons in these years. This might suggest some dampening effect of the market liberalization on peak season urea demand from rice farmers. 24Mean fertilizer demand is presented in Table 20 rather than medians because medians for the various types of fertilizer used were mostly zero. 46 Table 20: Change in Fertilizer Demand by Season, 1995 & 1999 1995 1999 % Change Type of (N=1165) (N=1165) in Mean Fertilizer Season No. of HHs Mean Std. Dev. No. of HHs Mean Std. Dev. 1995-99 using fert. using fert. Urea demand Rainy 616 122.7 204.6 770 141.8 171.0 15.6 (Kg/Ha) Dry I 459 100.0 208.1 522 106.5 180.1 6.5 Dry II 171 51.0 178.5 320 67.0 193.2 31.3 Annual 60 9.3 47.1 296 45.9 140.1 394.9 TSP demand Rainy 446 55.4 123.0 271 29.3 75.3 -47.1 (Kg/Ha) Dry I 348 47.4 129.2 210 27.8 87.0 -41.3 Dry II 144 24.1 101.4 144 19.9 76.2 -17.7 Annual 37 3.5 23.9 117 11.6 47.3 231.5 KCL demand Rainy 117 9.5 34.0 109 8.4 37.9 -12.1 (Kg/Ha) Dry I 99 10.6 47.0 99 10.3 47.7 -2.8 Dry II 55 7.1 39.5 52 7.0 69.0 -1.5 Annual 37 3.5 21.9 126 11.9 44.8 240.6 47 On-Farm Labor Demand The pattern of change in labor demand on farm may be quite different than for fertilizer because real wages fell during the crisis and there were changes in returns to off-farm employment for household members. Table 21 presents the change in median labor demand for crop production by 1995 effective per capita landholdings quintiles for households from the "non-business" sample with crop income in at least one year. Over the entire sample, median labor hours per hectare increased. The large increase in labor hours for the lowest quintile is due primarily to the entry of new farmers. Among the other four quintiles, the second quintile had the smallest increase in labor intensity. Table 21: Change in Median Labor Hours per Hectare On Farm, by PC landholding quintiles 1995 Effective PC N 1995 1999 % Change, Landholdings Quintile 1995-99 1 174 316 1050 232.4 2 260 1680 1844 9.8 3 260 1012 1531 51.3 4 261 856 1209 41.3 5 260 522 793 51.9 All farm households 1215 840 1242 47.8 The size of the increase in labor hours also varies with commodities grown. Table 22 shows that tobacco and sugarcane producers dramatically reduced labor hours per hectare as a cost saving measure in the face of falling prices. Rice and dryland crop producers increased labor intensity about as much as for the sample as a whole. However, tree crop producers had a much larger increase in labor demand as predicted by the terms of trade boom enjoyed by those farmers. Javanese households had a small decline in median crop labor demand (mostly in Central Java), while crop labor demand jumped in the outer islands. Disaggregated data on sources of labor show that the decline in labor demand on Java is due to a 27% reduction in median hired labor hours there. Median household labor hours on Java increased 25% from 1995 to 1999. Table 22 also shows that, for the whole sample, household labor hours increased much more than hired labor hours. This is consistent with results from Table 18 that show an increase in the mean share of hours worked by household labor from 62.5% to 66.8%. This trend suggests an increase in nominal wages when there is friction for household members entering the labor market. Rising wage costs increase the share of household labor on farm if household members are not able to take full advantage of the wage increase in off-farm employment. 48 Table 22: Changes in Annual Labor Hours per Hectare in Crop Production by Primary Commodity Group, Location, and Type of Labor, 1995-99 Hours Per Hectare 1995 1999 % Chg in N Median Mean Std. Median Mean Std. Median Dev. Dev. 1995-99 By primary commodity grown Rice 362 864 1395 2055 1210 1566 1339 40.1 Dryland Crops 296 991 2206 3363 1480 2650 3789 49.4 Tobacco & Sugar 126 4544 5235 4757 2548 3181 3275 -43.9 Tree Crops 287 582 1212 3287 1086 1686 2636 86.5 By location Java 344 2545 3778 4896 2378 3550 4350 -6.5 Off-Java 871 674 1098 1896 1031 1501 1727 52.9 By type of labor Hired 1216 175 585 1206 207 644 1592 18.4 Family 1215 445 1272 2859 775 1433 1995 74.1 All hours 1215 840 1857 3288 1242 2079 2884 47.8 Further evidence on the changing use of household and hired labor on-farm is presented in Figure 10, which shows nonparametric density estimates of the distribution of household and hired labor hours in crop production for each year. The distribution of hired labor per hectare is generally unchanged, while the distribution of household labor per hectare shifts right. If household labor is disaggregated by gender, median labor hours per hectare worked by female household members on farm increased 68% from 1995-99, while median on-farm labor hours per hectare worked by male household members increased 37%. 49 Figure 10: Household and Hired Labor Hours per Hectare in Crop Production, 1995 & 1999 Family, 1995 Hired, 1995 Family, 1999 Hired, 1999 .3 .2 .1 0 0 2 4 6 8 10 Ln of Labor Hours per Hectare 50 IV. Other Household Strategies for Adapting to Crisis Effects Migration The pattern of migration following the onset of the economic crisis demonstrates both crisis-related effects on rural households--as urban-based relatives return to their rural homes following employment loss in the cities--and the reaction of rural households to crisis impacts--as household members move to take advantage of employment opportunities in booming centers of tradable agricultural production. One of the earliest predictions at the start of the crisis was that there would be massive flight from depressed urban centers. It is also possible that the terms of trade windfall for farmers of tradable crops could produce a "pull" effect in which urban workers migrate to rural areas in search of work. However, reports from the sociologists who undertook a parallel qualitative study of the effects of the crisis in PATANAS villages state that urban-to-rural migration was not a major source of changing rural demographics. This was due, at least in part, to weak "pull" effects: urban workers were generally unskilled at agriculture activities and the urban migrants often considered farmwork to be undesirable. Instead, the sociologists found substantial rural-to-rural migration as skilled (usually male) farm workers sought jobs in centers of tree-crop production. They noted, for example, reports of workers moving from Lombok and Sumbawa in NTB to South Sulawesi to take advantage of higher-paying jobs there. In general, it is difficult to identify the dynamics of migration from micro-level household data sets such as PATANAS that have limited spatial coverage. However, retrospective questions from the household roster of the 1999 PATANAS survey provide some evidence concerning the hypotheses of increased migration and its sources. In 1999, 13.3% of all household members had entered the household since the 1995 survey and 12.6% of original household members had left. If births and deaths are excluded, the largest annual inflow of "entrants" and the largest annual exodus of "leavers" both occurred in 1998, after the start of the crisis. In fact, the pre-crisis rate of new household member entry into all PATANAS households was 12.9 entrants per month before the crisis (from October 1995 ­ July 1997) and 17.9 entrants per month after the crisis began (August 1997- April 1999). The rate of exit from the household was more even pre- and post-July 1997, with 14.4 people leaving per month before and 15.4 after. The destination of migrants also shifted after the start of the crisis. Those that entered or left the house after July 1997 moved farther away and migration from cities and other provinces increased. Only 42% of all movers (entrants and leavers) before the crisis traveled beyond their village or sub-district, but 55% did so after the crisis began. If migrants are defined as those that move beyond their village or sub-district, migration increased 44% (from 192 to 276 individuals) after the crisis began. The hypothesis of increased out-migration from the cities is generally supported by the data. The rate at which new entrants arrived from Jakarta or the capital city of the province or district doubled after the start of the crisis. There is also some support for the sociologists' findings of increased rural-to-rural migration across provinces. The share of 51 movers traveling to another province rose from 6% to 13%. While it is not possible to know if all of this migration was between rural areas, the most popular destinations after the crisis began were Lampung and South Sulawesi where there are a large number of tree crop plantations. The 1999 questionnaire also asked respondents the reason for migration of household members. Of the 1435 entrants and leavers that answered this question, only 56 claimed they moved because they lost work or were looking for work. However, three quarters of these moved after the crisis began. Of the 27% of leavers that moved because of a work- related change of assignment or location, 63% of these moved after the crisis. In other evidence of push or pull effects, landless households were no more likely to lose a household member to out-migration after the crisis than landed households, suggesting no crisis-related push effects for the landless. Although the probability that a tree-crop producing household (those with most of their crop revenue from tree crops in 1995) will have an entrant from 1995-1999 is no greater than for the rest of the sample, new entrants to tree-crop producing households are far more likely to arrive after the start of the crisis. 52 V. The Effectiveness of Government Programs to Protect the Poor In the formulation of household vulnerability to macroeconomic shocks outlined by Glewwe and Hall (1995), after a household experiences a shock and adapts to protect its level of expenditures, the third stage of events that determines the ultimate depth of the shock for the household is government spending to shore up the incomes of those effected. Below we will consider the evidence from the PATANAS survey on the breadth and effectiveness of government programs designed to boost incomes and protect consumption of households that were adversely effected by the economic crisis. One such program is a collection of public work programs called padat karya (literally, "labor intensive") that hired local men or women--usually without any direct screening--for temporary employment on projects that built roads or provided other public services. Padat karya programs are operated in Indonesia on an ongoing basis, but these projects were greatly intensified during the crisis and were intended to serve as a cornerstone of the government's crisis safety net. A second safety net program that we consider, called Operasi Pasar Khusus (OPK) or "special market operation", distributed subsidized rice to poor households. Households identified as poor were supposed to receive 20 kilograms of rice per month at a cost of Rp. 1,000/kg., generally less than half the market price. Padat Karya Work Programs and Other Government Income Transfers As part of the 1999 PATANAS survey module on off-farm household labor supply and income generation, data were collected on hours worked and income earned by household member from padat karya workfare projects. These data suggest that padat karya programs exist in many parts of the country but are heavily concentrated in some regions. Out of 1494 households in the 1995-99 PATANAS panel, 153 households (10.2%) had at least one member that participated in a padat karya program from April 1998 ­ March 1999. A total of 169 individuals reported earning income from padat karya during this period; 86% of these were males. While at least one household in each province reported padat karya income in 1998/99, most individuals working in these programs were concentrated in NTB (61%), Central Java (23%), or South Sulawesi (11%). Nearly three-quarters of the jobs in these programs were in construction; another 21% were in security (night watchmen) or other services. Wages paid varied by project and by region. In NTB, padat karya programs typically paid nominal wages between Rp. 1,000 per hour and Rp. 1,250 per hour (with a median of Rp. 1,250).25 Median padat karya wages were Rp. 1,000 per hour in Central Java and Rp. 750 per hour in South Sulawesi. Interestingly, median padat karya wages were higher than median wages for other non-agricultural employment in NTB and Central Java, but the padat karya wage was less than 50% of the comparable market wage in South Sulawesi. This suggests that there must have been some screening in NTB and Central Java to restrict the number of entrants into the program. 25The average exchange rate during the survey period was Rp. 9745/$US. 53 The 1999 PATANAS questionnaire also gathers information on other government transfers, excluding padat karya and farmer credit programs. Sixty-three households in the sample reporting receiving some income from other government programs. These households were primarily in Central Java (41%), North Sulawesi (25%) and Lampung (22%). Table 23 shows the distribution of government expenditure on PATANAS households in 1998/99 under the padat karya program and through other government income transfer programs by province. Average nominal per capita government expenditure on households participating in the padat karya program (or, equivalently, per capita household income from padat karya) were substantially higher in Central Java than in NTB. However, average per capita expenditures per recipient household were far greater in South Sulawesi. This probably reflects the fact that a few households in South Sulawesi worked the majority of their labor hours during the year under this program.26 Average per capita government expenditure per recipient in other government programs was not as varied across the provinces. Table 23: Government Expenditure on Safety Net Programs by Province, 1998-99 Padat Karya Public Works Program Other Government Programs (excl. Farm Credit) 1999 Non-gov. No. of Per Capita Province Mean PCE No. of Per Capita Province Mean PCE PC Income HHs in Expenditure Share of per recipient HHs in Expenditure Share of per recipient Quintiles Program (Rp '000) Expenditure HH, (Rp '000) Program (Rp '000) Expenditure HH, (Rp '000) Lampung 3 341 0.04 113.6 14 169 0.17 12.1 Central Java 37 1,791 0.21 48.4 26 293 0.30 11.3 East Java 3 1,528 0.18 509.5 2 9 0.01 4.5 NTB 96 2,263 0.26 23.6 4 242 0.24 60.5 N. Sulawesi 1 13 0.00 12.5 16 178 0.18 11.1 S. Sulawesi 13 2,638 0.31 202.9 1 96 0.10 96.0 All 153 8,573 56.0 63 987 15.7 These safety net programs had moderate success in targeting transfers to reach the poor during the crisis. Table 24 presents the number of households participating in the programs, total per capita program expenditure and average per capita transfers per recipient household by 1999 per capita (non-program) income quintile. Income quintiles were generated based on per capita household income earned outside the padat karya or other government programs. Out of 153 households with income from the padat karya work program, 98 households (64%) are in the bottom two per capita non-program income quintiles. The 40% of all households in the two lowest quintiles received 60% of padat karya expenditures on PATANAS households. While this demonstrates some effectiveness in targeting the poor, considerable resources were transferred to the non- 26Large mean per capita expenditures per recipient household in padat karya in Lampung and East Java may be due to respondents incorrectly identifying the income as deriving from a government program. 54 poor. The fourth column of Table 24 shows that, while the average per capita transfer to households in padat karya were relatively high for households in the poorest quintile (Rp. 59,800 per year), households in the middle income quintile received on average 37% more income per capita under the program. Targeting through other government programs was even less effective. While 70% of households receiving transfers were in the bottom two income quintiles, the size of average per capita transfers was (almost) monotonically increasing in per capita income quintiles; rich households received larger per capita transfers than the poor. Table 24: Government Expenditure on Safety Net Programs by Income Quintiles, 1998-99 Padat Karya Public Works Program Other Government Programs (excl. Farm Credit) 1999 Non-gov. No. of Per Capita Quintile Mean PCE No. of Per Capita Quintile Mean PCE PC Income HHs in Expenditure Share of per recipient HHs in Expenditure Share of per recipient Quintiles Program (Rp '000) Expenditure HH, (Rp '000) Program (Rp '000) Expenditure HH, (Rp '000) 1 50 2,988 0.35 59.8 25 241 0.24 9.6 2 48 2,123 0.25 44.2 19 275 0.28 14.5 3 28 2,301 0.27 82.2 7 95 0.10 13.6 4 16 697 0.08 43.5 8 129 0.13 16.1 5 11 465 0.05 42.3 4 247 0.25 61.7 All 153 8,573 56.0 63 987 15.7 The share of household income earned from safety net programs is another indicator that can be used to assess the effectiveness of these government programs in targeting the poor. Table 25 presents summary statistics for program-related income shares under padat karya and other government programs for the entire sample and for those households participating in the relevant program, by 1999 total per capita income quintiles. These figures support the evidence of mediocre targeting presented in Table 24. Also of interest is the generally small contribution of the safety net programs to household income. The median income share from padat karya for households from the lowest per capita income quintile that participated in the program is only 3.1%. These public work programs provided only a modest degree of support to poor households. The effectiveness of targeting under these safety programs is presented clearly in Figure 11, where we graph the cumulative benefit share received by 1999 per capita income percentiles under padat karya and under both padat karya and other government programs. The 45° line represents equal distribution of benefits or zero targeting. The lowest 25th per capita income percentile receives only 38% of the per capita benefits from padat karya. When other government transfers are included, effectiveness of targeting is worse: the lowest 25th percentile receives only 32% of benefits. 55 Table 25: Average Share of Household Income from Government Programs by 1999 Income Quintiles Entire Sample (N=1494) Conditional on Membership in Program 1999 Public Works Other Government Public Works Other Government Total Programs Programs Programs Programs PC Income Quintiles Mean Std. Dev. Mean Std. Dev. Median N Median N 1 0.92 3.40 0.16 0.75 3.14 45 2.15 24 2 0.86 5.68 0.11 0.67 1.15 50 1.35 19 3 0.65 5.04 0.06 0.59 0.95 28 0.93 8 4 0.49 4.41 0.02 0.21 0.62 19 0.24 8 5 0.02 0.24 0.02 0.30 0.14 11 0.55 4 Total 0.59 4.22 0.07 0.55 1.31 153 1.32 63 Figure 11: Targeting in Government Safety Net Programs 1 0.8 Share 0.6 Padat Karya Benefit PK & Other Prgs 0.4 0.2 Cumulative 0 0 20 40 60 80 100 Percentiles of 1999 PC Income 56 Subsidized Rice Program In addition to its income maintenance schemes, the government operated the OPK program to provide subsidized rice to poor households. Households were identified for this safety net program from a survey conducted by the family planning agency (BKKBN). Under the program, each poor household should have received 20 kilograms of rice per month at a price of Rp. 1,000. However, the program suffered several problems that effectively eliminated the targeting of poor households. One of the shortcomings was in the design of the survey instrument, which incorrectly identified households as poor or non-poor based on questions concerning ownership of a limited set of household assets. These survey design problems contributed to a further erosion of targeting effectiveness in implementation. In part because of the apparent arbitrariness in selection of poor households for inclusion in the roster of subsidized rice recipients, many local officials refused to rely on the roster, claiming that to do so would be politically unpopular. Citing such concerns, one village head told World Bank researchers that he simply divided the rice for the program evenly among all households in the village, and sold it to them at the subsidized price. The 1999 PATANAS data captures household purchases of subsidized rice from April 1998 to March 1999. There were 642 households in 1998/99 that said they received subsidized rice at some point during the year. Nearly all of these households were in Lampung, Central Java, East Java, or NTB. Households typically paid Rp. 1,000 per kilogram of subsidized rice regardless of location. This translated into an average savings of Rp. 1,374 per kilogram of subsidized rice. The average annual per capita benefit received by households in the program was Rp. 21,889 (compared to nominal per capita household income for the 10th percentile of Rp. 372,000 during the same period). However, this was based on an average supply of subsidized rice of 5.1 kilograms per household per month, well below the intended transfer of 20 kilograms per month for program participants. The reason for the reduced transfer is clear. Subsidized rice was provided to many non-poor households and was distributed beyond the program guidelines to households not identified as intended recipients. Figure 12 demonstrates the complete ineffectiveness of targeting under the subsidized rice program. The figure graphs the cumulative share of per capita benefits received against percentiles of 1999 per capita income. As in Figure 11, the 45° line represents equal distribution of benefits or zero targeting. The figure shows a complete failure to target this program to poor households. In fact, the 50th percentile of per capita income received exactly 50% of per capita program benefits. 57 Figure 12: Distribution of Benefits from Subsidized Rice Program 1 0.8 Share 0.6 Benefit 0.4 0.2 Cumulative 0 0 20 40 60 80 100 Percentiles of 1999 PC Income 58 VI. Conclusions and Policy Implications This study has used the 1994/95 and 1998/99 rounds of the PATANAS survey to address a number of issues concerning the effects of the economic crisis that began in July 1997 on rural households in Indonesia. In general, the effects of the crisis have been heterogeneous and varied considerably depending on household location, initial socio- economic status, income sources and commodity portfolio. However, a clear result of this analysis is that the economic crisis did not have broad damaging effects on rural households. The economic crunch that gripped urban areas in late 1997 and 1998 was not transmitted into declining incomes for most rural households. One reason for some positive effects from the economic crisis for rural households is the terms of trade boom that resulted from the large depreciation of the rupiah against the dollar. This boosted agricultural incomes for many rural households. Despite this growth in profits from farming, agricultural laborers did not benefit substantially from the enhanced profitability of agricultural products. Outside agriculture, non-farm incomes also rose substantially for all but the rural poor. Labor incomes increased sharply off Java, and business incomes rose on Java. The results also show that, in terms of distributional impacts, the poor received generally fewer benefits from the crisis than other households. Households on Java, the landless, and female-headed households enjoyed fewer benefits and were more likely to suffer income losses because of the crisis. The main conclusions of the report are the following: · The effects of the crisis on rural households were heterogeneous, but predictions that the economic crisis in urban areas would have substantial negative effects on the socio-economic status of rural households are not substantiated. Many households enjoyed considerable increases in per capita income during the period. Still, 35% of all households in the sample experienced a decline in per capita incomes from 1995 to 1999. · Farmers generally benefited from the crisis. As expected, crop portfolios largely determined gains in crop income; tree crop farmers, located off Java, did especially well, as did rice farmers. Growers of crops for domestic markets--especially tobacco and sugarcane--did worse. Farmers showed considerable resiliency in shifting cropping patterns to adapt to changing incentives during the crisis. · Surprisingly, non-farm income also rose substantially, though not for the poor. These gains were driven by business income on Java and by labor earnings off Java. Undoubtedly, some, but not all, of these gains were realized prior to the crisis (from 1995 to 1997). · The income boom enjoyed by many households in the outer islands generally did not emerge on Java as per capita incomes there stagnated. Landless Javanese households, in particular, were among the worst off in relative terms. 59 · In general, poorer households experienced smaller income gains than richer households. The lowest landholdings quintile in the sample had a small increase in median agricultural income and declining non-farm income. Results suggest that agricultural production was an important source of income security for the poor during the crisis. · Migration increased after the onset of the crisis. The share of arrivals to rural areas from cities increased. Also, people moved greater distances once the crisis began, with a much larger share of movers traveling outside their sub-district. · Landless households and agricultural wage earners performed worse than land- owning, farming households. Wage earners did not appear to share in the benefits of the terms of trade boom experienced by producers in agriculture. However, wage earnings outside of agriculture increased. · Average intensity of fertilizer use declined for most producers, with the exception of rice farmers. · Household labor use on farm increased sharply during the crisis while hired labor demand per hectare was unchanged. Within the household, female household members increased hours worked on-farm much more in percentage terms than did male members. The evidence suggests that rural women are a labor reserve for households adjusting to the new economic environment. · The effectiveness of targeting poor households through padat karya public works programs and other government income transfer schemes is poor. The bottom 25th per capita income percentile in 1998/99 received only 38% of the benefits from padat karya. · The benefits from the OPK program of subsidized rice were distributed almost evenly across the distribution of households by per capita income. Targeting of subsidized rice to the poor was undermined by an unwillingness of local authorities to accept an externally generated roster of program recipients that they believed inaccurately identified the poor. Given the heterogeneity of crisis-related experiences and the substantial gains in socio- economic status for many households in the PATANAS sample, it is difficult to construct effective policy implications concerning rural households. In general, agricultural policy is a blunt instrument to help the poor, who are primarily small landholders and landless workers. However, poor households with access to land benefited from a substantial increase in farm income during the crisis, suggesting that the effects of agricultural policy on the poor during a crisis should not be ignore. For landless poor households, the search for policy levers is more illusive. Agricultural labor income did not rise substantially for this group, despite the boom in agricultural output prices. Non-farm income-generating activities proved to be a vital source of income growth for all households except those 60 without any land. This suggests that policies that support income generation from outside agriculture should also be encouraged as a source protection from crisis effects for most households, but that poor households without land may need to be reached through direct transfers. Regarding social policy, targeted social programs for those left behind by the commodity boom, or otherwise hurt by the recession, are worth considering. However, the targeting performance of the padat karya program has been poor. Steps should be taken to reduce leakages of program benefits from padat karya to the non-poor. The effectiveness of targeting in the subsidized rice program was even worse than in padat karya. Improving targeting of cheap rice to the poor will require an alternative method of identifying poor households, one that is politically more acceptable in the villages. 61 References Bourgeois, Robin, 1999, Impact of the Crisis on Javanese Irrigated Rice Farmers, mimeo, CASER. Deaton, Angus, 1997, The Analysis of Household Surveys: A Microeconometric Approach to Development Policy: Baltimore, Johns Hopkins University Press. Feridhanusetyawan, Tubagus, August, 1999, The Social Impact of the Indonesian Economic Crisis, mimeo, Thailand Development Research Institute. Frankenberg, Elizabeth, Thomas, Duncan, and Beegle, Kathleen, March, 1999, The Real Costs of Indonesia's Economic Crisis: Preliminary Findings from the Indonesian Family Life Surveys, Labor and Population Program Working Paper Series 99-04, RAND. Glewwe, Paul, and Hall, Gillette, 1995, Who is Most Vulnerable to Macroeconomic Shocks? Hypothesis Tests Using Panel Data from Peru, LSMS Working Paper No.117, World Bank. Poppele, Jessica, Sumarto, Sudarno, and Pritchett, Lant, May, 1999, Social Impacts of the Crisis: New Data and Policy Implications, mimeo, SMERU/World Bank. Tabor, S. R., Dillon, H. S., and Sawit, M. Husein, 1999, Understanding the 1998 Food Crisis: Supply, Demand or Policy Failure?, in Simatupang, Pantjar, Pasaribu, Sahat, Bahri, Sjaiful, and others, Edited by, Indonesia's Economic Crisis: Effects on Agriculture and Policy Responses: Adelaide, Australia, University of Adelaide for CASER. Thomas, Duncan, Frankenberg, Elizabeth, Beegle, Kathleen, and others, March, 1999, Household Budgets, Household Composition and the Crisis in Indonesia: Evidence from Longitudinal Household Survey Data, mimeo, RAND. 62