IMPACTS OF GOVERNMENT MAIZE SUPPORTS ON SMALLHOLDER COTTON PRODUCTION IN ZAMBIA By Joseph Christopher Goeb A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE Agricultural, Food, and Resource Economics 2011 ABSTRACT IMPACTS OF GOVERNMENT MAIZE SUPPORTS ON SMALLHOLDER COTTON PRODUCTION IN ZAMBIA By Joseph Christopher Goeb In Zambia, cotton has been an agricultural success story led by private cotton ginneries and smallholder production. Since liberalization in 1994, the cotton sector has seen periods of dramatic growth and two severe crashes. Production recovered well after the crash in 2000, but recovery since 2007 has not been as strong. The Zambian government has drastically increased its supports to smallholder production of maize since the 2005 harvest year through maize purchases by the Food Reserve Agency (FRA) and subsidized fertilizer targeted to maize through the Farmer Input Support Program (FISP). Because cotton is almost entirely produced in the country‟s main “maize belt”, these maize supports in principle also affect the relative profitability of cotton, but any effects directly on smallholder cotton cropping decisions are largely unknown. This thesis attempts to move towards understanding the effects of the FRA and FISP maize supports on smallholder cotton production in Zambia. Two separate Cragg hurdle models are employed to determine the effects of the maize supports on i) smallholders‟ decisions whether to plant cotton, and ii) their land allocation decisions to cotton given that they decided to plant it. We also track household cotton planting decisions over a ten year period and analyze across several household indicators. Copyright by JOSEPH CHRISTOPHER GOEB 2011 ACKNOWLEDGEMENTS I am very grateful for the support and assistance that I received in the process of writing this thesis and I would like to acknowledge a few of the people that contributed to its completion. First and foremost, I need to thank my professors at Michigan State University. My major professor and thesis advisor, Professor Dave Tschirley, guided me through my entire master program and provided particular support and direction in the process of writing this paper. His considerable time and effort expended on my behalf is greatly appreciated. I am also thankful for Professors Eric Crawford, Thom Jayne and Robert Richardson for being on my committee and for each of their unique perspectives. The feedback given by each of my committee members was constructive and directly improved this paper‟s quality. I am also grateful for the comments, thoughts, and feedback provided by Professor Mike Weber. His efforts and attention to details helped me to refine this work. I also owe several thanks to David Mather for sharing his syntax of manually running the Cragg hurdle model. Likewise, I thank Bill Burke for sharing his Stata program, the “craggit”. Second, I would like to thank my fellow graduate students. Nicky Mason provided helpful advice and ideas in the early stages of this project and helpful feedback in the latter stages. For her time and patience, I am thankful. I am also thankful for the unbelievable efforts of Jenny Cairns. I know that it is neither exciting nor easy to run someone else‟s models, and I cannot thank her enough for her time – without which I would have no final results. Many light thanks are owed to Ross Bowmar, Josh Nielson and Karen Lewis for making my transition to and time in East Lansing so enjoyable. iv Last, and far from least, I need to thank my family. I am infinitely grateful for my wife, Joy, who provided support and late night dinners at the office. Her patience, strength and understanding during a very busy time in our lives were incredible. I also thank my mother, Mary Goeb, and sisters, Carrie and Sarah Goeb, for their love and guidance. v TABLE OF CONTENTS LIST OF TABLES ......................................................................................................... VIII LIST OF FIGURES ............................................................................................................ X LIST OF ABBREVIATIONS ........................................................................................... XI 1. INTRODUCTION......................................................................................................... 12 1.1 BACKGROUND INFORMATION ON ZAMBIA .................................................................. 1 1.2 THE IMPORTANCE OF COTTON ................................................................................... 3 1.2.1 Outgrower Schemes ........................................................................................... 3 1.2.2 The “Cotton-4” Example ................................................................................... 5 1.2.3 Zambian Cotton Production ............................................................................... 4 1.3 MAIZE SUPPORTS ....................................................................................................... 8 1.4 POTENTIALLY PROBLEMATIC RELATIONSHIP BETWEEN COTTON PRODUCTION AND MAIZE SUPPORTS ............................................................................................................. 9 1.5 PURPOSE OF THE STUDY ............................................................................................. 9 1.6 ORGANIZATION OF THE STUDY ................................................................................ 10 2. ZAMBIAN COTTON SECTOR ................................................................................... 12 2.1 INTRODUCTION......................................................................................................... 12 2.2 SPATIAL DISTRIBUTION OF PRODUCTION ................................................................. 13 2.3 COTTON SECTOR EVOLUTION SINCE LIBERALIZATION ............................................. 18 2.4 SMALLHOLDER COTTON DECISIONS TRACKED AND ANALYZED ACROSS HOUSEHOLD INDICATORS ................................................................................................................... 24 2.4.1 Drivers of Aggregate Production ..................................................................... 24 2.4.2 Characteristics of Households Entering and Exiting Cotton ........................... 26 2.4.3 Crash and Recovery Households by Indicators ............................................... 38 2.5 SUMMARY OF KEY FINDINGS ................................................................................... 42 3. ZAMBIA‟S MAIZE SECTOR ...................................................................................... 44 3.1 INTRODUCTION......................................................................................................... 44 3.2 FARMER INPUT SUPPORT PROGRAM (FISP) ............................................................. 45 3.3 FOOD RESERVE AGENCY (FRA) .............................................................................. 52 3.4 SPATIAL DISTRIBUTION OF MAIZE PRODUCTION AND MAIZE SUPPORT VOLUMES... 55 3.5 HOUSEHOLD FRA SALES AND FISP RECEIPTS BY INDICATORS ............................... 60 3.6 EFFECTS OF MAIZE SUPPORTS .................................................................................. 66 3.7 SUMMARY OF MAIZE SECTOR AND GOVERNMENT SUPPORTS .................................. 68 4. CONCEPTUAL MODEL, ESTIMATION TECHNIQUES, AND RESULTS ............ 71 4.1 INTRODUCTION......................................................................................................... 71 4.2 CONCEPTUAL MODEL .............................................................................................. 71 4.3 DATA ....................................................................................................................... 74 4.3.1 Supplemental Survey Panel Data ..................................................................... 74 vi 4.3.2 Supplemental Survey Data Benefits ................................................................ 76 4.3.3 Post Harvest Surveys ....................................................................................... 77 4.3.4 Post Harvest Survey Data Benefits .................................................................. 78 4.4 MODEL SPECIFICATION ............................................................................................ 78 4.4.1 Cragg Hurdle Model and Partial Effects .......................................................... 80 4.5 ESTIMATION AND RESULTS ...................................................................................... 82 4.5.1 Model 1: SS Household level Panel Data for Harvest years 2003 and 2007 ... 82 4.5.2 Model 2: PHS SEA Level Panel Data for Harvest Years 2003 and 2006 ....... 93 4.6 EXPLORING THE VALIDITY OF THE POLICY VARIABLES .......................................... 102 5. CONCLUSIONS, POLICY IMPLICATIONS AND FURTHER RESEARCH ......... 105 5.1 CONCLUSIONS ........................................................................................................ 105 5.2 POLICY IMPLICATIONS ........................................................................................... 107 5.3 FURTHER RESEARCH .............................................................................................. 109 BIBLIOGRAPHY ........................................................................................................... 111 vii LIST OF TABLES TABLE 2.1: SHARE OF COTTON PRODUCTION AND % OF HOUSEHOLDS THAT PLANTED COTTON IN EACH PROVINCE ..................................................................................................................... 16 TABLE 2.2: COTTON GINNERY CAPACITIES AND THROUGHPUTS, HARVEST YEARS 2004/06 ........... 23 TABLE 2.3: PERCENTAGE OF SMALLHOLDER HOUSEHOLDS ENTERING AND EXITING COTTON BY YEAR, 1999-2007 ................................................................................................................... 28 TABLE 2.4: HOUSEHOLD INCOME INDICATORS FROM SS08 ACROSS THE NUMBER OF YEARS THAT A HOUSEHOLD PLANTED COTTON FROM 1997/98 TO 2006/07 ................................................... 31 TABLE 2.5: HOUSEHOLD INDICATORS FROM SS08 ACROSS THE NUMBER OF YEARS THAT A HOUSEHOLD PLANTED COTTON FROM 1997/98 TO 2006/07 ................................................... 32 TABLE 2.1: FRA AND FISP INDICATORS FOR THE 2002/03 AND 2006/07 HARVEST SEASONS ACROSS THE NUMBER OF YEARS THAT A HOUSEHOLD PLANTED COTTON FROM 1997/98 TO 2006/07 .. 33 TABLE 2.7: YIELDS AND AREAS PLANTED FOR DEDICATED COTTON GROWERS AND NON-DEDICATED COTTON GROWERS, 2003 AND 2007 HARVEST YEARS............................................................. 37 TABLE 2.8: SMALLHOLDER MOVEMENTS INTO AND OUT OF COTTON FOR CRASH YEARS, HARVEST YEARS 2000 AND 2007 ........................................................................................................... 41 TABLE 2.9: SMALLHOLDER MOVEMENTS INTO AND OUT OF COTTON FOR RECOVERY YEARS, HARVEST YEARS 2001 AND 2003............................................................................................ 41 TABLE 3.1: FISP DISTRIBUTION, GROWING SEASONS 2002/03 THROUGH 2009/10 ......................... 49 TABLE 3.2: FISP FERTILIZER RECEIPTS IN 2006/07, % OF HOUSEHOLDS PURCHASING FERTILIZER FROM PRIVATE DEALERS, AND % PRODUCING COTTON, BY DISTRICT IN COTTON PRODUCING PROVINCES ............................................................................................................................. 51 TABLE 3.3: FRA MAIZE PURCHASE QUANTITIES, MEAN FRA PRICES, AND MEDIAN MAIZE PRICES RECEIVED BY FARMERS .......................................................................................................... 55 TABLE 3.4: HOUSEHOLD INDICATORS BY SALE OF MAIZE TO FRA DURING 2006/07 CROPPING SEASON .................................................................................................................................. 61 TABLE 3.5: HOUSEHOLD INDICATORS BY RECEIPT OF FISP FERTILIZER IN 2006/07 CROPPING SEASON .................................................................................................................................. 62 TABLE 3.6: PERCENT OF SMALLHOLDERS GROWING MAIZE AND MEAN AREA OF MAIZE PLANTED FOR EACH PROVINCE FOR YEARS 2000 THROUGH 2007 ................................................................. 69 viii TABLE 4.1: VARIABLE NAMES AND DEFINITIONS ........................................................................... 83 TABLE 4.2: FIRST STAGE CONDITIONAL APES, CRE CRAGG ......................................................... 90 TABLE 4.3: SECOND STAGE CONDITIONAL APES, CRE CRAGG ..................................................... 90 TABLE 4.4: UNCONDITIONAL APES, CRE CRAGG ......................................................................... 92 TABLE 4.5: AVAILABILITY OF VARIABLES IN PHS ......................................................................... 95 TABLE 4.6: FIRST STAGE CONDITIONAL APES, FIXED EFFECTS CRAGG MODEL ............................. 99 TABLE 4.7: SECOND STAGE CONDITIONAL APES, FIXED EFFECTS CRAGG MODEL ....................... 100 TABLE 4.8: UNCONDITIONAL APES, FIXED EFFECTS CRAGG MODEL ......................................... 1002 TABLE 4.9: MAIZE CRE CRAGG MODEL RESULTS OF FRA AND FISP VARIABLES ....................... 104 ix LIST OF FIGURES FIGURE 1.1: ZAMBIAN NATIONAL POVERTY RATES ......................................................................... 1 FIGURE 1.2: ZAMBIAN SEED COTTON PRODUCTION, 1994 – 2010 (MT)............................................. 7 FIGURE 2.1: ZAMBIA'S AGRO-ECOLOGICAL ZONES ......................................................................... 15 FIGURE 2.2: COTTON PRODUCTION BY DISTRICT IN CENTRAL, EASTERN AND SOUTHERN PROVINCES FOR THE 2007 HARVEST YEAR (MT) ........................................................................................ 17 FIGURE 2.3: ZAMBIAN COTTON PRODUCTION, GINNERY ESTIMATES AND PHS ESTIMATES............. 19 FIGURE 2.4: COTTON MEAN AREAS PLANTED AND MEAN YIELDS AMONG GROWERS, 1993 - 2007 . 25 FIGURE 2.5: PERCENTAGE OF HHS GROWING COTTON AND TOTAL PRODUCTION, 1993 – 2007...... 26 FIGURE 3.1: NATIONAL SHARES OF MAIZE PRODUCTION, FRA MAIZE SALES, AND FISP FERTILIZER RECEIVED IN 2007 HARVEST YEAR BY PROVINCE ................................................................... 56 FIGURE 3.2: SPATIAL DISTRIBUTIONS OF MAIZE PRODUCTION, FRA MAIZE PURCHASES, AND FISP FERTILIZER, 2006/07 GROWING SEASON ................................................................................. 59 x LIST OF ABBREVIATIONS AEZ AMIC APE ATC CMB CRE CSO C4 DAC ESA FISP FRA FSRP ha HH HHH kg mt MACO PCO PHS SEA SS SS01 SS04 SS08 SSA US WCA Agro-Ecological Zone Agricultural Market Information Center Average Partial Effect Authority to Collect Cotton Marketing Board Correlated Random Effects Central Statistical Office (Zambia) Cotton Four District Agricultural Committee East and Southern Africa Farmer Input Support Program Food Reserve Agency Food Security Research Projects hectare(s) Household Household Head kilogram(s) metric ton(s) Ministry of Agriculture and Cooperatives Program Coordination Office Post Harvest Survey Standard Enumeration Area Supplemental Survey Supplemental Survey 2001 Supplemental Survey 2004 Supplemental Survey 2008 Sub-Saharan Africa United States West and Central Africa xi 1. INTRODUCTION 1.1 Background Information on Zambia Zambia is one of the poorest countries in the world by several measures. According to Zambia‟s Central Statistical Office (CSO), 64% of households were living below the country‟s poverty threshold in 2006 (Figure 1.1), down from a peak of 73% in 1998. According to CSO, most of this recent decline can be attributed to rural areas where the incidence of poverty decreased from 88% in 1991 to 78% in 2006. Figure 1.1: Zambian national poverty rates 76 74 72 70 68 66 64 62 60 58 1991 1993 1996 1998 2004 2006 For interpretation of the references to color in this and all other figures, readers are referred to the electronic version of this thesis. Source: Zambia Central Statistical Office (CSO), Living Conditions Monitoring Surveys While the recent decline in rural poverty is encouraging, 78% of rural households living in poverty is a striking number and in terms of per capita income Zambia was still ranked in the bottom thirty countries in the world (CIA, 2010). While contemporary poverty definitions extend beyond incomes and into health factors, water access and other measures, this study focuses on incomes and their direct contributions to household food security. 1 Most of these rural and impoverished people can be classified as smallholder farmers; that is, a large portion of their economic activity involves farming small areas of land, typically not larger than 2 hectares (about 5 acres). About 65% of Zambia‟s population was rural in 2005 (UN, 2009). These smallholder households, in order to meet their daily caloric consumption needs, must either harvest enough food from their own production or secure enough cash income throughout the year to consistently purchase food; most smallholder households obtain food in both ways. Some 57% of these households earned income from crop sales following the 2007 harvest (SS08), and others earned income by working on nearby farms for wages; so many rural households depend on agriculture both for their own food consumption and for their cash income. In Zambia, as in most of Sub-Saharan Africa (SSA), rural poverty, food security and agricultural productivity are inextricably linked. 1.2 The Importance of Cotton Africa as a whole has been losing ground with the rest of the world in agricultural productivity and trade. According to Tschirley et al (2009), from 1980 to 2005, Africa‟s share of total world agricultural trade dropped by about half. But hidden within this dire agricultural picture is a promising success: cotton. Over the same period when Africa‟s share of total agricultural trade was falling, the continent‟s share of cotton trade more than doubled as SSA cotton production grew three times faster than in the rest of the world. 1.2.1 Outgrower Schemes Nearly all of the cotton production in SSA is done by smallholder farmers, making the increasing cotton production in SSA a success for the rural poor. Growing cotton requires 2 quality seed and pesticide inputs, and very few Zambian smallholder farmers have enough cash in November (the typical planting time) to purchase from a dealer outright. Furthermore, as in nearly all of Africa, underdeveloped rural credit markets mean that smallholder farmers have almost no access to seasonal credit to finance input purchases. If smallholder farmers were forced to purchase their own inputs, few farmers would have the assets and income required to plant cotton. The high input costs would place cotton outside many households‟ choice set of crops to plant. Outgrower schemes typically channel the inputs from a processor to the smallholders prior to planting in return for the processor‟s right to purchase all of the crop output after harvest (often at a preset price) less the value of the inputs supplied earlier in the year. Essentially, the inputs are provided on loan at planting time and the loan is paid off when the crop is sold. Government involvement in the execution of the outgrower scheme and in the provision of inputs has varied by country and over time. For instance, some West and Central Africa (WCA) countries continue to utilize government managed parastatals while Zambia and most other countries in East and southern Africa (ESA) provide minimal government support, relying almost entirely on private cotton ginneries to provide inputs. A persistent problem with many outgrower schemes has been the ability to recover the value of the input loans and prevent side-selling (selling to a buyer that did not provide input credit to the farmer) of the output. Parastatal monopolies in WCA have done a good job of preventing side-selling, but Zambia‟s private led scheme has been susceptible to the problem. The outgrower scheme‟s success depends in large part on its ability to mitigate side-selling while promoting increased productivity and total production. 3 1.2.2 The “Cotton-4” Example Among SSA‟s cotton producing countries the “Cotton-4” (C4) in WCA has received almost all of the attention. The C4 consists of Mali, Benin, Burkina Faso and Chad; three of these countries (Mali, Benin and Burkina Faso) led SSA in cotton production from 2004 to 2008. The government managed parastatal cotton companies in the C4, enjoying near monopolies on ginning capacity and thus almost entirely able to control the side-selling problem, had used 1 cotton production over many years as a means of spurring rural development . Also, with their increased cash incomes from cotton sales, rural cotton farmers were able to invest in agricultural assets like better tools and animal traction which increased their yields and productivity in cotton and other crops as well. A well-functioning cotton sector can increase smallholder incomes, and, as shown by the example set in WCA, it can also contribute to reaching broader rural development goals that benefit all rural farmers, including those who do not grow cotton. The government led systems in WCA showed signs of weakness in the mid-1980s through early 1990s as the cotton sectors saw stagnant production volumes and financial problems engendered by high sector costs. The sectors have also experienced problems with parastatal management that have led to increased inefficiencies (Tschirley et al, 2009). Despite these difficulties, the C4 countries, and WCA as a whole, proved that a well-functioning cotton sector can lead to effective rural development. 1.2.3 Zambian Cotton Production While WCA‟s cotton production stagnated in the early 1990s, Zambia implemented more liberalized agricultural policies and opened up its cotton sector to competition. Over the next 1 Note that Burkina Faso allowed limited private entry in 2006 but had a single monopoly prior to that time. 4 decade, cotton‟s importance among agricultural commodities in Zambia grew significantly: nationwide output grew by over 1,000% in only eleven years (Figure 1.2). The development of the cotton industry in Zambia has the potential to increase rural smallholder incomes and improve household food security for some farmers. Tschirley and Kabwe (2007) found that the returns to a day‟s labor in cotton production for small plots of land was well above the rural wage rate in Zambia for the 2004/05 and 2006/07 cropping seasons. Furthermore, Govereh and Jayne (2003) found evidence in Zimbabwe that producing cotton brought benefits beyond increased incomes from crop sales in two main ways. First, the study found that cotton producers are able to obtain skills and key inputs that increase their productivity in other crops as well as cotton. Second, the study found that the presence of cash cropping and specifically cotton production brought increased investment to the region that benefited all farmers. Thus, cotton production can increase smallholder incomes and overall farming productivity among growers and act as a catalyst for rural development. National cotton production in Zambia has seen two periods of dramatic growth since 1994. As shown in Figure 1.2, both of these growth periods – 1994 to 1998 and 2000 to 2005 – saw aggregate seed cotton production volumes more than triple. While the overall story since 1994 is one of large and promising growth, the sector has faced several challenges that have contributed to production crashes in the 2000 and 2007 harvest years. These crashes and production instabilities are pressing concerns for Zambia‟s cotton sector. 5 Figure 1.2: Zambian seed cotton production, 1994 – 2010 (mt) * denotes forecast prediction of production Source: Cotton ginnery throughput estimates Causes of each crash can be categorized as internal or external to the sector. Internal sources of instability primarily include issues related to the outgrower scheme. The success of the outgrower scheme is directly related to cotton ginneries‟ ability to prevent side-selling of cotton output and recoup the value of the loaned inputs. As a result of increased farmer frustrations with decreasing cotton prices, poor transparency in price by cotton ginning companies, and widespread side-selling, which peaked during the 1998/99 growing season, rates of credit default among smallholder cotton growers increased. This led cotton ginneries to scale back the number of farmers that they contracted with and production collapsed in the 2000 harvest year (Govereh et al, 2000). One external source of instability was the considerable appreciation of the Zambian kwacha (ZMK) relative to the US dollar that occurred from late 2005, just after cotton planting, into mid-2006, when cotton was being purchased, processed, and readied for export. This 6 relative rise in currency value harmed Zambia‟s exports including cotton. Cotton ginneries were forced to lower their prices paid to farmers for the 2006 harvest and repayment rates dropped causing another production crash in the 2006/07 season. Another potential external challenge to cotton production involves the Zambian government‟s heavy emphasis on promoting domestic maize production. 1.3 Maize Supports While cotton has been somewhat neglected by policy makers, one agricultural commodity that has received no shortage of attention from the Zambian government is maize. Maize is the dietary staple in most of East and Southern Africa (ESA), and it is eaten at least once daily by nearly all Zambians. As a result of their people‟s strong preferences for maize, the agricultural policies of most ESA countries can be described as maize-centric. Zambia has recently proven its maize preferences by dramatically increasing its supports to maize production since the 2005 harvest season. The two largest support programs are government purchases of maize through the Food Reserve Agency (FRA), typically at prices above the market price, and subsidized maize seed and fertilizer through the Farmer Input Support Program (FISP) – formerly the Fertilizer Support Program (FSP). Both programs are large and costly: the FISP consistently accounts for about 35-40% of Zambia‟s agricultural budget annually (Xu et al, 2009). The FRA purchases several agricultural commodities each year, but the scale of their maize purchases dwarfs those of other commodities. The FRA purchased 390,000 metric tons of maize in the 2007 harvest year, which was over 30% of estimated smallholder maize production (PHS). In the 2008 harvest year, maize accounted for 98% of all planned FRA purchases. The 7 FRA purchases maize at an above market price and attempts to reach into areas with limited market access to buy from rural smallholders. The FISP tries to supply fertilizer and quality maize seed to smallholder farmers who would not otherwise be able to afford it, the idea being that increased maize production from fertilizer use would significantly contribute to household and national food security. Fertilizer subsidies affect maize production heavily because Zambian smallholders apply fertilizer almost exclusively to maize fields: in the 2006/07 growing season, over 90% of the fertilizer applied by smallholders went into maize fields (SS08). 1.4 Potentially Problematic Relationship between Cotton Production and Maize Supports Unfortunately, the maize supports do not always have their expected positive results. For instance, there is evidence that FISP fertilizer often does not reach its intended beneficiaries, being sold off to wealthier smallholder farmers (see chapter 3). Furthermore, Xu et al (2009) concluded that FISP fertilizer quantities received in areas with already strong fertilizer markets actually decrease the amount of fertilizer used in the area. Meanwhile, the execution of FRA maize purchases in rural areas can be very expensive and diverts funds and efforts away from other programs. While these unintended effects are fairly straightforward, there is perhaps another unintended consequence of Zambia‟s maize supports that is hidden from first glance. The FRA and FISP maize supports to smallholder farmers have the combined effect of increasing the profitability of maize production and they make maize relatively more appealing compared to other crop options. The hybrid seed varieties provided by FISP are more responsive to fertilizer than cotton and other crop options for Zambian smallholders, and it is widely understood that 8 fertilizer is predominantly applied to maize. Thus, when a household is faced with their decisions at planting time of what area to devote to maize, if said household expects to receive FISP fertilizer and hybrid seed or expects to be able to sell their maize output to the FRA at the higher price, then, ceteris paribus, they will most likely plant maize in a larger area than they otherwise would. This potentially increased maize area may come at the cost of lower areas planted in other crops. Crawford et al (2006) refer to this potential problem as it specifically relates to fertilizer promotion programs as an “inefficient substitution of crops towards those that use the subsidized fertilizer.” In Zambia, maize supports go well beyond fertilizer promotion, and could potentially magnify this problem. This opportunity cost is relevant to the once booming cotton sector as nearly all cotton is produced in Zambia‟s maize belt. It is possible that maize competes with cotton for smallholders‟ land areas, and, in some cases, the maize supports may contribute to farmers electing to plant more maize and less cotton. Thus, the strong government maize supports may be inadvertently damaging Zambia‟s private sector agricultural success, cotton, by luring farmers away from cotton production and into increased maize production. 1.5 Purpose of the Study The overall purpose of this research is to determine what the effects of the aforementioned FRA and FISP maize supports are on the private cotton sector. The paper has three specific objectives: i) identify what characteristics drive smallholder decisions to plant cotton, ii) evaluate the effects of FRA and FISP supports on smallholders‟ decisions to plant cotton or not, and iii) evaluate how the same programs affect areas planted in cotton among cotton growers. 9 If the Zambian government is going to continue promoting maize production among smallholder farmers, it is important that they have an understanding of the full costs of the policies. Such costs would include any adverse effects on the production of other crops; and the effects felt by the cotton sector, with its potential to increase smallholder incomes, are of particular importance. We meet the above mentioned objectives through quantitative descriptive analysis and empirical estimation of econometric models on panel data. The main strength of our econometric approach is that we study the proposed problems with two different models applied to two independent data sets: two separate Cragg hurdle models (1971) are used in this study. The first model uses two years of a household level panel data set and employs a correlated random effects (CRE) Cragg model. The second model uses two years of a household survey that is a panel at the standard enumeration area (SEA) level, and applies a SEA level fixed effects Cragg model to these data. A full discussion of our two approaches can be found in Chapter 4. 1.6 Organization of the Study This paper is organized in five chapters. Following this introduction, Chapter 2 analyzes Zambia‟s cotton sector in detail. It begins by looking at the spatial distribution of cotton production and continues by expanding upon the sector‟s structure and the evolution of production over time. It then tracks a sample of smallholder farmers and their production decisions over a ten year period and characterizes them across several household indicators. The chapter concludes with a summary of key findings. 10 Chapter 3 discusses operational structures of the Zambian government‟s major maize support programs, the FRA and the FISP. It then examines the spatial distribution of these supports along with maize production and it characterizes the households participating in these support programs by several household indicators. The chapter continues with an assessment of the effects of maize supports on farmers‟ maize cropping decisions, and concludes with a brief summary of findings. Chapter 4 explains the conceptual model used in the study and provides technical details on the Cragg hurdle model and its properties. It also discusses both of our econometric techniques in detail and describes the data that each uses. The results of each model are then presented and explained. Chapter 5 summarizes this paper‟s key findings and concludes with a discussion of implications for Zambia‟s agricultural policies. 11 2. ZAMBIAN COTTON SECTOR 2.1 Introduction This chapter discusses Zambian cotton production in detail. It begins by showing the spatial distribution of cotton production across Zambia‟s four Agro-ecological Zones (AEZ). Next, the chapter continues the discussion on Zambia‟s cotton sector structure found in the introduction to this paper and provides more detailed information on the sector‟s formation. Then it evaluates cotton production over time focusing specifically on production trends and instabilities. It concludes by summarizing key findings in the data analysis of cotton production and smallholder cotton producers. We use data from several sources in this chapter. The Post Harvest Survey (PHS) is an annual survey of more than 6,300 smallholder farmers executed after the growing season by Zambia‟s Central Statistical Office (CSO). We take weighted annual cotton production estimates from these data sets for the years 1993 through 2007. All years listed in this chapter are harvest years and signify the year in which the growing season ended and crops were harvested. We also use ginner estimates of cotton production as a cross reference against the PHS data. The two production estimates show the same trends and tell the same story of year-toyear production changes, but the ginner estimates are consistently larger than the PHS estimates. We take the ginner production estimates to be the most accurate for aggregate production and we use these data for discussion on total production levels. Because ginner estimates are difficult to allocate over space, we use the PHS data to discuss the spatial distribution of cotton production, including the percentage of smallholder farmers growing cotton. We also use a three year panel data set collected by CSO in cooperation with the Food Security Research Project (FSRP). The panel data questionnaires were given as a “Supplemental 12 Survey” to the PHS and were implemented in 2001 (SS01), 2004 (SS04) and 2008 (SS08), after attrition, 4,340 households were interviewed during each of these three years. The 2001 questionnaire retrieves information on households planting cotton for the harvest years 1998, 1999, 2000, and 2001: the 2004 questionnaire retrieves cotton planting information for 2002, 2003 and 2004: and the 2008 survey retrieves the same information for 2005, 2006 and 2007 harvest years. Together, these data make it possible to track 1,985 smallholder households in Eastern, Southern and Central provinces and whether or not they planted cotton each year from the 1998 harvest season through the 2007 harvest season. We use the panel data to analyze movement of specific households into and out-of cotton production across this ten year period and to characterize and compare the households that have steadily produced cotton over the years and the households that have not. We use both the panel data and the PHS data to discuss the agronomic practices employed by cotton farmers. 2.2 Spatial Distribution of Production To understand Zambia‟s cotton sector, it is useful to begin with a quick overview of the agronomic practices employed by cotton farmers. Cotton is grown in pure stands and is most frequently rotated with maize, although there are some fields in continuous cotton and others rotated with groundnuts. Smallholders apply almost no fertilizer or manure to their cotton fields; however cotton fields require additional attention and inputs in other ways. Over the growing season, cotton fields are weeded once more than maize fields on average. Additionally, cotton plants need to be sprayed with pesticides to control pests including aphids and Lepidoptera species. In 2006, cotton production required an estimated 110 days of labor while maize 13 required only 90 days (Tschirley and Kabwe, 2007). Despite the high labor costs of cotton relative to maize, a good cotton yield can be highly profitable for a smallholder household. A successful cotton yield requires fertile soils with the correct amount of water – not too much, but not too little – quality inputs – seeds and pesticides – and proper care during germination and growth. Households obtain the quality inputs and extension advice on farming practices through contract farming, which is discussed in the introductory chapter. In Zambia, rainfall quantities and soil types limit where cotton is grown. Cotton is bred to be drought tolerant and excess water and flooding damages the crop. Also, clay soils are better than sandy soils for cotton production. Figure 2.1 shows Zambia‟s AEZs. AEZ 1 and AEZ 2a are ideal for cotton production. AEZ 1 receives the right amount of rainfall to sustain drought tolerant cotton. AEZ 2a receives slightly more rainfall, but has clay soils which support cotton particularly well. AEZ 2b has the same annual rainfall as AEZ 2a, but has sandy soils that are poor for cotton production. AEZ 3 receives too much rainfall to support healthy cotton production although cotton is grown in some of the lower rainfall areas of the zone. The Central, Eastern, and Southern provinces make up most of AEZs 1 and 2a. It is no surprise, then, that these three provinces have accounted for more than 95% of the cotton production in Zambia annually since 1993 (PHS). Table 2.1 shows the percentage of smallholder households planting cotton and the share of nationwide cotton production for each province. In 2007, cotton was grown by only 10.8% of Zambian smallholder farmers, but in the three main cotton growing provinces, 23.1% of smallholders grew cotton and accounted for about 98% of all the cotton farmers in the country. 14 Figure 2.1: Zambia's agro-ecological zones Source: Reprinted from Nielson (2009) 15 Table 2.2: Share of cotton production and % of households that planted cotton in each province 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Production % Eastern of Total % Growing Province Cotton Production % Central of Total % Growing Province Cotton Production % Southern of Total % Growing Province Cotton Production % Lusaka of Total % Growing Province Cotton Production % Western of Total % Growing Province Cotton Production % Copperbelt of Total % Growing Province Cotton Production % Northwestern of Total % Growing Province Cotton Production % Northern of Total % Growing Province Cotton Production % Luapula of Total % Growing Province Cotton 66% 60% 62% 63% 65% 64% 67% 60% 63% 79% 31% 29% 19% 31% 38% 28% 42% 51% 43% 34% 23% 27% 32% 27% 23% 24% 20% 24% 23% 10% 18% 16% 7% 11% 17% 18% 21% 27% 26% 7% 9% 11% 5% 9% 10% 10% 11% 15% 13% 11% 7% 9% 3% 7% 12% 9% 12% 18% 18% 6% 1% 2% 0% 1% 2% 1% 2% 1% 1% 1% 3% 4% 0% 5% 7% 1% 6% 7% 4% 1% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 1% 1% 0% 0% 0% 0% 1% 1% 1% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% Source: Author‟s calculations from PHS data 16 Eastern province is the definite leader in cotton production. About 34% of smallholders in Eastern province grew cotton in the 2007 harvest season and produced 44,700 metric tons, which was over 78% of the national production for the season. Southern province ranked a distant second in production in 2007 with about 6,100 metric tons of cotton produced by 5.8% of its smallholder households. Central province ranked a close third with 5,530 metric tons produced and 7.3% of its smallholder households growing cotton. The remaining six provinces accounted for less than one percent of Zambia‟s total cotton production in 2007. Figure 2.2: Cotton production by district in Central, Eastern and Southern provinces for the 2007 harvest year (mt) 0 115,000 112,500 131,300 112,000 Cargill 60,000 48,976 44,196 42,023 Great Lakes 10,000 0 0 10,000 Alliance Cotton No data 0 0 8,000 Continental 25,000 5,000 7,000 8,000 Mulungushi 10,000 5,820 8,314 5,140 Chipata-China Cotton 15,000 0 No data 12,000 Ginnery Mukuba 500 43 113 24 Birchand Oil Mills 0 0 0 No data Total >215,000 172,339 190,923 197,187 Source: Tschirley and Kabwe, 2007 After the second production collapse, the sector showed signs of a rebound in 2008 as cotton production increased to around 100,000 metric tons. However, cotton harvest estimates for the 2009 and 2010 harvests were 110,000 metric tons and 120,000 metric tons respectively. The recent recoveries in aggregate cotton production have been modest relative to the recovery following the first sector crash. Part of the difference in the sector recoveries may be due to the fact that situations and conditions presented to farmers and cotton ginning companies have changed in recent years. One such change has been the recent increases in government led supports to smallholder farmers for maize production – discussed in detail in the next chapter. The future status of Zambia‟s cotton sector remains uncertain. It has shown some damaging instability with the two production crashes, but we can safely say that Zambian cotton 23 production has been a strong example of private led agricultural success in Zambia. History has shown us that future cotton production volumes will be influenced by several different factors, including; the number of cotton ginning companies in the market and their commitment to providing inputs and extension to smallholders, the effectiveness of the Cotton Board in regulating side-selling and fostering healthy competition, the outgrower scheme‟s reliability for both farmers and ginners, macroeconomic factors including exchange rates, and the level of government involvement in smallholder agriculture – specifically the cotton and maize sectors. 2.4 Smallholder Cotton Decisions Tracked and Analyzed Across Household Indicators This section discusses several key findings of our Zambian cotton research and is broken into three subsections. Section 3.3.1 analyzes changes in national production volumes to help determine what farm level factors drive changes in aggregate seed cotton production. Section 3.3.2 tracks the movements of smallholder households into and out-of cotton production by examining the differences in several household characteristics among the households that chose to plant cotton and the households that did not. Section 3.3.3 presents further information on the types of farmers that entered, stayed in, exited, and stayed out of cotton production during crash years – 2000 and 2007 – and recovery years – 2001 and 2003. 2.4.1 Drivers of Aggregate Production Overall levels of cotton production are directly affected by cotton yields, the area of cotton planted by each grower, and the number of farmers choosing to plant cotton. Figures 2.4 and 2.5 use PHS data to highlight these three determinants of production and identify which of them are most associated with aggregate production changes. 24 Figure 2.4: Cotton mean areas planted and mean yields among growers, 1993 - 2007 Source: Author‟s calculations from Post Harvest Surveys, PHS 92/93 – PHS 06/07 Figure 2.4 shows that mean household level cotton yields and areas planted have remained relatively stable over time. Figure 2.5 shows that same cannot be said for the number of farmers planting cotton, which has shown a good deal of variability. Figure 2.5 shows very similar movements in the percentage of smallholder farmers growing cotton and overall cotton production. This is confirmed by the high correlation (0.93) between the proportion of smallholders growing cotton and the national cotton production estimates and the low correlations between these estimates and both the mean area planted (0.26) and mean yield among growers (0.19). These patterns suggest that it may be important to examine the characteristics that distinguish smallholders that choose to remain in cotton even during crash years and those that choose to exit. 25 Figure 2.5: Percentage of HHs growing cotton and total production, 1993 – 2007 140,000 16% 120,000 Percentage 14% 100,000 12% 10% 80,000 8% 60,000 6% 40,000 4% 2% 20,000 0% Cotton Production (mt) 18% 0 Growing Season % of HH's Growing Cotton Total Production Estimate (mt) Source: Author‟s calculations from PHS data 2.4.2 Characteristics of Households Entering and Exiting Cotton A descriptive analysis of the Supplemental Survey panel data provides several insights into these issues. Table 2.3 shows the percentages of cotton farmers that moved into and out of cotton production each year from 1999 to 2007, by province. A household “exits” the market if said household grew cotton the previous growing season, but did not grow cotton this growing season. Likewise, a household “enters” the market if the household did not grow cotton in the previous growing season, but did grow cotton this season. Higher enter percentages than exit percentages are an indication of growth in the number of smallholders producing cotton and, therefore, total cotton production. In all but one year – 2000, the year of the first cotton collapse – the enter percentage was higher than the exit percentage in Eastern province. The percentages for Southern and Central provinces show much more volatility from year to year. The years 26 when total cotton production was increasing show large entry percentages and low exit percentages and the contraction years show the opposite. In 2001, Eastern province and Southern province had much higher entry rates than exit rates; and in 2003, all three provinces had entry rates that were much higher than their exit rates. These two years highlight the sector‟s strong recovery after the 2000 collapse. All in all, the Eastern province percentages for entry and exit are smaller than those for Central and Southern provinces, suggesting lower turnover among cotton farmers in Eastern province. This finding is consistent with PHS production data, which show less production volatility in Eastern province than in Central province and Southern province. These entry and exit percentages are useful for looking at the cotton sector as a whole, but they do not tell anything about the actual household decisions nor do they show the differences between smallholder households that grow cotton and those households that do not, and those that enter and exit at different times. 27 Table 2.4: Percentage of smallholder households entering and exiting cotton by year, 19992007 CENTRAL EASTERN SOUTHERN % Exit % Enter % Exit % Enter % Exit % Enter Harvest Year 1999 22.07% 26.70% 20.03% 37.58% 35.26% 48.46% 2000 27.28% 27.91% 31.04% 24.96% 53.62% 21.61% 2001 44.91% 32.37% 19.58% 29.50% 33.09% 60.97% 2002 No data No data No data 2003 14.66% 53.65% 29.78% 44.61% 33.70% 64.60% 2004 25.67% 21.51% 22.27% 22.65% 49.71% 36.56% 2005 59.24% 44.94% 42.85% 43.88% 63.86% 61.83% 2006 33.58% 54.27% 29.85% 34.34% 35.65% 52.15% 2007 58.46% 30.11% 23.45% 24.24% 62.95% 50.93% Note: “% Exit” is the percentage of growers from the previous year that did not grow cotton during the current year; “% Enter” is the percentage of growers during the current year that did not grow cotton during the previous year. Source: Author‟s calculations from Supplemental Survey panel data Tables 2.4, 2.5 and 2.6 use Supplemental Survey panel data to show the total number of seasons a household planted cotton in the past ten years along with household level indicators for 2007 (the last season for which the panel collected a full set of household information). Zero years planted cotton means the household did not report planting cotton in any of the ten years; about 56% of households in Eastern, Central and Southern provinces did not plant cotton over the recorded period; about 44% did plant cotton during at least one year. The distribution of the number of years a household planted cotton is as expected – highest at zero years grown and decreasing each year (except from eight to nine years) with ten years having the lowest number of observations. Only 3% of households planted cotton in every year and less than 10% grew cotton in eight or more of the reported years. For the rest of this analysis that group of households will be referred to as “dedicated cotton growers”. These dedicated cotton growers account for only about 30% of all households that planted cotton at some point over the ten year 28 period, which suggests that movement into and out-of cotton production over this time period was much more common than consistent cotton production every season. 29 Table 2.5: Household income indicators from SS08 across the number of years that a household planted cotton from 1997/98 to 2006/07 YEARS PRODUCED COTTON 0 1 2 3 4 5 6 7 8 9 10 # of Observations 990 194 134 116 112 94 92 68 61 75 49 Weighted # of Obs. 149,302 29,835 20,864 19,608 17,654 13,530 14,882 9,967 8,724 9,179 7,329 Per Capita Mean 723 604 700 388 660 546 685 496 535 1,044 926 HH Income Median 228 200 217 210 265 271 251 222 359 412 367 Mean 1,887 1,380 1,489 1,024 1,531 1,346 856 1,068 1,070 2,127 3,709 Cash Income Median 380 513 584 416 538 654 448 615 669 906 952 From Farm % 52.20% 64.33% 72.36% 75.14% 79.00% 87.64% 91.01% 95.15% 96.71% 99.54% 100% Sales Received Mean 2,260 3,686 2,168 1,075 1,700 1,738 1,500 1,179 1,143 4,581 3,044 Business Median 735 900 770 600 525 410 940 400 640 265 750 Income % 38.15% 34.90% 45.33% 30.91% 41.09% 32.69% 47.41% 33.91% 40.25% 45.94% 44.65% Received Mean 856 283 225 129 407 186 190 118 43 72 203 Ag-Sector Median 148 40 112 70 50 170 212 120 26 75 25 Wages % 12.63% 5.74% 7.81% 22.39% 20.83% 15.28% 14.16% 11.88% 15.39% 20.71% 10.06% Received Mean 4,014 3,703 2,441 1,582 2,457 1,858 4,638 7,756 4,180 3,737 614 Non-AgMedian 980 800 600 300 500 700 2,100 7,200 2,664 1,761 860 Sector % Wages 13.96% 8.86% 11.16% 12.80% 9.74% 13.05% 13.78% 6.07% 14.22% 8.79% 4.98% Received Note: All means and medians are for the portion of the sample that received the given income type; and all incomes are in „000 ZMK. Source: Author‟s calculations from supplemental survey data (SS08) 30 Table 2.5: Household indicators from SS08 across the number of years that a household planted cotton from 1997/98 to 2006/07 YEARS PRODUCED COTTON 0 1 2 3 4 5 6 7 8 9 10 Househol %Female 28.65% 22.14% 19.04% 15.80% 18.06% 15.94% 15.83% 15.74% 11.69% 11.71% 6.62% d Head Mean Education 4.91 5.13 5.34 4.91 5.27 4.85 4.90 4.90 5.43 4.27 4.14 (HHH) Median Ed. 5 5 6 6 5 6 5 6 6 4 4 Mean HH Size 4.65 4.88 4.63 4.57 4.81 4.91 5.23 5.14 4.96 4.90 5.07 HH Size Mean Labor and Labor 2.49 2.50 2.50 2.46 2.44 2.67 2.85 3.00 2.77 2.67 2.91 AE's Mean Value 5,348 5,404 4,663 3,247 3,447 6,081 3,454 3,594 3,946 8,826 7,297 Agricultu- Median Value 660 895 780 685 800 870 985 1,255 1,475 2,765 1,659 ral Assets % Own Animal 24.72% 30.15% 29.22% 22.35% 24.40% 31.85% 24.28% 23.24% 33.48% 39.84% 39.62% Traction Area Mean 2.94 2.94 3.03 2.62 2.98 2.76 3.30 3.07 3.15 5.66 4.75 Cultivated Median 1.56 1.82 2.06 1.91 2.19 2.03 1.94 2.84 2.63 3.21 3.18 Mean Area 1.27 1.45 1.53 1.14 1.33 1.30 1.13 1.23 1.20 1.75 1.68 Median Area 0.81 0.81 0.91 0.81 0.81 0.81 0.81 0.81 0.81 1.22 1.22 Maize Mean Yield 1,498 1,612 1,504 1,252 1,366 1,528 1,366 1,402 1,434 1,800 1,661 Median Yield 1,136 1,438 1,325 1,065 1,136 1,323 1,014 1,183 1,278 1,489 1,136 0 0.91 0.71 0.68 0.84 0.81 0.68 0.93 0.94 1.24 1.66 Mean Area 0 0.75 0.6075 0.405 0.81 0.625 0.5 0.81 0.81 0.81 0.81 Median Area Cotton Mean Yield . 909 843 902 869 1,056 938 841 884 1,191 1,013 . 700 672 731 667 938 800 668 790 956 840 Median Yield Note: All calculated areas are listed in hectares (ha); all yields are calculated as kilograms produced divided by area (kg / ha). Source: Author‟s calculations from supplemental survey data (SS08 31 Table 2.6: FRA and FISP indicators for the 2002/03 and 2006/07 harvest seasons across the number of years that a household planted cotton from 1997/98 to 2006/07 YEARS PRODUCED COTTON 3 4 5 6 7 0 1 2 Share of cotton 0 2.40% 6.55% 8.71% 12.68% 13.25% growers 2003 Cotton Share of cotton 0 1.40% 4.86% 6.41% 9.39% 10.46% Statistics production Share of cotton 0 1.86% 5.14% 14.10% 9.00% 12.67% area Share of cotton 0 3.18% 8.21% 10.90% 11.75% 11.13% growers 2007 Cotton Share of cotton 0 2.28% 4.50% 6.03% 8.36% 8.96% Statistics production Share of cotton 0 3.15% 5.63% 7.87% 10.75% 9.52% area 2002/03 0.60% 0.14% 0.00% 0.86% 1.45% 3.08% % Sold to FRA 2006/07 10.18% 14.60% 16.65% 10.43% 9.05% 11.95% 8 9 14.78% 10.52% 10.72% 10.95% 10 9.43% 12.25% 10.35% 11.33% 17.87% 15.67% 12.31% 9.28% 11.05% 13.49% 11.11% 13.73% 10.66% 9.45% 11.25% 8.82% 8.42% 7.69% 19.24% 25.70% 10.14% 10.57% 9.64% 15.18% 17.54% 0.13% 3.85% 5.05% 0.00% 16.50% 12.74% 11.55% 16.73% 4.50% 0.00% 9.73% Mean kg Sold to FRA 2002/03 6,884 9,488 - 2,013 575 1,387 2,070 - 639 518 - 2006/07 4,511 3,296 3,455 2,628 4,524 6,032 1,372 3,343 2,656 3,252 37,207 Median kg Sold to FRA 2002/03 6,030 14,070 - 2,013 575 1,150 2,070 - 1150 518 - 2006/07 1,925 1,668 1,150 2,530 3,220 3,680 575 1,725 1,495 3162.5 51750 %Received FISP Fert 2002/03 10.16% 14.36% 10.28% 13.28% 7.34% 9.78% 14.07% 14.28% 12.12% 22.60% 7.17% 2006/07 11.05% 13.87% 10.76% 10.82% 10.50% 16.20% 17.03% 14.92% 11.69% 11.88% 5.59% Mean kg Fert Received 2002/03 426 376 440 236 243 260 224 307 319 297 157 2006/07 368 243 443 244 331 274 216 413 366 621 312 Median kg Fert Received 2002/03 200 200 350 200 200 200 200 350 200 200 100 2006/07 300 200 400 200 400 200 200 200 400 400 400 Source: Author‟s calculations from supplemental survey data (SS08 and SS04) 32 These tables show several expected trends among cotton producers and those that did not produce cotton. Most of the variables show a general trend across the number of years cotton was planted, though with some inconsistent up or down movements in a few of the years. The percentage of households receiving cash income from crop sales and the percentage of households that are female headed (both based on 2007 data) are the exceptions. These percentages increase and decrease, respectively, almost monotonically across the number of years cotton was planted. Household per capita income and value of agricultural assets are particularly important statistics to this study, because they offer an idea of how well off a household is. Interestingly, mean per capita income and agricultural asset value are higher for smallholder households that never planted cotton than for all groups of households who cultivated cotton except for our dedicated growers group; these dedicated cotton growers have the highest median values of per capita income and agricultural asset holdings. Based on these two indicators, dedicated cotton growers are much better off than all other smallholder households. Not surprisingly, the higher incomes of the dedicated cotton growers appear to be driven by cash sales: because cotton is a cash crop and the outgrower scheme provides a direct line for sales, it is logical that dedicated cotton growers have high incomes from cash sales. Another expected outcome seen in Table 2.4 is that cotton farmers as a group, and dedicated cotton farmers in particular, earned much less in agricultural sector wages than households that did not plant cotton. This is an expected result because smallholders with limited land available for cultivation and low per capita incomes tend to work on nearby fields in return for cash; but cotton farming households as a whole had higher land areas in cultivation and dedicated growers had high per capita incomes, so they did not need to work for agricultural wages as often as non- 33 cotton growing households did. Also, because cotton production requires more time and attention than many alternative crops, cotton farmers are more likely to forgo wages in other sectors in favor of spending more time working on their fields. This is made apparent by the low mean non-agricultural sector wage earnings of dedicated cotton producers and the low percentage of households receiving this income. Perhaps more surprisingly, the dedicated cotton farmers have higher mean business incomes and are more likely to receive business income than most other smallholders. Median per capita incomes tell a different story than mean per capita incomes. Smallholders who did not grow cotton at all have a lower median per capita income than cotton growers despite having a higher mean per capita income. This discrepancy between mean and median per capita income is a result of households that did not plant cotton having the highest percentage of households in the lowest quintile of per capita income (23%), but also having several households with high incomes that pull the mean upward. In this case, the median value is more representative of the group, and we put more weight on its value for analysis purposes. Another interesting finding is that median asset values are much higher for cotton growers than for non-growers and increase with each additional year of cotton grown, yet the percentage of households owning animal traction is not meaningfully different among most cotton growing households and non-growing households. The group of dedicated cotton farmers, however, shows a much higher percentage of animal traction ownership than other smallholders. Median total land area cultivated is lower for non-growers than for every group of cotton growers. Mean values are higher for almost every year of cotton grown as well – only the households that grew cotton three and five years had lower means than the households that did 34 not grow cotton. Again, the dedicated cotton farmers show the largest means and medians. These growers also had much higher areas planted in both maize and cotton than other smallholders. Yield values show that dedicated cotton farmers also get the most out of their land. Mean and median cotton yields in 2007 were higher for dedicated cotton growers than for most other cotton growers, which suggests that experience and dedication to cotton matter in household level production. Mean and median maize yields vary greatly across groups and there is no observable pattern. However, a t-test revealed that dedicated cotton growers had a higher mean maize yield (1,701 kg/ha) than all other households (1,529 kg/ha) at the 5% significance level. Furthermore, these dedicated cotton producing households are not appreciably more likely than other cotton producing households to have sold some of their 2007 maize harvests to the FRA or to have received FISP fertilizer (observations based on a relatively small sample). This implies that the dedicated cotton growing households are not particularly better connected to the village power structures than other smallholders; while ability to sell to FRA may not be strongly influenced by local authorities, the FISP‟s allocation process (as described in chapter 2) appears to be more susceptible to favored access by households connected to village powers. An analysis of household characteristics reveals some interesting information as well. Most notably, the percentage of households that were female headed in 2007 decreases almost monotonically across the number of years cotton was grown, strongly suggesting that female headed households are less likely to plant cotton. The education level of the household head shows no strong relationship to the number of years cotton was planted, but to the extent that a trend does exist it appears that more dedicated cotton growing households have heads with lower education levels than less dedicated households. This finding is consistent with most research on 35 formal education and involvement in agriculture in Africa, which typically finds either no relationship or a slightly negative one. Previously in this chapter, we established that cotton production is labor intensive, so it is reasonable to expect a positive relationship between cotton cultivation and household size. The data provide some support for this expectation, with households that planted cotton in five or more seasons having more household labor, a mean of 2.98, than households that planted cotton in four or fewer seasons, a mean of 2.56 (significantly different at 1% by a t-test). To summarize across indicators, cotton production – in particular dedicated cultivation of eight or more years over the past 10 – appears to be associated with higher household per capita incomes and agricultural asset values. This higher level of economic wellbeing is achieved despite possibly lower levels of education and no apparent greater access to government maize supports than other smallholder households. While nearly half of all smallholder households in Eastern, Southern and Central provinces planted cotton in at least one of the ten recorded growing seasons, only about 30% of these cotton cultivating households – less than 10% of all households -- chose or were able to plant cotton in eight or more of the 10 seasons. Smallholder cotton production appears to be represented better by entrance to and exit from cotton production than by consistent and dedicated cotton cultivation. Comparison of the cotton statistics from 2003 (a recovery year) and 2007 (a crash year; see bottom portion of Table 2.5) brings to light some interesting information. In 2007, dedicated cotton growers accounted for about 53% of Zambia‟s cotton production and 42% of its area, while in 2003 these same farmers generated only 45% of production and 36% of area. The difference is even more pronounced when looking only at households that planted cotton in all ten of the analyzed seasons. These farmers constitute about the same share of all cotton growing 36 farmers each year – 9.4% in 2003 and 9.7% in 2007 -- but in 2007 their shares of total cotton production and area planted increased by 64% (from 15% to 25%) and 58% (from 11% to 18%), respectively. These increased shares for dedicated growers could come from i) an increase in their productions or areas planted from 2003 to 2007, ii) a decrease in productions and areas planted by non-dedicated growers, or iii) a combination of the two. Table 2.7 helps clarify the situation. Table 2.7: Yields and areas planted for dedicated cotton growers and non-dedicated cotton growers, 2003 and 2007 harvest years Non-dedicated growers Dedicated growers Mean Area Planted (ha) 0.88 1.05 Median Area Planted (ha) 0.66 0.87 Mean Yield (kg/ha) 1054.33 1246.86 Median Yield (kg/ha) 848.81 1069.96 0.79 0.64 1.28 0.81 908.06 1029.49 2003 Cotton 2007 Cotton Mean Area Planted (ha) Median Area Planted (ha) Mean Yield (kg/ha) Median Yield (kg/ha) 739.35 Source: Author‟s calculations from supplemental survey data (SS08 and SS04) 861.89 Mean and median yields decreased substantially for both groups from 2003 to 2007. When considering that 2007 was the largest crash in aggregate production, the decreased yields are not surprising. The group of non-dedicated growers had a lower mean and median area planted in cotton in 2007 than they did in 2003, whereas the dedicated growers had a much higher mean area planted in 2007. We find that the increased shares of cotton production and 37 areas planted for dedicated cotton growers resulted from a decrease in areas planted by nondedicated growers and an increase in areas planted by some of the dedicated growers. This suggests that the 2007 crash not only saw smallholders move out of production, it saw nondedicated households broadly decrease their areas planted to cotton while some of the dedicated growers increased their plantings. 2.4.3 Crash and Recovery Households by Indicators We now continue our analysis of household level indicators, but we turn our attention directly to crash years and recovery years and the differences between them. We have already demonstrated the instability of cotton production (Figure 2.3), and have shown that aggregate production movements closely follow the percentage of smallholders deciding to plant cotton in a given year; so we are directly interested in the differences across indicators of households that moved into and out of cotton production during crash years and recovery years. Table 2.8 uses data from the 2001 and 2007 supplemental surveys to show household indicators for two crash years, 2000 and 2007, across four cotton planting groups: i) households that stayed out of cotton production during the crash year (i.e., did not produce during the crash year nor the previous year), ii) households that exited cotton production during the crash year, iii) households that stayed in cotton production during the crash, and iv) households that entered cotton production during the crash year. Table 2.9 uses the 2001 and 2004 supplemental survey data and the same indicators and planting groups as Table 2.8 to analyze two “recovery” years, 2001 and 2003, in which cotton production increased following the 2000 crash. All income and asset values are nominal, so absolute comparisons can only be made within each year, but relative observations 38 across years are valid. Four important observations regarding the quality of farmers entering and exiting cotton production during crash and recovery years stand out from these tables First, in the 2000 crash, the households that stayed in cotton had substantially higher median asset values than other groups and they also had much higher mean and median land areas, which might imply that they are the better endowed or more capable farmers. The contrast between the group of households that stayed in production and the group of households that exited production is large in terms of median asset values and land areas: households that were able to stay in cotton were clearly the best endowed in these two categories. Second, in the 2007 crash, the differences between households that exited and those that stayed in cotton are less striking and we can no longer make the assertion that the households that stayed in production were the best endowed. The group that exited cotton had higher mean and median asset values than the group that stayed in, and they had only slightly lower land areas and median incomes. When this observation is considered along with the first observation it becomes evident that more households with better endowments left cotton production during the 2007 crash than during the 2000 crash. It can then be implied that the 2007 crash was more harmful and severe as some of the most capable cotton farmers were driven from producing the crop, an implication that is consistent with the fact that cotton production fell more precipitously in 2007 than it did in 2000 (Table 2.3). Third, to contrast crash years with recovery years, we focus on 2000 and 2001 because of their temporal proximity and because both years of data were obtained in the same data set (SS01) and therefore the asset and income values are on the same nominal terms. In 2001 (the recovery year), households with lower land endowments and asset values relative to 2000 were able to remain in cotton production, which suggests that cotton production was seen as profitable 39 Table 2.7: Smallholder movements into and out of cotton for CRASH years, harvest years 2000 and 2007 2000 Stayed Out Exited Stayed In 77.14% 6.09% 12.54% 181,389 138,737 196,341 50,000 53,333 128,564 Entered 4.23% 169,225 83,116 Stayed Out 65.35% 686,741 211,857 mean value 677,500 944,550 1,165,320 524,785 median value 66,000 64,667 246,333 56,667 Assets % owning AT No data mean land area 2.43 2.56 3.63 3 median land area 1.62 1.82 2.7 1.94 mean # years 1.01 4.33 7.62 4.62 Cotton median # years 0 4 8 5 mean 2.89 3.03 3.17 2.8 Labor AEs median 2 2 2.83 2 Note: Incomes and assets are nominal values. Source: Author‟s calculations from supplemental survey data (SS04 and SS01) 5,145,001 650,000 24.68% 2.89 1.62 0.6 0 2.49 2 % of total hhs mean Income Per Capita median 40 2007 Exited Stayed In 9.20% 18.53% 762,023 624,333 240,000 297,483 5,346,134 1,390,000 34.43% 3.28 2.19 3.46 3 2.42 2 5,169,204 1,260,000 28.91% 3.67 2.63 6.46 6 2.79 2 Entered 6.91% 519,523 262,620 3,299,071 935,000 32.06% 2.81 2.06 4.19 4 2.59 2 Table 2.8: Smallholder movements into and out of cotton for RECOVERY years, harvest years 2001 and 2003 2001 Stayed Out Exited Stayed In Entered % of hhs 77.00% 4.11% 12.71% 6.18% mean 184,093 132,361 210,020 94,456 Income Per Capita median 51,333 78,000 128,564 42,057 mean value 690,115 849,208 1,029,449 733,121 median value 66,000 120,000 205,000 60,000 Assets % owning AT No data mean land area 2.37 3.88 3.35 3.39 median land area 1.62 2.12 2.495 1.82 mean # years 0.97 4.52 7.62 4.86 Cotton median # years 0 5 8 5 mean 2.91 2.93 3.12 2.88 Labor AEs median 2 2 2.83 2 Note: Incomes and assets are nominal values. Source: Author‟s calculations from supplemental survey data (SS04 and SS01) 41 2003 Stayed Out Exited Stayed In 69.25% 4.74% 16.96% 600,000 316,878 602,808 166,800 99,553 317,533 3,050,334 1,959,644 2,499,901 415,000 462,000 640,000 20.22% 23.31% 26.62% 2.09 2.22 2.84 1.59 1.68 2.25 0.64 3.78 6.76 0 4 7 2.28 2.23 2.45 2 2 2 Entered 9.05% 424,765 217,667 2,516,046 580,000 24.43% 2.86 2.03 4.48 4 2.3 2 for households with smaller land areas and that cotton ginneries were willing to continue their relationships with these smaller households during the 2001 recovery. Fourth, the group of households that exited cotton production in 2000 is similar to the group that entered cotton production in 2001. These groups have comparable median asset values and land areas as well as similar means for the number of years households planted cotton. These observations suggest that the 2000 crash did not have long-run impacts on the quality of farmers that planted cotton. We cannot explore similar relationships following the 2007 crash because we do not have data for the 2008 harvest year. 2.5 Summary of Key Findings This chapter has discussed several important findings in Zambia‟s cotton sector. A few key points are highlighted and briefly summarized in this section. First, Zambia‟s cotton sector has been characterized by tremendous, though unstable, growth since liberalization in 1994. The concentrated sector has implemented the outgrower schemes successfully for the most part, but twice low input loan repayment rates contributed to production crashes. Low repayment rates prior to the second crash were engendered by the appreciation of the kwacha relative to the US dollar. Ginneries were forced to pay prices below their announced pre-planting prices and repayment rates fell again, followed by a production crash the following season as ginneries were more selective in their input provisions on loans. A second important finding is that aggregate cotton production volumes move very closely with changes in the percentage of households that plant cotton in any given year, and much less with mean household yields and areas planted. This emphasizes the importance of households entering and exiting cotton production each year. Eastern province has shown 42 relatively stable enter and exit percentages over time while Central and Southern provinces have shown more variability in these percentages. Also, households that entered cotton production for the 2001 harvest year were similar across indicators to the households that exited cotton production during the 2000 crash, implying that the sector‟s first crash did not have persistent effects on the quality of farmers that chose to plant cotton. The 2007 crash saw relatively higher quality farmers exit cotton production. This fact coupled with the anemic recovery in aggregate cotton production in 2008 and 2009 raises serious questions about the health of the cotton sector and its future production growth. Third, we find that the increased shares of cotton production and areas planted for dedicated cotton growers from the 2003 harvest year to the 2007 harvest year resulted from a decrease in areas planted by non-dedicated growers and an increase in areas planted by some dedicated growers (Table 2.5). This suggests that not only did smallholders move out of cotton production in 2007, but that non-dedicated growers that still chose to plant cotton, and even some of the dedicated cotton growers reduced their field sizes. Lastly, dedicated cotton cultivators – those households that planted cotton in eight or more of the ten seasons – had much higher per capita incomes (at least partially explained by high incomes from cash sales), asset values, and land areas under cultivation than the other household groups. These dedicated cotton growers also achieved significantly higher maize yields than non-dedicated growers but do not appear to be any more likely to have sold maize to the FRA or to have received FISP fertilizer than the other groups. 43 3. ZAMBIA’S MAIZE SECTOR 3.1 Introduction This chapter begins by expanding upon maize‟s importance as a food crop in Zambia. It continues by discussing the Zambian government‟s involvement with the maize sector, as two smallholder maize cultivation support programs – the Farmer Input Support Program (FISP) and maize purchases by the Food Reserve Agency (FRA) – are explained in detail. Then it analyzes the spatial distribution of smallholder maize production, FRA maize purchases and FISP fertilizer allocation. It continues by analyzing indicators across groups of households that sold different quantities of maize to the FRA and received different volumes of FISP fertilizer. The chapter then looks at the association between trends in maize supports and smallholders‟ cropping decisions, and concludes with a summary of key points. As mentioned in the introductory chapter, Zambia is accurately described as a maizecentric country. Maize meal is the food staple of choice for nearly all Zambians. Consumers consider breakfast meal to be the highest quality meal as it eliminates all the germ and pericarp, keeping only the starchy portion of maize kernels. Roller meal is considered to be of lower quality, being less refined and including some of the germ along with the starchy portion of maize kernels. Breakfast meal is preferred for its taste, and according to Agricultural Marketing Information Centre (AMIC) data from 1994 to 2009, is on average thirty percent more expensive than roller meal. Consumers‟ preference for maize is echoed in production patterns, with maize being the primary crop grown among Zambian smallholder farmers. In Central, Eastern and Southern provinces, where the soil and water conditions are particularly favorable – and where 98% of all cotton is produced – 97% of smallholder farmers grew maize in the 2006/07 growing season. 44 The average smallholder household in this region devoted over 76% of their available farmland to maize. Due to maize‟s direct influence on the livelihoods of most Zambian smallholder farmers, the national government takes an interest in the crop‟s production. Smallholder maize production has been supported through several different programs since Zambia achieved independence from the United Kingdom in 1964. In the 1990s, Zambia‟s agricultural policies became more liberalized and government financed supports to maize production were reduced but not abandoned. Since the 2005/06 growing season, the Farmer Input Support Program (FISP) and the Food Reserve Agency (FRA) have been expanded and are currently the leading agricultural supports in the country. Detailed explanations of these programs follow. 3.2 Farmer Input Support Program (FISP) Most countries in SSA attempt to spur rural agricultural productivity and increase food security through programs that promote fertilizer use among smallholder farmers. While there has been considerable variation in the implementation details across time and countries, the general framework for these programs includes subsidized fertilizer prices and some level of government involvement in distribution, ranging from full distribution authority to distribution supports to private retailers through rural credit supports. The economic benefits of these programs, especially when other possible agricultural supports are considered, vary greatly across program designs (Crawford et al 2006). Zambia has employed several different fertilizer promotion programs since independence in 1964. The post-colonial Government of Zambia managed the import, distribution, and pricing of fertilizer through NAMBOARD, the state marketing board, until the early 1990s, at which 45 time the government relinquished its managerial position in the industry and allowed private sector participation (Jayne et al 2002). The government‟s reduced involvement in the fertilizer industry was one of many policy changes made to liberalize Zambia‟s agricultural sector. This liberalization emerged in response to a number of factors: an upsurge in ideological commitment in western (donor) countries and multi-lateral aid organizations to “free market” solutions, political pressures brought on by the government‟s broken promises of low maize meal prices to consumers, and macroeconomic mismanagement in the 1980s including large external debts and currency overvaluations (Pletcher, 2000). Since this period of liberalization, however, the Zambian government has continued to play a significant role in the fertilizer industry with fertilizer subsidies and supports to smallholder farmers. The government‟s continued efforts to support fertilizer usage are not necessarily misguided, as research by Deininger and Olinto (2000) found strong evidence that increasing the number Zambian smallholders applying fertilizer to their maize fields would significantly increase maize production and that fertilizer application in small doses would be profitable for many farmers. Donovan et al (2002) found fertilizer application to maize fields could be quite profitable for some smallholders, but they also found that maize fertilizer response rates were highly variable in Zambia. Throughout the 1990s the government‟s involvement in fertilizer distribution varied as several different distribution methods were employed. One program of meaningful size was FRA loans used to make fertilizer obtainable for smallholder farmers. The FRA fertilizer loan program distributed fertilizer to smallholders for the 1996/97 and 1997/98 growing seasons. Although the program distributed more fertilizer than the private sector for both of these seasons (Jayne et al, 2002), the program faced several challenges including high costs of distribution and 46 very low loan recovery rates. These problems made the program expensive and unsustainable, and the fertilizer distribution function of the FRA was eliminated for the 1998/99 season. The problem of low loan recovery rates was not unique to the FRA program: even private agents contracted by the government to distribute fertilizer never achieved a loan recovery rate greater than 43% prior to 2002 (Govereh et al, 2002). In the 2002/03 growing season, the government heavily involved itself in fertilizer distribution once again with the implementation of the Fertilizer Support Program (FSP). The program‟s name was changed prior to the 2009/10 growing season to the Farmer Input Support Program (FISP). For consistency purposes, this paper retroactively applies the program name FISP to all FSP activities. The FISP is designed to allow smallholder farmers to purchase packets containing maize seed and fertilizer at a price equal to or less than 50% of the market price. FISP packets sold to smallholders contain inputs calculated for farming one hectare of maize. A few of the explicit FISP purposes as listed in their annual reports are to increase private sector participation in input provision, ensure timely input delivery, improve farmer access to inputs, and break the monopolies of input provision. The program has been expensive, consistently accounting for 35-40% of the public budget to agriculture (Xu et al, 2009). Table 3.1 shows the changes in FISP fertilizer quantities distributed and the number of intended recipients of FISP fertilizer for every growing season since the program was initiated. Only the 2004/05 and 2007/08 growing seasons had lower fertilizer quantities and fewer intended recipients than the previous year. That is to say the FISP has typically been increasing its distribution since inception. Since 2007/08, the distribution increases have been particularly striking. From 2007/08 to 2009/10 the number of intended beneficiaries increased by over 340% and the tonnage of subsidized fertilizer distributed increased by 120%. 47 Table 9: FISP distribution, growing seasons 2002/03 through 2009/10 Number of Intended Total Fertilizer Growing Season Subsidy (%) Recipients Distributed (mt) 2002/03 120,000 2003/04 150,002 2004/05 75,000 2005/06 125,000 2006/07 145,375 2007/08 120,250 2008/09 200,000 2009/10 534,190 Source: MACO FISP manuals 48,000 60,000 30,000 50,000 58,000 48,500 80,000 106,838 50% 50% 50% 50% 60% 60% 75% NA The FISP‟s structure for distributing subsidized fertilizer and maize seed was changed following the 2008/09 growing season. We do not wish to ignore the current FISP fertilizer distribution techniques, but because this study is conducted using data collected prior to the 2009/10 growing season we directly focus on the FISP framework used prior to that season. In principle, farmers wanting FISP fertilizer must complete an extensive application process. The farmer must be a member of a cooperative or a farmer organization and they must apply to receive FISP fertilizer through their respective organization – individuals cannot apply by themselves and farmers must have the ability to farm between one and five hectares for their application to be accepted. The regional cooperatives and farmer organizations must then submit 3 an application to the District Agricultural Committee (DAC) on behalf of their members. The DAC reviews the applications and reports their desired fertilizer distributions to the Program Coordination Office (PCO). With all the application information from the DACs, the PCO 3 The DAC was changed to the more localized Camp Agricultural Committee (CAC) in 2009 with the intent of ensuring more accurate distribution of fertilizer. Each CAC consists of seven community members, one from each of the following sectors or groups; church, NGO, civil servant, chief representative, youth, agricultural cooperative, and MACO. The CAC verifies the names on the cooperative applications and must be present when fertilizer is handed out. 48 determines who should receive subsidized fertilizer. When a cooperative learns that some of its members will be receiving FISP fertilizer, they are responsible for depositing these members‟ portion of the fertilizer costs (usually 50%, but it varies year to year) into a bank account set up to make payments to the private fertilizer distributors. The government will then match their portion of the payment in the same account. When the full payment from both the farmer coops and the government is received, the bank transfers the funds to accounts held by the private traders who then begin distributing the fertilizer as it becomes available. As one of the explicit goals of FISP is to increase private sector input provision, the government does not actively involve itself in fertilizer distribution; private fertilizer companies are selected for distribution to the districts. In order for farmers to receive fertilizer from the traders, they must first pick up their Authority to Collect (ATC) slips from the DAC. Then, when the fertilizer is distributed by the private retailers, the famers must take their ATCs and a form of identification to their district depots to receive their fertilizer in person. Each person is supposed to receive only one packet from FISP which contains enough maize seed and fertilizer for one hectare of maize. In principle, this design is meant to allow Zambian smallholder farmers who otherwise might not apply fertilizer to their fields to fertilize one hectare of maize and increase their yields; smallholders who already buy fertilizer to receive more fertilizer or decrease their input costs and increase their profits from maize sales; and private fertilizer distributors to benefit from the increased volume of fertilizer sales, lower their average fixed costs, and increase profits. But in the real life execution of FISP, there is evidence that these benefits are much smaller and in some cases nonexistent. Research by Xu et al (2009) found that in some cases FISP fertilizer can “crowd out” private fertilizer purchases. A World Bank (2009) study estimated that in the 2007/08 growing season FISP fertilizer displaced about 10% of private fertilizer sales. 49 Furthermore, Xu et al (2009) concluded that in places with an already strong fertilizer market an increase in FISP fertilizer received did more than displace private retailers, it resulted in a net decrease in total fertilizer used. The authors explain this result by suggesting that government announcements of planned FISP fertilizer distributions in an area might cause private fertilizer sellers to consider distributions to said area to be unprofitable and to discontinue their business activities there, which could decrease total fertilizer volumes sold and used in the area. Table 3.2 shows that the districts with the highest percentages of smallholders already purchasing fertilizer through private retailers have higher average FISP fertilizer receipts per household. The FISP‟s success in increasing rural productivity and alleviating poverty is sensitive to properly targeting “small farmers lacking effective demand” (Crawford et al, 2006). Unfortunately, another problem with FISP‟s implementation is that the intended recipients are not the actual recipients in many cases. In the 2005/06 growing season, FISP distributed packets intended for farming one hectare of land, which included 400 kg of fertilizer – 200 kg of basal 4 fertilizer and 200 kg of top dressing . Following FISP‟s implementation design, a farmer could not receive more than one packet and, therefore, no more than 400 kg of fertilizer. However, the PHS data for the same season suggest that 15% of households that obtained FISP fertilizer received more than 400 kg, and the mean and median kg received among them were 998 kg and 800 kg, respectively. This observation coupled with the information presented later in Table 3.5, which shows that wealthy farmers appear much more likely to receive FISP fertilizer than poorer farmers, suggest that actual distribution of FISP fertilizer differed from mandated distribution in ways that favored better-off farmers. 4 For the 2009/10 growing season, FISP reduced the intended packet size to 200 kg of total fertilizer per farmer – 100 kg of basal fertilizer and 100 kg of top dressing – with the direct intentions of reaching more famers and improving the distribution. 50 Table 3.2: FISP fertilizer receipts in 2006/07, % of households purchasing fertilizer from private dealers, and % producing cotton, by district in cotton producing provinces Mean FISP fertilizer % of HHs purchasing % of HHs District receipts across all fertilizer from private producing cotton household (kg) input dealer in 2003/04 Kabwe 112.2 57% 0% Kalomo 80.7 22% 4% Mazabuka 67.9 32% 20% Choma 66.2 27% 5% Mkushi 62.6 56% 0% Chibombo 53.5 56% 8% Mumbwa 52.5 25% 29% Kapiri mposhi 47.6 33% 8% Chadiza 42.5 30% 37% Monze 41.6 16% 14% Saivonga 40.0 7% 9% Chipata Petauke Lundazi Katete Serenje Gwembe Livingstone Itezhi tezhi Namwala Kazungula 39.5 38.7 35.7 35.4 26.1 24.0 13.3 10.7 8.7 7.7 35% 5% 24% 5% 13% 24% 16% 14% 9% 0% Nyimba 6.4 3% Chama 5.8 2% Mambwe 1.2 4% Sinazongwe 0.0 3% Source: Author‟s calculations from supplemental survey data (SS08) 30% 29% 38% 69% 0% 16% 0% 3% 13% 2% 30% 58% 46% 24% Yet another issue with the FISP‟s fertilizer distribution methods is late delivery. Several researchers, including Xu et al (2009) and Minde et al (2008), have emphasized that the timing of fertilizer application is very important, and that yields are maximized when fertilizer is applied “on time”. Estimates of the frequency of late delivery vary, but all are high: World Bank (2009) estimates that 70% of FISP fertilizer was delivered late during the 2007/08 growing 51 season, while CSO/MSU Supplemental Survey data for the same season indicate that 30% of FISP recipients reported that FISP fertilizer was not available when they needed it. To summarize, Zambia‟s government has tried to increase fertilizer use by smallholder farmers for several decades with different support programs. Most recently, the FISP has tried to increase smallholder access to cheap fertilizer nationwide. FISP volumes have grown considerably since the program‟s inception in the 2002/03 growing season and its plans for the 2009/10 season include more targeted smallholders and more fertilizer than ever before, with 534,190 intended recipients and about 107,000 metric tons of fertilizer to be distributed. Unfortunately, the program has shown evidence of some harmful unintended effects including side-selling of fertilizer, late fertilizer delivery, displacement of private fertilizer purchases and even less total fertilizer use in some areas with already strong fertilizer access. 3.3 Food Reserve Agency (FRA) The other main agricultural support program in Zambia is the Food Reserve Agency (FRA). FRA‟s explicit goal according to the agency‟s website is “to stabilize National Food Security and market prices [through] a sizeable and diverse National Strategic Food Reserve in Zambia by 2010.” FRA buys several agricultural commodities directly from rural farmers and stores them in the storage facilities that the agency manages. The primary commodity purchased by FRA is maize. For the 2007/08 growing season, maize accounted for over 98% of FRA‟s planned tonnage of agricultural purchases. FRA was founded in 1996 under the Food Reserve Act. The agency is operated by Zambia‟s Ministry of Agriculture and Cooperatives (MACO) and is funded through MACO‟s share of government funds. The Minister of MACO appoints an advisory board that oversees the 52 management of the program. The managers work with district agricultural cooperatives who communicate with the smaller, primary cooperatives of which smallholder farmers are members. Information is passed up through the chain of command, but the pertinent decisions (i.e., targets for commodity prices and quantities) are made by the advisory board. For the 2009/10 growing season, FRA planned about seven purchasing depots in each district for a total of 469 nationwide. Smallholder farmers can sell their maize to FRA agents at these depots at a fixed, previously announced FRA price, which in most years has been well above the open market price (Table 3.3). The maize can be delivered by any individual smallholder or by a member of any cooperative or a farmer‟s association. FRA buys maize from June through September each year or until funds for purchases run out. In four of the past five growing seasons the maize purchase period was extended through December as additional funds became available. This timing is not an issue for most farmers because 76% of all maize sales by Zambian smallholders in the 2006/07 harvest season were made in July, August and September. Each maize transaction must be a minimum of 10 bags of 50 kg and a maximum of 153 bags of 50 kg, and all bags must be free of “foreign matter”. Payment for maize is not immediate: farmers are expected to leave their maize at the depot and trust that their payment will come later. The Food Reserve Act was amended in 2005 to include the functions of distributing wealth to rural farmers and providing market access to farmers in remote areas. In the years following these amendments, FRA has dramatically increased the tonnage of maize it purchases from smallholder farmers (Table 3.3). Table 3.3 shows that the FRA began by purchasing moderate quantities of maize at modest prices during the 1996 and 1997 harvest years. The following four years saw no FRA maize 53 purchases due to a lack of funding for the program. Over this period of inactivity, Zambia experienced annual inflation rates of more than 25% in 2001 and 2002, which contributed to the median price received by smallholder farmers doubling from under 13,000 kwacha per bag of 50 kg in 1999/00 to over 29,000 kwacha in 2001/02. When the FRA resumed maize purchases in 2001/02, it offered farmers 44,400 kwacha for their 50 kg bags of maize, which is the highest price offered through at least 2007 and about 15,000 kwacha more per bag than smallholders were receiving in sales to private traders. The FRA increased its purchase volumes by about 100% in each of the next two years and decreased their purchase price to within 5,000 kwacha of the market price. After a minor decrease following the Food Reserve Act amendment in the 2005 season, the FRA drastically increased their maize purchases to around 390,000 metric tons in each of the next two seasons. This increase is made even more significant by the fact that FRA maize purchases had exceeded 100,000 tons in a season only once before. The FRA allowed their maize price to grow slowly over this time from 36,000 kwacha in 2003/04 to 38,000 in 2006/07. In every year that the FRA has made maize purchases, the FRA purchase price for maize has been above the market price received by farmers. 54 Table 3.3: FRA maize purchase quantities, mean FRA prices, and median maize prices received by farmers Market Price Received FRA Maize Harvest Year FRA Price per 50 kg Bag by Farmers per 50 kg Purchases (mt) Bag 1996 10,500 11,800 No data 1997 4,989 7,880 7,700 1998 0 No purchases 13,250 1999 0 No purchases 13,800 2000 0 No purchases 12,550 2001 0 No purchases 20,400 2002 23,535 44,400 29,100 2003 54,847 30,000 25,150 2004 105,279 36,000 No data 2005 78,566 36,000 32,200 2006 389,510 37,000 29,450 2007 396,450 38,000 33,550 Note: FRA prices are national means. Market prices received by farmers are national medians reported in PHS. Source: FRA and author‟s calculations from PHS data 3.4 Spatial Distribution of Maize Production and Maize Support Volumes In this section, we look at the spatial distribution of smallholder maize production, FRA maize purchase quantities and FISP fertilizer allocation. Our analysis begins at the province level, but concludes at the district level. We use PHS data along with annual FISP implementation manuals produced by MACO as our data sources for this chapter. Maize is cultivated by smallholders throughout Zambia primarily for household consumption, although about 30% of households sold some of their maize production in 2007 (SS08). The southern and eastern areas of Zambia are best suited for maize production, while the northern and western regions of the country have higher annual rainfalls, less fertile soils, and less maize production. Referring back to Figure 2.1, AEZ 1 and AEZ 2a are relatively better 55 suited for maize production – as they are for cotton production -- compared with AEZ 3a and AEZ 3b. Figure 3.1: National shares of maize production, FRA maize sales, and FISP fertilizer received in 2007 harvest year by province Source: Author‟s calculations from PHS data (PHS 2007) Figure 3.1 shows that Central, Eastern and Southern provinces led the country in maize production in 2007; these provinces‟ smallholders harvested two thirds of Zambia‟s maize. Luapula, Lusaka, Northwestern, Western and Copperbelt provinces produced disproportionately small shares of the country‟s maize: these provinces combined for about one fifth of the national smallholder output. Northern province produced less than 10% of the country‟s smallholder maize in 2007, which places it in between the high and low share groups. Also shown in Figure 3.1 are the province shares of FRA maize purchases and FISP fertilizer received. The FRA purchase shares are close to the maize production shares for each 56 province. However, Central and Western provinces accounted for much smaller percentages of FRA sales than their production percentages, while Eastern province accounted for a disproportionately high percentage of FRA sales despite leading the country in maize production. FISP fertilizer distribution shares show more consistency with maize production shares than the FRA purchase shares do. For the largest three maize producing provinces, FISP fertilizer shares were lower than maize production shares. Conversely, all but one of the other six provinces had FISP fertilizer shares that were higher than their maize production shares. Although the largest maize producing provinces received the highest portions of FISP fertilizer, the relative ratios of production shares to FISP fertilizer shares seem to suggest that FISP fertilizer distribution is more deliberately designed to increase smallholder maize production in the provinces where production is lowest. This point is corroborated by district level data. In the 2006/07 growing season, all of the country‟s 72 districts received subsidized fertilizer through FISP. FRA made maize purchases in 64 districts, but was absent from 8 districts. The widespread reach of these two programs reflects the government‟s clear efforts to extend supports for maize production to most smallholders countrywide. However, as Figure 3.1 shows, the FRA and FISP supports remain heavily concentrated in Southern, Eastern and Central provinces. For this reason, and because these provinces are the dominant cotton producers, we focus on these three provinces. Figure 3.2 highlights the district level distribution of smallholder maize production in Central, Eastern and Southern provinces, and displays FRA maize purchases and FISP fertilizer distribution respectively in the same area. These maps show a great deal of overlap and consistency in the spatial distributions of maize production and maize supports. Maize production appears to more closely related to FISP fertilizer distributions than to FRA maize 57 purchases, and this is, in fact, the case in the cotton growing provinces and the country as a whole. In the cotton growing provinces, an analysis of district level maize production from SS08 data along with FISP fertilizer distribution quantities and FRA maize purchase quantities for the 2006/07 growing seasons revealed a stronger correlation between maize production and FISP fertilizer quantities (0.82) than between maize production and FRA purchase quantities (0.50). The correlation between FRA and FISP (0.59) is strong as well. The biggest inconsistency between the FRA maize purchases and maize production maps is in Central province where it seems that the western districts produced a lot of maize but evidently did not have a proportionate opportunity to sell their harvests to the FRA. The biggest difference between the FISP and maize production maps lies in Southern province. The FISP fertilizer map shows that districts either received large amounts or small amounts of fertilizer with no districts in between. The smallholder maize production map shows a more even distribution across districts with several districts in each of the three harvest quantity range. 58 Figure 3.2: Spatial distributions of maize production, FRA maize purchases, and FISP fertilizer, 2006/07 growing season X <= 5 X<10,000 5 < X <=12 10,000 0, otherwise Si = 0, Stage 2 Wit* = β2 x2t + ui 2 ui ~ N(0, σ ) (4.2b) where Wi = Wi* if Wi*> 0 and Si = 1, otherwise Wi = 0, Sit* is the latent variable of Sit which is the observed binary variable representing a household‟s decision to plant cotton or not. Wit* is the latent variable of Wit which is the observed continuous variable of actual land area planted in cotton. The subscripts i and t refer to the ith 79 household during time period t. β1 and β2 are vectors of estimated parameters from their respective variable vectors x1t and x2t . An additional assumption required for this two stage model is D(W*|S, x) = D(W*| x), which means that the processes that determine W* and S are independent (Wooldridge, forthcoming) or that there is no correlation between the error terms, ei and ui (Mather, Boughton, and Jayne, 2009). Furthermore, in the above equations, x1t and x2t need not contain the same set of explanatory variables. There are a few different partial effects that can be calculated out of this two stage model. The effects are differentiated by being “unconditional” or “conditional” on the binomial decision variable. Conditional partial effects are the partial effects of a variable, xj, in only one of the two stages; xj will have a different conditional partial effect in stage one and in stage two, if xj is included in both estimation stages. The unconditional partial effect of xj takes into account both stages of the model regardless of whether or not xj is in both stages or only one stage. In essence, the unconditional partial effects of xj examine xj‟s partial effect on the process as a whole, while the conditional partial effects of xj look only at xj‟s partial effect on each individual stage of the model. Following Burke‟s (2009) example, the conditional partial effect of xj in the first stage is: 80 PW 0 | x1 xj x1 1j 1 where β1j is the maximum likelihood estimated coefficient of xj from the probit, and (4.3) is the standard normal probability density function (pdf). The conditional partial effect of xj in the second stage is: (4.4) 5 where λ represents the inverse mills ratio , β2j is the estimated coefficient of xj from the truncated regression and is the estimated variance from the truncated regression. Calculating the unconditional partial effects from the two estimation stages is more complicated and can be expressed as a single equation with two parts: (4.5) 5 The inverse mills ratio (IMR) is the probability density function divided by the cumulative density function (pdf / cdf). 81 where is the cumulative density function. The average partial effects (APEs) are obtained by averaging xj‟s partial effects across all observations. Conditional and unconditional APEs are reported in our results. 4.5 Estimation and Results This study employs two separate Cragg hurdle models to address our research questions with more confidence. Both models follow the typical Cragg two stage estimation process, employing a probit regression for the first stage and a truncated normal regression for the second stage. Different panel data are used for the two models, and the actual estimation procedures are modified to meet the needs of the data and panel estimation techniques. These techniques and their respective results are explained in this section. 4.5.1 Model 1: SS Household level Panel Data for Harvest years 2003 and 2007 4.5.1.1 Variables The household level panel Cragg model uses the supplemental survey panel data discussed in section 4.3.1. However, there are a few discrepancies in the information obtained in the three panel years. SS01 differs from SS04 and SS08 in the way it identifies household members: the two latter implementations of the panel collect information on all household members, but SS01 collected detailed data for “adults” only (age 12 and up), making calculations of household size and dependency ratios difficult. SS01 also does not obtain information on ownership of animal traction and it calculates household asset values using a smaller set of productive assets. Because we view animal traction ownership and productive asset values as important factors in a household‟s planting decision, we use only SS04 and SS08 in our 82 estimations, providing data on harvest years 2003 and 2007. Table 4.1 shows our complete list of variables used in both stages of our first Cragg model along with a brief description of each. Table 4.1: Variable names and definitions Variable Definition tot_hect total cultivated land area (hectares) animal_trac mech_trac age_hd educ_hd fem_hd ownership of animal traction dummy ownership of mechanical traction dummy age of household head education of household head (years) dummy variable for female head of household dep_ratio (number of household members age<15, age>60) / (number of household members 15