Report No. 40166-ZM Zambia The Relevance of a Rules-Based Maize Marketing Policy An Experimental Case Study of Zambia December 18, 2007 Agriculture and Rural Development Unit Sustainable Development Department Country Department AFCS2 Africa Region Document of the World Bank ReportNo: 40166 - ZM The Relevance o f a Rules-Based Maize Marketing Policy: An Experimental Case Study ofZambia December 18,2007 Agriculture andRuralDevelopmentUnit SustainableDevelopmentDepartment CountryDepartmentAFCS2 Africa Region The Relevanceof a Rules-BasedMaizeMarketingPolicy: An ExperimentalCase Study of Zambia CONTENTS Page ACNOWLEDGMENTS ................................................................................. i ABSTRACT..... ......,,, ..,, .. . .,.,,,.,.,....,,..............,.....,,,,,,.,,...,.,,...,.,.....,...,,,,...ii .. Section 1.Introduction.. .....................................,,.....,...........,,,..........,.....,......1 Section 2. The Political Economy o f Maize Market Reform inZambia.. .... ..... ............2 , . ., Section 3. The Model and ExperimentalDesign.. .......,......,.......,...,,.,...,..................5 Section 4. Results of the Main Experiment.. ........................ .........12 Section 5. The Workshop Experiment.. ,.............................................................18 Section 6. Conclusion. ..,,,...,.,..., ..,..,.,....,,.....,.,,..,,.,.,,,,,,..,.,..,.,...,,.................20 References...............................................,.....,.....,.,....,.............................22 Appendix Appendix A. Calibration o fthe Demand Function., .,,..,.,.,.,.,...,,.,...,........................24 Appendix B. Instructions for the Experiment.,...., ., .. ,.,., ,.,.., ...,,....,......... .... 26 . . . . .. .. .. Appendix C. Results o f the Workshop Experiment.. ..............................................28 Tables Table 1 ThePayoffTables.............................................................................. . 9 Table2 GovernmentQuantity:Announcedvs . . Implemented.................................... 14 Table 3 MarketOutcomesfor Alternative GovernmentPolicyRegimes . ...................... -18 Table A.1 Annual Maize Supply andPriceEstimatesinZambia, 1994-2006 ..................24 Table C.1 Market 1:Decisions andPayoffs......................................................... 28 Table C.2 Market2: Decisions andPayoffs ......................................................... 28 Figures Figure 1 DiscretionaryPolicy:Trader andGovernmentSupply . ................................. 13 Figure2 PrecommitmentPolicy:Trader andGovernmentSupply . .............................. 14 Figure3.PrecommitmentPolicy:PrivateSector Supply.......................................... 16 Figure4.Trader SupplyunderAlternative GovernmentPolicyRegimes........................ 17 Figure5.FrequencyDistributionof TotalSupplybyPolicyRegime............................. 18 Acknowledgements The report was prepared under the guidance o f Michael Baxter (AFCS2), Richard G. Scobey (AFTQK), Frank K.Byamugisha(AFTAR), and Paavo Eliste (AFTAR). This report was prepared by Lars C. Moller (LCSPE), Klaus Abbink (Associate Professor, University o f Amsterdam, AFTAR consultant) and Thom J a p e (Professor, University o f Michigan, AFTAR consultant). Excellent and extensive advice was received from Julie Dana (ARD),and Xavier Gin6 (DEC), who were the peer reviewers for this report as well as Jos Verbeek (AFTP1) and Anke Reichhuber (AFTAR). The report also benefited greatly from the participation of high-level government representatives and private sector traders andmillers at the Zambia Maize Marketing Policy Dialogue held at Fringilla, Zambia on 10 March 2007. The workshop was organized jointly by the Ministry o f Agriculture and Cooperatives, Agricultural Consultative Forum (ACF) and the World Bank. Special thanks to Hyde Hantuba (ACF) for assistance in preparing and conducting the workshop. Thanks also to Si1 Boeve and Lucas Molleman (University of Amsterdam), andMasiyeNawiko (ACF) for assistanceinrunningthe experiments. Financial support from the World Bank Finnish-Norwegian TFESSD Trust Fund (Grant TF057365) is gratefully acknowledged. Similarly, the Multi Donor Trust Fund on Commodity Risk Management (Swiss Secretariat o f Economic Affairs and the Netherlands Ministry o f ForeignAffairs) also helped support this work. Meseret Kebede (AFTAR) and Korotimi Sylvie Traore (AFTAR) provided excellent assistance invarious aspects of the management and administration of the project, and the preparation of the final document. 1 Abstract A critical barrier to achieving food security and rural income growth in the `mixed' food marketing systems characterizing many Eastern and Southern African countries revolves around the way that governments and the private sector interact. In shortage years, governments may question the capacity o f the private sector to import maize, and thus arrange imports on their own to cover the shortfall. At the same time, traders' import decisions depend on their expectations regarding governments' response to food shortages. Social dilemmas can arise if traders are uncertain about future government behavior or lack trust in official announcements. This paper argues that well-functioning markets depend on transparent and predictable government behavior underpinned by mutual trust and cooperation. We report on an economic policy experiment based on a stylized model o f the Zambian maize market. The experiment facilitates a comparison between the current government policy o f discretionary interventionism and a rules- based policy in which the government precommits itself to a future course o f action. A simple precommitment rule can overcome the social dilemma by reducing the risk o f food crises and provide appropriate incentives for private traders' participation inthe market, thereby enhancing economic efficiency. Exploring mechanisms that can support more predictable and rules-based policy responses to food shortages may therefore be beneficial to the Government o f Zambia and to other governments inthe region. .. 11 TheRelevanceof a Rules-BasedMaizeMarketingPolicy: An ExperimentalCase Study of Zambia 1.Introduction Over the past several decades, the role o f markets in supporting national food security, price stability, and rural income growth has become widely recognized. In Eastern and Southern Africa, several different approaches have been pursued. Since colonial times up to the early 1 9 9 0 ~ ~ grain marketing systems were controlled by the state inefforts to stabilize food prices and promote production through integrated input delivery, farm credit and output markets. The performance o f these systems was highly varied, contributing to impressive smallholder grain production growth in some cases, and retarding agricultural sector growth in other cases (Rohrbach 1989, Howard and Mungoma 1997, Karanja 1997, J a p e and Jones 1997). Invirtually all cases, however, these state-led systems imposed massive costs on the treasury, contributing to the fiscal crises that compelledmost governments inthe region to adopt market reform measures inthe 1990s. These dynamics gave rise to the second type o f approach to grain marketing policy, commonly understood as `market liberalization'. Inmuch o f eastern and southern Africa, the liberalization process was markedby ostensible attempts to transfer critical marketing functions from the state to private traders, but where in reality governments retained a great deal o f discretionary influence over prices and supplies (Jayne et al. 2002, Goldsmith 2002). In most cases, the liberalization process has been marred by lack o f trust, cooperation and coordination between the private andpublic sectors. Neither o f these approaches - controlled markets or liberalization - as implemented have produced sustained rural income growth, nor have they successfully avoided periodic food crises. There i s therefore an emerging consensus that the status quo food marketing situation inmany African countries is not working and that new approaches will need to be found urgently. Perhaps the most critical barrier to achieving these objectives revolves around the way the private sector and the government relate to each other. Despite the widespread perception that food markets have been liberalized, government intervention is pervasive in Eastern and Southern Africa. The food marketing systems in countries such as Ethiopia, Kenya, Malawi, Zambia, and Zimbabwe are most accurately characterized as `mixed systems'. Foodprices and availability are highly politicized issues in the region and there is a widespread view that governments are responsible for ensuring people's access to food (Bratton and Mattes 2003). This kind o f political economy has often led to a social dilemma inwhich the private sector is reluctant to undertake certain marketing activities for fear o f government interveninginways that impose private sector losses. The resulting low private sector activity then forces government to intervene in order to achieve its social objectives in the market. Since the private sector tends to be more timely and efficient, this situation results in a welfare loss. However, much larger than these short-run welfare losses are the inhibiting effects o f uncertain government behavior on long-term investment and the overall development of the marketing system (North 1987, 1994). Strategic interaction between the public and private sector i s therefore an issue that fundamentally affects food security outcomes within these mixedmarketing systems. 1 This paper introduces a novel approach to analyze strategic interaction between government and private traders in food markets, based on the case o f Zambia. An economic experiment was designed based on a variation o f the Cournot-Stackelberg oligopoly model with parameters informed by real data wherever possible. A specific objective o f the experiment was to compare the current government policy o f discretionary interventionism with a rules-based policy inwhich the government pre-commits itself to a future course o f action with robust replicable data. Experimental sessions with the `real' maize market players in Zambia were also conducted, including government officials and private sector participants. These sessions were intended as a learningdevice to facilitate apolicy dialoguerather thanto collect generalizable data. The results o f the maize market experiment underscores the importance o f predictable and transparent rules for governing the state's involvement inmarkets, andhow such operations inthe market could reduce the risks o f a food crisis and enhance economic efficiency. Specifically, government pre-commitment to a fbture course o f action i s found to be theoretically and empirically superior to a discretionary policy in this particular model and experiment. The Government o f Zambia should therefore consider mechanisms which can help make maize market policy more predictable or rules-based inthe future. The remainder o f the paper i s structured as follows. Section 2 discusses the difficulties of implementing maize market reforms in Zambia. Section 3 presents the model and experimental design. Section 4 presents the results o f the main experiment with subjects drawn from outside the context o f the Zambian maize market. Section 5 discusses the outcome o f the experiment in which Zambian government officials and private sector participants took part. Section 6 gives examples o f how the government could practically implement the policy recommendations arising from the analysis. Section 7 concludes. 2. The PoliticalEconomy of MaizeMarket Reform inZambia The Government o f Zambia adopted maize marketing reforms as part o f loan conditionality agreements with the World Bank and IMF in the late 1980s while facing extreme fiscal pressure. However, starting in 1993 the government reversed some o f these reforms and progressively re- introduced a number o f measures to control food prices and supplies. By 1995, a new parastatal, the Food Reserve Agency (FRA), was formed to hold strategic food stocks. Since the early 2000s, the FRA has taken on many o f the activities formerly carried out by the marketing board o f the 1980s (Namboard), albeit on a smaller scale. While private trade has developed steadily since the early 199Os, the current market environment i s remarkably similar to that o f the late 1980s, when external donors were urging the government to curtail the activities o f the grain marketing board, open up the borders to regional trade, and rely more on the private sector to carry out grain marketing and trade. During the past five years, the Mwanawasa government has introducedprogressively greater state intervention infood marketing and trade. Why have successive governments inZambia, and elsewhere inthe region, tended not to pursue the market reform and liberalization agenda recommended by international development 2 agencies?' There are two possible explanations. The first i s that government objectives are varied, inherently political, and vulnerable to influence and capture by elites. As argued by Lopez (2003), the allocation o f public expenditures tends to be biased in favor o f private goods, such as input subsidies, that can be captured bypolitically influential groups and against the provision o fpublic goods that would improve the overall performance o f markets and thus have broad-based benefits for the poor. The political landscape inmuch o f Africa can also be described as being dominated by neo-patrimonial relationships, in which government commodity distribution is an important tool by which leaders maintain loyalty and patronage among rural leaders and their constituents (van de Walle 2001, Bird et a1 2003, Pletcher 2000). Evenwithout resorting to neo-patrimonial arguments, it i s clear that the prospect o f an upcoming election compels decisions by policy makers to be dominated by what can be achieved in the short term while the payoffs from most policy reforms accumulate over the long term. The second class o f explanations has to do with genuine government concern for the social welfare o f smallholders as well as urbandwellers. White maize is the strategic political crop in this region o f Africa. Maize became the cornerstone o f an implicit and sometimes explicit `social contract' that the post-independence governments made with the African majority to redress the neglect o f smallholder agriculture during the colonial period (Jayne and Jones 1997). The controlled marketing systems inherited by the new governments at independence were viewed as the ideal vehicle to implement this objective. The benefits o f market controls designed to produce rents for European farmers during the colonial period instilled the belief that the same system could also promote the welfare o f millions o f smallholders if it were simply expanded (Jenkins 1997). The social contract also incorporatedthe understandingthat governments were responsible for ensuring cheap food for the urbanpopulation. While the social contract approach achieved varying levels o f success inpromoting smallholder incomes and raising consumer welfare, a common result in all cases was an unsustainable drain on the treasuryS2The cost o f supporting smallholder production - through input subsidies, credit programs with low repayment rates, commodity pricing policies that subsidized transport costs for smallholders in remote areas, and the export o f surpluses at a loss - contributed to fiscal deficits and in some cases, macroeconomic instability. Under increasing budget pressure, international lenders gained leverage over domestic agricultural policy starting in the 1980s, which culminated in structural adjustment programs in each country (Jayne and Jones 1997). While structural adjustment i s commonly understood to be a decision that international lenders imposed on African governments, a more accurate characterization o f the process i s that some sort o f adjustment was unavoidable due to the mounting fiscal crises that the social contract policies were imposing on governments. Continuation o f the status quo policies was not an option in countries such as Malawi, Tanzania, Zambia, Zimbabwe, and Kenya, and in some of these countries, the controlled marketing systems had already broken down prior to `market liberalization' as parallel markets swiftly became the preferred channel for most farmers and consumers. 'Inrecentyears, this includes the following World Bank studies andreports: DelNinno et a1(2005), Deininger and Olinto (2000), Siege1andAlwang (2005), World Bank (2003, 2004 and 2006) and World Bank andIFRPRI(2005). RelevantUSAID-sponsoredwork include: Mwanaumoet al. (1997), Jayne andJones (1997), Jayne et al. (2002). 'To illustrate,by the late 1980s, Zambia's subsidies to the maize sector reached 17%of the nationalbudget(Howard andMungoma 1997). 3 The rise o f multi-party electoral processes in the early 1990s has, however, made it difficult for governments in these countries to withdraw from the `social contract' policies. Elections can be won or lost through policy tools to reward some farmers with higher prices and reward others with lower prices, and this is hardly unique to developing countries (Bates 1981, Bates and Kmeger 1993, Bratton and Mattes 2003, Sahley et al. 2005). Because they provide obvious demonstrations of support for millions o f small farmers and consumers, a retreat from the social contract policies exposes leaders to attack from opposition candidates (Sahley et al. 2005). For this reason, it remains difficult for leaders to publicly embrace grain market and trade liberalization, even as they accepted structural adjustment loans under conditionality agreements from international donors to reform their internal and external markets. And starting in the late 1990s, the transition o f the World Bank and other development partners from structural adjustment loans with ex-ante conditionality to direct budget support with ex-post conditionality made it easier for states to reinstate some elements o fthe social contract policies. By the early 2000s, grain marketing boards have once again become the dominant players inthe market in Kenya, Malawi, Zambia, and Zimbabwe (Jayne et al, 2002). Each o f these countries have a highlyunpredictable anddiscretionary approach to grain trade policy, commonly imposing sudden and unanticipated export and import bans, changes in import tariff rates, or issuing government tenders for the importation o f subsidized grain. Problems frequently arise due to uncertainty about when and whether governments will alter import duties or import intentions in response to a short crop (e.g., Zambia in 2000/01, 2001/02; 2005/06; Malawi in 2001/02). Traders otherwise willing to mobilize imports early are likely to incur financial losses if the government later waives the duty and allows competing firms (or the government parastatal) to import more cheaply. When governments create uncertainty over import intentions or tariff rates duringa poor crop season, the result is commonly a temporary under-provision o f imports, which can produce a situation o f acute food shortages and price spikes far above the cost o f import (Nijhoff et al. 2003, Mwanaumo et al. 2005, Tschirley et al. 2004). Analysts not familiar with the details o f these situations often erroneously interpret them as evidence that markets fail and that the private sector i s weak, leading to a rationale for continued direct government involvement in marketing. The above illustrations highlight the importance o f strategic interaction, in determining food security and improving market performance. Many analysts have concluded that predictable and transparent rules governing state involvement inthe markets would reduce market risks, allow for greater coordination between private and public decisions inthe market, and enable governments to more effectively achieve food security policy objectives (Kherallah et al. 2002, Jayne et al. 2002, Mwanaumo et al. 2005, Byerlee et al. 2006). However, these conclusions have generally not been tested in a rigorous manner as we do in this paper. Perhaps more importantly these recommendations may seem unconvincing or abstract to policy makers. The constant re-shuffling o f Ministers o f Agriculture and Permanent Secretaries makes it difficult for them to invest enough time in understanding the agricultural sector and develop a greater evidence-based appreciationo f the way the sector actually works. From their vantage point, they have not been in a position to see how the performance of markets may be influenced by their own actions. An important purpose o f the Zambia maize policy experiment was therefore to provide first-hand experience, through participating in a simulated market game, o f how government and trader 4 behavior influences market outcomes. The subsequent section explains in detail how this experiment was designed. 3. The ModelandExperimentalDesign When designing the experimental model we faced a number o f challenges. First, the model had to capture the most essential features o f the Zambian maize market. We therefore decided to inform the model using data from the actual market rather than usingartificial pay-offs which a common instandardeconomic experiments.Second, the modelhadto be simple enoughto beplayable ina short experimental session. Third, since the game was also designed as a learning tool, it had to be entertainingto play andnot too complex to handle. 3.1. The players The Zambianmaize market constitutes o f four economic agents: farmers, millers, traders and the government. Farmers, who grow and harvest the crop, are mostly small family enterprises, many producing for subsistence. Each farmer's influence on market outcomes i s small, so we do not model farmers as strategic players. We instead assume that their production level is determined exogenously, predominantly by rainfall. The second group, the millers, buy the harvest and turn the maize into maize meal. They then sell the meal to consumers, who amongst others use it as the basis for nshima, the staple diet in Zambia. Millers do not play a strategic role either and is therefore also omitted from the game. We represent the behavior o f consumers and millers with a demand function. The higher the price, the less consumers are willing to buy. The specification o f the demand fhction is described the subsequent section. The remaining two types o f players are the key strategic actors in the maize market game: the traders and the government. In a shortage year traders import maize from nearby countries (mainly South Africa) and sell it to millers. The Zambian maize market comprises o f about 1,000 small traders who make up about 60 percent o f the trading volume. Four large trading companies (AFGRI,Amanita, Zdenakie, and CHC Traders) cover the remaining 40 percent o f the market (Jayne et al. 2007). Their trading volume is sufficiently high to exert market power, so they can be assumed to make their decisions strategically, taking the actions o f the other players into account. We assume traders to be profit-maximizers. Finally, the behavior o f the government, through the Food Reserve Agency (FRA), strongly affects market outcomes. In shortage years the FRA imports maize in competition to the private sector. In this sense it can be seen as an additional big trader on the market. In contrast to the private traders, the government is not a profit-maximizer, but is assumed to pursue a political agenda aimed at re-election. To gain popular support from consumers the government tries to keep consumer prices low. On the other hand, since many households inZambia are small maize farmers, the government also has an interest inhighproducer prices. This set o f goals essentially conflicts with that o f traders as discussed later. 3.2 The consumer market As the core model we chose a Cournot (1838) oligopoly game. Inthis model suppliers choose their quantities and the price i s determined by the market. The model is appropriate for a basic 5 agricultural product with a high degree o f product homogeneity. Further, it has very natural predictions for the market outcome. It reacts smoothly to small changes inthe traders' behavior, and changes inthe competitive environment leads to the expected change inmarket outcome (e.g. an increase inthe number o f firms results inlower prices and profits). These propertieshave also been confirmed in a plethora o f experimental studies (e.g. Huck et al. 1999, 2004; Offerman et al., 2002; Bosch-Domknech and Vriend 2003). Traders face a downward-sloping demand hnction, where the consumer price, P", is a decreasing function o f the total quantity supplied by the market. For simplicity, we assume a linear demand function with the inverse form:3 P" =a-b(Q+G+S) (1) Where Q is the total quantity supplied by the traders, G i s the government quantity and S is the baseline supply offered by small traders. Exogenous parameters a and b specify intercept and slope o f the demand function, respectively. As mentioned, the suppliers on the consumer market consist o f four big traders and a large number o f small traders. The small traders are price takers with a capacity constraint. Theyjointly supply a fixed quantityS, which they sell irrespective o f the market price, without strategic considerations. Each trader faces constant marginal costs, c. For simplicity we assume that marginal costs are identical to the producer price (the price that the farmers receive), assuming other costs (notably transport costs) to be fixed and thus not affecting optimal choices. Note that this assumption implies that traders have the same cost structure, since the producer price i s the same for all.4 Thediscretionary policy case Inthebaseline model, traders andthe government choose their quantities simultaneouslyafter the government has made an announcement about its supply intention. The total quantity is then given by &i + G + S, where qi is the quantity chosen by trader i. We assume that the big traders are not capacity constrained, i.e. they can import unlimited ~upplies.~Traders i's profit i s given as: ri=(P" -c)q,=[a-b(q, + +G+S)-c]q, (2) Empirical estimates ofien yield a relatively constant demand elasticity over the relevant range o f market outcomes. This invites the use o f a constant elasticity demand fimction o f the formp=Q7. We have tried estimations o f such demand functions, but they turned out to have very undesirable properties in a strategic market model. In extreme cases they would lead to corner equilibria, inwhich firms would optimally sell one grain o f maize at an i n f i i t e price. The reason i s that we cannot expect the constant elasticity assumption to hold over the entire price range, including those prices not empirically observed. Ina strategic model, however, the unobserved range can affect the equilibria dramatically and we therefore decided against the use o f this functional form. It i s important to note that although intuitive, this model o f price determination o f the maize market is a simplification. Inreality, the government announces consumer and producer prices at the start o f the season and is at liberty to change this price later inthe season or cease purchasing at any time depending on its rate of intake and in light o fchanges inmarket conditions. In Zambia, the issue o f trader's import capacity is a contentious one with govenunent questioning whether the private sector has sufficient capacity and the private sector eager to demonstrate that it does. 6 where Q-i denotes the total quantity provided by the traders less player i's supply. This i s similar to the profit ina standardCournot model except that we introduce government supply, G, and the bulk quantity S supplied by the small traders. The market equilibrium can be obtained by maximizing trader i's profit function and solving for qi. The equilibrium quantity qi* i s then given bythe expression: qi* = a -b(S b(n 1) ++G)-e (3) where n denotes the total amount o f traders. Note that expression (3) contains the government's quantity G which is endogenous. Since the government is not a profit maximizer, we can only solve for the eventual equilibrium quantities once the payoff function o f the government has been specified. The case ofpolicy pre-commitment Inthis variant of the game the government chooses its quantity before the traders. Hence the Cournot game o f the discretionary case turns into a variation o f a Stackelberg oligopoly model, with the government as the leader and the n traders as followers. Equilibrium quantities are computed the same way as inthe discretionary case. However, since the government's quantityi s now known when traders make their decision, the market outcome may be different, as we shall study inmore detail later. 3.3. The demandfunction For both research and training purposes, it was important that the model's parameters were not invented, but at least informed by real-world data. This increases the relevance o f the experimental results, and makes the game more recognizable as the Zambian maize market environment to the real players in the workshop experiment. Efforts to generate real-life parameters, o f course, find their limits inthe availability o f robust data to feed into the model. In the current framework only very sparse data were available, so the market model we develop cannot claim statistical robustness nor a highlevel o f accuracy. On the other hand, given that the alternative was to assume arbitrary parameter values we decided to proceed with parameter estimation. As explained inmore detail inAnnex A, we obtain the following demand h c t i o n for badweather years: Pc =436-0.99(Q A') + (4) where Q+S i s the total quantity suppliedjointly by large and small traders (excluding government supply). This demand function was subsequently used as a basis for calculating trader payoffs in the experimentalmodel. 3.4. The governmentpayofffunction Government maize trading i s not aimed at making a profit. Indeed, due to the comparatively higher operatingcosts it often takes place at a loss. Inthis paper, the objectives o f the government are assumed to be political in nature. Food security and maize price stability are concerns frequently expressed by government officials. Further, because Zambia is a multi-party democracy the ruling party i s concerned about its chances o f being re-elected, so it aims to 7 increase popular support. As previously discussed, the maize price is a crucial variable for voter satisfaction, because o f the vital role maize plays in the staple diet. Virtually all Zambians are consumers o f maize. Highconsumer prices are a likely cause o f public dissent, so the government is interested in keeping consumer prices low. At the same time the majority o f Zambians are small farmers, where maize is the predominant crop. These maize farmers benefit from high producer prices, so the government also has an interest innot letting producer prices drop too far. In sum, the objective function of the government is consistent with the social contract notion introduced insection 2. We abstract from all other goals the government may have.6 Any estimation about the relative weight o f the government's two price objectives can naturally only be guesswork, since hard data on governments' payoff functions are inherently absent. For the model we therefore used a payoff function that was linear decreasing in the consumer price, P", and linear increasing in the producer price (Le. the marginal cost o f traders, c). In lack o f qualified data we assumed the natural prior that both goals have equal weight. Finally, to capture the fact that government imports are generally carried out at a higher cost relative to private sector imports, we assume that there is a constant cost to each metric ton o f maize supplied by the government, k>Oq7Government payoff, u, i s thus given by: u=c-PC-kG (5) Note that the government's goals directly conflicts with the interests o f the traders, since the difference between producer and consumer prices i s essentially the traders' profit margin, c.f. equation (2) Inreality, the Zambian government has awide range ofmaizemarketingpolicy instrumentsat its disposal. For example, it can influence market outcomes by setting import tariffs, granting export licenses or banning exports altogether. In this paper, however, we focus only on direct FRA activity as a buyer and seller o f maize. In a shortage year the government's main activity is to import maize from neighboring countries to ensure adequate domestic supply. It also buys some quantity from the domestic market for the strategic food reserve. All other things equal, increased government import lowers the consumer price, since it increases total supply. Since the government also buys some maize from the domestic producers, it increases the demand for domestically produced maize, and hence exerts an upwardpressure on producer prices. Ina shortage year this effect i s relatively small, since the excess demand must be filled with imports and domestic contributions to the strategic food reserve play a small role. 6 Other relevant government objectives include, for example, development and modernization o f the food marketing system, reducing the treasury costs o f grain marketing operations, andprice stabilization. 'The assumption k>O i s critical for the results of the paper. It captures the fact that the private sector has a cost advantage over government inimporting maize. Economic efficiency therefore increases inprivate sector imports. 8 Table 1. The Payoff Tables A Trader's payoff if the Government chooses a LOW quantity I 20 40 60 80 % II 201 2763 -- I 2763I 2367 4733 I Ig7l5912 I 6299 U C An 4733 394I 3149 2357 T I 2367I 3941 I 4724I 4715 C 60 5912 4724 3536 2348 1971 3149 3536 3131 4715 3131 1547 "" I 1575II 2357 II 2348 II 1547I My own payoffis written inred,the other trader's payoffinblue. A Trader's payoff if the Governmentchooses a HIGH quantity I The other trader's quantity I 20 40 60 80 20 614 218 -178 -574 .->, 614 436 -534 -2296 +I C -I m 40 436 -356 -1148 940 3 v 218 -356 -1722 -3880 60 -534 -1722 -2910 -4098 P -178 -1148 -2910 -5464 80 -2296 -3880 -5464 -7048 -574 -1940 -4098 -7048 My own payoffis written inred,the other trader's payoffinblue. The Government's payoff Government's 1 The traders' TOTAL quantity I quantity 40 60 80 100 120 140 160 Low 913 1111 1309 1507 1705 1903 2101 Hinh 1528 1726 1924 2122 2320 2518 2716 9 3.5 Adaptationsof the modelto the experiment The real Zambian maize market has four big traders. However, with four suppliers the game would have been hard to present transparently to experimental participants. Moreover, the principal analytical interest i s the strategic interaction between government and traders rather thaninteractionbetweentraders. For those reasons, we reduced the number oftraders to two.* By reducing the number o f active traders we also understate the competitiveness o f the real market. However, it turns out that the main characteristics o f the market, mainly with respect to the strategic environment, remain preserved. Further, the effect o f government supply on domestic producer prices needs to be taken into account. The government must buy its supply from the market first. As mentioned, in a shortage year this effect i s not supposed to be large, since most o f the maize the government sells i s imported. It i s therefore assumed that in the `high government supply' case producer prices are only 10 percent higher than in the `low quantity' case. This figure is well within the empirical range o f observed prices (see Appendix table A.1). Finally, the strategy space was reduced inorder to makethe payoffs presentable intables. Traders therefore have only four options. They can each choose quantities of 20, 40, 60, or 80 kMT (thousand metric ton). The government's options are reduced even further. It can either supply a low quantity (of zero kMT) or a highquantity (assumed to be 80 kh4T). With the reduction o f the strategy space o fplayers it is now possible to represent the game using relatively compact payoff tables (see table 1). The government's payoff depends on its own choice and the aggregate quantity supplied by the two traders. Thus, one table i s sufficient to display the government's possible payoffs. Since the government's choices is restricted to two (either a highor a low quantity), the traders have to take two different payofftables into account, one for each o f the government's possible choice^.^ The reduction o f the market from a tetraopoly to a duopoly facilitated a presentationo f the game in bimatrix form, as it is tradition in game theory. O f course most experimental subjects and virtually all workshop participants were not trained in game theory and thus unfamiliar with bimatrix games. The bimatrix representation often looks unintuitive and confusing to game- theoretic laymen. All payoff tables were therefore printed in color, marking all choices and payoffs for one trader in red and for the other in blue. Color-coding turns the bimatrix into a transparent and easy-to-use representation o f a game. 3.6 Game-theoreticanalysis The game-theoretic analysis o f the two variants is straightforward. Consider the discretionary variant o f the game. In stage 1, government announces its intended quantity. In stage 2, the The calibrations o fthe demand function continue to assume a four-trader market, since this corresponds to the real- life constellation (see Appendix A). Itmay seem very restrictive to let the government choose only between two rather extreme alternatives. However, the game theoretic analysis will show that the fkndamental characteristics of the game do not get lost. For the government higher quantities are always better than lower ones, while for the social optimum the lowest government quantity would be preferable. 10 government and traders decide simultaneously on the quantity that they supply. Note that the government announcement at stage 1o f the discretionary game i s 'cheap talk' andwill not affect the game theoretic prediction. From table 1, it is observed that the government's dominant strategy is to supply a high quantity, as its payoff is always higher regardless o f what the traders do. The traders foresee that the government will always choose high, and only take the payoff table for the government's high choice into account. In this case each trader has a dominant strategy to choose the lowest possible quantity o f 20. The corresponding Nash equilibrium payoffs are (u; TI; ~ 2 = ) (1,528; 614; 614) for the government andthe two traders. The Nash equilibrium, however, i s a Pareto-inferior allocation. To realize this, suppose that the government can credibly commit itself to choosing a low quantity. The mutual best response occurs if each trader submits a quantity o f 60. Inthis allocation the corresponding payoffs are: (u ;TI ; ~ 2 = ) (1,705; 3,536; 3,536). This allocation represents a Pareto improvement since both government and private sector would be better off. However, in the discretionary variant the government cannot credibly make such a promise, as both traders know that once the decision stage i s reached, a rational government will play its dominant strategy. A rules-based policy can overcome this strategic dilemma." Inthe precommitment treatment the government i s a Stackelberg leader. It makes a binding decision before the traders make theirs, thus the traders know what the government will do. The subgame perfect equilibrium (Selten 1965) o f the game i s identified as follows. If the government chooses a high quantity, then the traders choose 20 each, and the government receives 1,528 just like in the equilibrium o f the discretionary game. If the government chooses a low quantity, then the traders respond with choosing 60 each, which leads to a government payoff o f 1,705. Thus, the government's best strategy is to commit to a low quantity. 3.7 The conduct of the mainexperiment The experiment was first conducted with 96 volunteer participants from the University o f Amsterdam. It was run as a pen-and-paper experiment ina classroom. A computerized setup was not used for two reasons. First, to maintain some parallelism with the workshop experiment. Second, to enable a re-run o f the exact same set-up in other Southern African countries in hture studies. In these countries computerized laboratories, which are the norm in most standard university experiments, are virtually non-existent. Each subject was allowed to participate in one session only, and no subject had participated in experiments similar to the present one. The subjects were undergraduate students from a wide range o f disciplines, with a balanced gender distribution. The experiment was conducted in English, which i s the language o f instruction for most students in Amsterdam. The subject pool was very international, with only a relative majority o f Dutchcitizens. In each session between four and six experimental markets were run in parallel. Subjects interacted in fixed groups o f three subjects. Subjects were not told who o f the other participants were in the same market, but they knew that the composition o f the markets did not change lo mayseemcounter-intutive that apolicyrulecan It be preferable to discretionary intervention, but this i s a well- established generalized result in economic theory first derived by Kydland and Prescott (1977) which applies to many fields ofpolicy making. 11 during the experiment. Subject roles (governmenthrader) were also held constant. The subjects were seated distantly from one another in order to ensure that they could not influence each other's behavior except through their decisions inthe game. The players' decisions were communicatedusingdecision sheets and results sheets. At each stage o f the game subjects filled in a decision sheet. Ifone role was inactive at one stage o f the game, the relevant players were given a `dummy sheet' asking for their expectations o f the other players' behavior. These sheets were administered to avoid revealing the roles o f participants which would have been the case if sheets were distributed to a subset o f participants only. The dummy sheetswere not usedto collect any data. Six rounds o f the game were played in each two-hour session, representing six years o f the Zambian maize market. This is a slightly longer time horizon than an election term in Zambia where the President is elected for a five-year term. Longer play allows learning and stabilization o f behavior. However, a length o f many rounds, as common in computerized experiments, was not possible in the pen-and-paper set-up and was also unrealistic, given that decision makers in the Zambian government frequently change.l1 At the outset o f the experiment, a capital balance o f 2,000 talers (the experimental currency) was granted to each subject, to account for possible losses. The total earnings o f a subject from participating inthis experiment were equal to this balance plus the sum of all the profits he made during the experiment, minus all losses. A session lasted for about two hours (this includes the time spent to read the instructions (see Appendix B).At the end o f the experiment, subjects were paidtheir total earnings anonymously incash, at a conversion rate o f one euro for 1,500 talers. A show-up fee o f 5 was given to each subject showing up on time. Subjects earned between 14.35 and 49.50 with an average o f3 1.21, which is considerably more than a student's regular wage in Amsterdam. At the time o f the experiment, the exchange rate to other major currencies was approximately US$1.30 and60.70 for one euro. Three sessions were conducted in each o f the two treatments. Since participants did not interact except within their own market, each market can be considered a statistically independent observation. Intotal, 16 independent observations were gathered ineach treatment. 4. Resultsof the MainExperiment The central purpose o f the main experiment was to test different policy options for the Zambian maize market with robust replicable data. The game theoretic analysis o f the model suggests the rules-based policy, in which the government precommits to its decisions, to be strongly superior to the discretionary regime. However, whether this advice i s empirically valid is another matter. The theoretical inferiority o f the discretionary policy stems from the social dilemma, i.e. the conflict between individual and social rationality, present in the maize market. Numerous experimental studies, however, have shown that subjects are frequently able to overcome such To illustrate, the Minister o fAgriculture, who participated inthe workshop, came into office only inOctober 2006, half a year before the event. 12 dilemmas and reach stable optimal outcomes through trust and reciprocity (see Ledyard 1995 for an overview). 4.1. The discretionary treatment Inthediscretionarygamegovernmentplayershaveadominant strategy to supplyahighquantity. They thus have a strong incentive to choose a high quantity and earn a short-term profit. Inorder to reach a Pareto-superior cooperative arrangement, traders must trust that the governments can resist this temptation. At the same time the government also needs to trust the traders. If traders supply low quantities, then the government's payoffis very small ifit also chooses low. Total quantities Figure 1 shows the average total quantity supplied by the traders (left axis) and the frequency o f high choices by the government (right axis). The figure illustrates that cooperation i s frequently attempted inearly rounds, but it i s very short-lived. Over time, high choices from the government become increasingly frequent. By the end o f the experiment, cooperation has collapsed in all but one market. Inaccordance with the rising frequency o f high choices, quantities supplied by the traders decrease from the third round onwards. There is some evidence to suggest that it is the governments which first cease to cooperate and that the traders respond to this. In the disaggregated data, however, no predominant response pattern i s evident.l2 Figure 1. Discretionary Policy: Trader and Government Supply II I I I I I 0.00 1 2 3 4 5 6 Round Misleading announcements Before government and traders choose their quantities, the government sends a non-binding signal to the traders, indicating which quantityit intends to choose. The government can use this signal to encourage traders to supply highquantities, ifit announces that it will itself choose low. 12In 13 o fthe 16markets high government frequencies rise from the f i s t to the second half o fthe game, whereas in the remaining three markets this frequency remains unchanged. The binomial test rejects the nullhypothesis o f equal likelihood o f rising and falling frequencies at p=O.OOOl (one-sided). Trader quantities fall in 10 of the 12 markets in which there is a change and this fall i s significant (p=0.0193, one-sided). Inthe six rounds o f play trader quantities do not fully converge to the noncooperative equilibrium, but the trend points toward that outcome. Notably government high frequencies rise fromthe thirdround on; trader quantities appear to follow with a one-round lag. 13 However, it can also use the messaging device to send a misleading signal, i.e. to lure the traders into believing the government would choose a low quantity, while it in fact intends to choose a highone (note that government payoffis monotonically increasing inmaize availability). Some observers believe that the Zambian government has occasionally made such misleading announcements, and in fact the strategic environment seems conducive to this behavior. Table 2 shows the distribution of the four possible combinations o f announcement and actual choice. In 36 o f 96 rounds (37.5 percent) the government chooses high after announcing low. A misleading signal inthe opposite direction (choosing low after indicating high) was made only once, possibly bymistake or after anhonest change o fmind. Announced Implemented Low I High I Total Low 38 1 39 High 36 21 57 Total 74 22 96 4.2. The precommitmenttreatment In the precommitment, or rules-based, treatment, the strategy dilemma between rational own- payoff maximization and social efficiency concerns is absent. The sub-game perfect equilibrium i s for the government to choose the low quantity, since it knows that it is in the traders' best interest to supply high quantities themselves. One may therefore expect that precommitment improves total supply. Looking at the overall picture, however, the improvement is surprisingly small. Average total trader quantity rises only slightly from 74.4 kMT to 79.2 kMT. The frequency of government high choices decreases from 58.3 percent to 49.0 percent, but this difference i s statistically insignificant.l3 Figure2. Pre-commitmentPolicy:Trader and GovernmentSupply 100 1.oo &- 80 z 0.80 2n != S5 60 .-2? 0.60 5 0 E 40 0.40 q Y !3 CJ 20 0.205' I' s 0 0.00 1 2 3 4 5 6 Round l3 'Fisher's two-sample permutation test cannot reject the null hypothesis o f equal trader quantities and equal government choices at any conventional significance level. 14 Two factors explain this phenomenon. First, the overall figures mask the strong deterioration in cooperation that i s present inthe discretionary treatment, but not inthe precommitment treatment. This can be seen infigure 2, which is analogous to figure 1for the precommitment treatment. In earlier rounds players make an effort to cooperate in the discretionary treatment, but cooperation eventually breaks down. Taking the second half of the experiment only (the last three rounds), there is a statistically significant difference in government choice frequencies, i.e. the precommitment policy dominates the discretionary one towards the end of the game.I4 Second, the unexpectedly low supply response o f the precommitment regime can be traced back to a phenomenonthat we term theparanoia effect. Recall that governments move first and traders second. When governments choose a low quantity, they must rely on the traders responding with high output levels, otherwise governments can be severely hurt by the resulting food crisis. Governments need only to rely on the traders to act intheir own best self interest, thus one would not expect the exposure to this risk to be very high. Nevertheless, figure 2 shows that in almost halfo f the rounds government refrained from choosing the efficient (high) quantity, arguably out o f fear to be hurt.15Such fears could be based inlack o f confidence inthe rationality o f the trader players or fear that these will act spitefblly. The question arises whether the governments' fear is warranted. In other words, did the trader players behave irrationally or spitefully in ways that reduced government payoff! The data reveals that this was not the case. Figure 3 shows the distribution o f total quantities conditional on the government's choice inthe precommitment treatment. It can be seen that inthe majority of cases traders responded to a government's low quantity choice with the equilibrium quantity of 120 and sometimes even 140 was achieved. These quantities are preferable to the government over the payoff the government obtains when choosing high (in which case virtually all traders respond with choosing 20 each). Only in about one-fifth o f all rounds did the traders supply a total o f 100. This allocation is only marginally worse for the government than the high quantity outcome yielding a payoff o f 1,507 instead o f 1,528. Thus, the fear o f exposure that many experimental governments apparently hadwas actually unfounded. Note that the paranoia effect reverses the behavioral patterns obtained in the large body o f standard trust game experiments.16 Intrust (or reciprocity) games a first mover can sendmoney to a second mover, who in turn can voluntarily reward the trustor by sending money back. The games are constructed such that by doing so, both players can be better off with respect to final payoffs, but in equilibrium no trust and no rewarding would be exhibited. Nevertheless, first movers frequently trust second movers, who reward the trust. So in trust games experimental participants do show trust though according to selfish rationality assumptions they should not. In 14 Fisher's two-sample randomization test rejects the null hypothesis o f equal frequencies o f government High choices at p=0.015 (one-sided). The analogous comparison for trader quantities is not significant (one-sided, p=0.30). l5 An alternative explanation could be that these subjects have a strong dislike o f disadvantageous inequity. Inthe efficient equilibrium traders earn more than the government, while inthe inefficient allocation the government earns more. However, most standard inequity aversion theories (e.g. Fehr and Schmidt, 1999) assume that individuals dislike inequity even if it is in their favor. The substantial occurrence o f `paranoid' choices inthe present experiment i s unlikely to be explained by inequality aversion alone. 16 See for instance Fehr et al. (1993), Berg et al. (1995), Dufwenberg and Gneezy (2000), Abbink, et al. (2000), FershtmanandGneezy (2001). See also Haile et al. (2006) for an application in the Southern African context. 15 our experiment, however, selfishly rational first movers should trust the second movers, since it i s inthe latter's owninterest to playa strategy favorable to the first mover. Nevertheless almost half o f the first movers unwarrantedly fail to trust the second movers. We are not aware o f existing experimental results that report this reversed patterno ftrustingbehavior. Figure 3. Precommitment Policy: Private Sector Supply for Alternative Government Choices (High or Low) 40 60 80 100 120 140 160 Private sector supply (krUrr) 4.3. Policy conclusions arising from the data Though the experiment consisted o f only two treatments, there are in fact three distinct policy options available to the government. Ifthe government chooses to establish a rules-based regime, it must also specify the rule to follow. Inthe framework o f the experiment, this means that, in addition to the discretionary regime, two cases can be distinguished in the precommitment treatment: commitment to a highquantity and commitment to a low quantity. One rationale behind the precommitment policy i s that it may encourage private sector activity and hence raise economic efficiency. Figure 4 illustrates that this goal is largely achieved. The figure shows traders' average total quantity for the three available policy regimes, over the six rounds o f the experiment. Precommitment to a low government supply induces the highest and relatively stable supply from the traders. A discretionary policy induces a lower trader supply which declines from round 3 onwards. Finally, precommitment to a high government quantity lead traders to respond with the lowest, butrelatively stable quantity. 16 Figure4. Trader Supply underAlternative GovernmentPolicyRegimes For food security purposes one needs not only to examine private sector activity, but the combined quantity supplied by government and traders and, notably, the variations inthis supply. Figure 5 shows the frequency distribution o f total quantities provided by the government and the private sector. While the mean total quantity supplied in each o f the three policy treatments i s quite similar, there i s substantially more fluctuation around the mean with a discretionary policy. Average total maize supply i s much more unpredictable under a discretionary policy due to frequent occurrences o f over-shooting or under-shooting with both government and traders supplying highor low quantities simultaneously. This lack o f coordination results in a substantial number o f `crises years' where the supply i s very low. Almost one-fifth o f all years results in a total supply o f less than 100 kMT - an outcome virtually nonexistent under a policy o f government precommitment. Precommitment to a highquantityi s results inthe most stable maize supply, but since this policy i s less efficient that precommitment to a low quantity, the latter is preferable. Figure5. FrequencyDistributionofTotalSupply by PolicyRegime 0.70 h 0.60 0.50 2 0.40 LL 0.30 0.20 0.10 0.00 40 60 80 100 120 140 160 180 200 Total supply(1,OOO kMT) 17 Finally, table 3 shows a range o f market performance measures including quantities, trader profit and government payoff.17 There is an almost complete crowding out o f the private sector if the government precommits to a high quantity compared to when it precommits to a low quantity. Total quantities are very similar, on average, inthe three regimes (around 120WT), although in our experiment the two government choices are at the rather extreme ends o f the scale (OkMT vs. 80kMT). Since the private sector i s more efficient in supplying maize to the market than government, economic efficiency is highest inthe `pre-commitment low' regime. Table 3. Market Outcomes under Alternative Government Policy Regimes (Averages). Policy Regime Government Trader Total Trader Government Quantity Quantity Quantity Profit Payoff Discretionary 46.6 75.3 121.9 1,446 1,618 Precommitment Low 0.0 115.5 115.5 2,707 1,661 Precommitment High 80.0 41.3 121.3 611 1,541 5. The WorkshopExperiment In addition to the main experiment with student subjects, the same experiment was also conducted with participants from the real maize market inZambia. This happened in the context o f the Zambia Maize Market Policy Dialogue which was a one-day workshop attended by 20 high-level government officials and private sector maize market players (traders, millers and farmers). Government representation included the Minister and a Permanent Secretary o f the Ministryo f Agriculture and Cooperatives andrepresentatives of the FoodReserve Agency. The private sector was represented by inter alia the Chief Executives o f the Grain Trader's Association, the Miller's Association of Zambia and the Zambia National Farmers Union. The experiment was conducted as one o f the first events o f the workshop, immediately after the official opening remarks and introductory comments. Care was taken that no substantial information about the nature o f the experiment was passed to participants beforehand, and that the introductory comments only made vague statements about what was to follow. The instructions were then read aloud. The participants were given the additional information that the game was designed using Zambian maize market data and that it represented a shorta e year, such as in2005, inwhich the maize supply o f traders and government would be imported. ?* A few modifications hadbeen appliedto the game compared to the mainexperiment with student subjects. One important change was that all players were represented by teams o f four participants, while the student subjects played individually. Individual play i s effective in data gathering, but is less suitable for training purposes. It was important for a successful workshop outcome that the game was entertaining to play, and individual play sessions with their long inactive phases can be quite tedious for participants. Further, the workshop was intended to "Thesefiguresdonotincludethe ''Note bulk supply fromthe small traders, whichis held constant, that in the season preceding the workshop rainfall conditions had been excellent and a bumper harvest was expected (though minor flooding had occurred insome areas). Insuch circumstances, the policy issues inZambia are quite different from those occurring when a bad harvest is expected, including e.g. concerns from trader's regarding the government's export policy. Without this information, therefore, participants may have beenconfused. 18 stimulate a dialogue between the different sides o f the market. The team discussions naturally inspired a lot o f debate duringand after the game. During play, teams were seated in separate rooms, where they could discuss their decisions without influencing or being influenced by discussions o f other teams. Each group had a facilitator (a member o f the organizing committee) to assist the group inansweringquestions and to remindthem towards focusing on the facts o f the game itself as opposed to the more complex reality that the real players may refer to whenmaking decisions. The experiment simulated two parallel maize markets. Eachparticipant was randomly assigned to a team. As a consequence, the participants did not necessarily play the role that they play in reality. The Minister o f Agriculture, for instance, played in a trader group. Group composition was often mixed with representation from both sides o f the market. It turned out that this feature was very useful for the purpose o f the workshop, since it enabled participants to experience the game from different perspectives, either by own experience or discussion with a team-mate from a different camp. Due to intense discussions within the teams the workshop experiment proceeded slower than the student sessions. Nevertheless, the teams managed to play three rounds o f the discretionary variant and two rounds o f the pre-commitment treatment within 2% hours. In contrast to the students workshop participants playedbothtreatments. Workshop participantswere not incentivized with monetarypayoffs due to ethical considerations. The concern was that handing out prizes to government and business representatives at a workshop aimed at improving food security in a poor country could have adverse reputational effects for all parties involved (monetary prizes are commonplace inexperimental economics, but experiments are unusual in the given context). As a substitute, the best government and the best two trader teams received symbolic prizes in the form o f certificates recognizing outstanding performance at the work~hop.'~Despite the lack o f a proper proportional incentivizing mechanism, intrinsic motivation provedhighand debates inthe teams were lively. Due to the limitednumber o f observations (only two markets and five rounds) it i s not possible to generalize the outcome o f the workshop experiment. The results should therefore be regarded as anecdotal. The two markets had substantially different outcomes. Market 1 had a government intend on cooperation with players who quickly identified the optimal outcome (low government imports and highprivate sector imports). In fact, this government identified this strategy already after the first round (though it sent confusing signals to the traders) and cooperation quickly evolved. This market behaved more efficiently than a typical market in the main experiment. Market 2, in contrast, exhibited characteristics which were much less cooperative than in the main experiment. According to statements made by the subjects after the experiment, the government players deliberately tried to punish traders by announcing low, but implementing high, maize imports. Moreover, the traders were relatively slow in responding to the l9Paying cash proportional to success, as usual, would have looked bad. One possibility was to award desirable but not too extravagant material prizes, such as portable music players or digital cameras, and make each participant's probability of winning proportional to their points earned. This would have been theoretically sound. For the symbolic prizes, however, playing out the lotteries would have time-consuming and irrelevant. 19 government's malevolent strategy. This resulted in a total negative payoff for both traders (effectively they went bankrupt). The results o f the workshop experiment are presented in Appendix C. 6. Conclusion Discretionary and unpredictable government intervention is one o f the greatest policy problems plaguing the food marketing systems and food security in the Southern Afiica region. This i s because actual and potential government interventions generate private sector uncertainties and inaction leading to additional government intervention needs. This problem has underlied virtually all o f the recent food crises inZambia andMalawi since 2000, where food supplies have dwindled and prices surged above the cost o f importing it. Effective coordination between the private and public sector would require greater consultation and transparency with regard to changes in parastatal purchase and sale prices, import and export decisions, and triggers for release o f stocks. This approach does not imply that government need be passive. Instead, it implies that government responses need to be transparent, reliable, and predictable in order to create the space for the private sector to play its role. The private sector role, inturn, includes the reliable andpredictable management o f commercial imports, when economically viable. The results o f the maize market experiment underpinthis policy recommendation. A simple pre- commitment rule was found to be superior to discretionary policy makingby reducing the risk o f food crises and providing appropriate incentives for private actors to participate in the market thereby enhancing economic efficiency. More specifically, total maize quantities and market prices are quite similar under the two different policy modes. Importantly, however, situations o f food shortage (and over-supply) were much more frequent under a discretionary policy because o f the risk o f poor coordination between the government and the private sector. Government pre- commitment to a low quantity also resulted insubstantiallyhighertrader profits, and hence higher economic efficiency because o f the larger volume traded by them. The Government o f Zambia may therefore want to consider mechanisms which can help make maize market policy more predictable or rules-based inthe future. Our experimental results also highlight the difficulties that the strategic environment o f the Zambian maize market imposes on rules-based policies. Eventhough it was inthe private sector's best interest to supply high quantities if the government pre-committed to a low quantity, many experimental governments still failed to put sufficient trust into the private traders, most likely out o f fear o f failure on the side o f the private sector. It turned out that such fear was unwarranted, which is why we coin the term paranoia effect for this behavioral phenomenon. This observation is surprising since it reverses the pattern observed in usual experimental trust games, in which subjects are far more trusthl and cooperative than they should be according to theoretical predictions. In the present setting subjects fail to trust even though there were no material reasons to mistrust. Nevertheless, those governments that did pre-commit to a low quantity were rewarded with highprivate sector activity, a result that is encouraging for current maize marketingpolicy making. This would not be the first time that policy makers have been encouraged to reform maize marketing policy by introducing higher degrees o f transparency, predictability and cooperation 20 towards the private sector, yet policy makers have thus far been reluctant in adopting such recommendations. An important reason, as explained in the paper, i s the predominance o f neo- patrimonial relationships in which national leaders maintain loyalty and patronage among rural leaders through commodity distribution. A second explanation i s that market controls enable governments to adhere to a `social contract' in which it supports smallholder agriculture while simultaneously ensuring cheap food for the urban population. For those same reasons, the policy recommendations presented inthis paper, should not be expected to be adopted overnight. Inrecognitionof the realities ofthe Zambianpolitical economy, we made a substantial effort of engaging policy makers and private sector participants over the issue o f maize market policy reform through their active participation in the economic experiment. This policy dialogue, and hence the use o f an economic experiment as a learning tool, was successful for several reasons: First, it allowed participants to develop a greater evidence-based appreciation, ina repeated-game setting, o f how maize market outcomes are affected by their own strategic interaction with other players. Secondly, it encouraged active discussion within each group due to the mixed group composition. Finally, the experiment facilitated a more constructive dialogue between parties with potentially conflictive interest. In afler-game discussions participants frequently referred to their experience in the game as a way to make their arguments while avoiding reference to controversial and painful episodes o f the past. This experience augers well for use o f economic experiments as a learning tool in many other contexts where the nature o f strategic interaction plays a critical role for social outcomes. More importantly, it augers well for the future o f the policy dialogue between the government andthe private sector inZambia. 21 References Abbink K., B. Irlenbusch, and E. Renner, 2000, `The Moonlighting Game - An Experimental Study on Reciprocity and Retribution',Journal of Economic Behaviour and Organization 42: 265-277. Berg, J. J. Dickhaut, and K. McCabe, 1995, Trust, Reciprocity and Social History, Games and Economic Behaviour, 10, 122-142. Bird, K., D. Booth, and N.Pratt, 2003, `Food Security Crisis in Southern Africa: The Political Background to Policy Failure.', Forum for Food Security in Southern Africa, Theme paper No. 1,Overseas Development Institute, London. Bratton, M. and R. Mattes, 2003, `Support for Economic Reform? Popular Attitudes in Southern Africa' WorldDevelopment 3l(2): 303-323. Byerlee, D., R. Myers, and T. Jape, 2006, `Managing Food Price Risks and Instability in an Environment o fMarket Liberalization', World Bank, Washington, D.C. (Processed). Cownot, A.A., 1838, `Recherches sur les Principes Mathbmatiques de la ThCorie des Richesses.' Hachette: Paris. Deininger, K. and P. Olinto, 2000, `Why Liberalization Alone Has Not Improved Agricultural Productivity InZambia: The Role of Asset Ownership and Working Capital Constraints', Policy Research Working Paper 2302, Washington, D.C. Del Ninno, C., P. Dorosh, and K. Subbarao, 2005, `Food Aid and Food Security inthe Short- and Long Run: Country Experience from Asia and sub-Saharan Africa.', Social Protection Discussion Paper 0538, World Bank, Washington, D.C. Fehr E., G. Kirchsteiger, and A. Riedl, 1993, `Does Fairness Prevent Market Clearing? An Experimental Investigation,' Quarterly Journal of Economics 108:437-459. Fehr E. and K. Schmidt, 1999, `A Theory o f Fairness, Competition and Cooperation', Quarterly Journal of Economics 114(3): 817-68. Fershtman C., and U. Gneezy, 2001, `Discrimination in a Segmented Society: An Experimental Approach,' Quarterly Journal ofEconomics 116: 351-377. Goldsmith, A., 2001, `Institutions and Economic Growth in Afiica' in M. McPherson (Ed), Restarting and Sustaining Growth and Development in Africa, Washington DC: USAID, andCambridge, MA:HarvardInstitute for International Development. Howard, Julie, and Catherine Mungoma, 1997, `Zambia's Stop-and-Go Maize Revolution' inD. Byerlee and C. Eicher (Eds.) Africa 's Emerging Maize Revolution, Colorado: LynnRienner Publishers. Huck, S., H.T. Nonnann, and J. Oechssler, 1999, `Learning in Coumot Oligopoly - An Experiment', Economic Journal 109: 80-95. Huck, S., H.T. Normann, and J. Oechssler, 2004, `Two are few and four are many: On number effects in experimental oligopolies' ,Journal of Economic Behavior and Organization 53: 435-446. Jape, T.S., J. Govereh, A. Mwanaumo, J. Nyoro and A. Chapoto, 2002, 'False Promise or False Premise: The Experience of Food and Input Market Reform in Eastern and Southern Africa', WorldDevelopment 30(11): 1967-1986. Jape, T.S., B. Zulu, and J.J. Nijhoff, 2006, `Stabilizing Food Markets in Eastern and Southern Africa', Food Policy 31(4): 328-341. Jape, T. S., A. Chapoto, B. Zulu, A. Mwanaumo, J. Shawa, and S. Haggblade, 2007, `Zambia's Maize Value Chain: Toward Improved Food Security and Small-Fanner Income Growth' Working Paper, Food Security ResearchProject andthe World Bank, Lusaka. 22 Jenkins, C., 1997, `The Politics of Economic Policy-Making in Zimbabwe', Journal of Modern African Economies 575-602. Kherallah, M., C. Delgado, E. Gabre-Madhin, N. Minot, and M. Johnson, 2002, `Reforming Agricultural Markets inAfrica', Baltimore: Johns Hopkins UniversityPress. Lopez, R. 2003, `Under-investing in public goods: evidence, causes, and consequences for agricultural development, equity, and the environment', Agricultural Economics 32(1): 211- 224. Kydland, F. and E. Prescott, 1977, `Rules rather than Discretion: The Inconsistency of Optimal Plans', Journal o fPolitical Economy 85(3): 473-492. McPherson, M. (ed.), 2002. `Restarting and Sustaining Growth and Development in Africa: Equity and Growth through Economic Research', Cambridge, Harvard Institute for International Development. Mwanaumo, A., W. Masters, and P. Preckel, 1997, A Spatial Analysis o f Maize MarketingPolicy Reforms inZambia, American Journal of Agricultural Economics, 79(2): 514-523 Nijhoff JJ., D. Tschirley, T. Jayne, G. Tembo, P. Arlindo, B. Mwiinga, J. Shaffer, M. Weber, C. Donovan, and D. Boughton, 2003, `Coordination for Long-term Food Security by Government, Private Sector and Donors: Issues and Challenges.', Policy Synthesis 65, Michigan StateUniversity. North, D., 1994, `Institutions and Credible Commitment', Economic History, WPA, 9412002. North, D., 1987, `Institutions, Transaction Costs, and Economic Growth', Economic Inquiry 25(5): 419-428. Pletcher, J., 2000, `The Politics of LiberalizingZambia's Maize Markets', WorldDevelopment 28 (1): 129-142. Sahley, C., B. Groelsema, T. Marchione, and D. Nelson, 2005, `The Governance Dimensions o f Food Security in Malawi', USAID Bureau o f Democracy, Conflict, and Humanitarian Assistance, Washington, D.C. Siegel, P., and J. Alwang, 2005, `Poverty Reducing Potential o f Smallholder Agriculture in Zambia: Opportunities and Constraints', SDV Working Paper Series No. 85, World Bank, Washington, D.C. Tschirley, D., J. Nijhoff, P. Arlindo, B. Mwinga, M.Weber andT. Jayne, 2006, `Anticipating and Responding to Drought Emergencies in Southern Africa: Lessons from the 2002-2003 Experience',International Development WorkingPaper 89, Michigan State University. van de Walle, N.,2001, `African Economies and the Politics o f Permanent Crisis, 1979-1999.' Cambridge: Cambridge Univ. Press. World Bank, 2003, `Project Performance Assessment Report, Zambia Agricultural Sector Investment Program', Report No. 26086, Washington, D.C. World Bank, 2004, `Zambia Country Economic Memorandum. Volume I: Main Report' ,Report No. 28O69-ZAyWashington, D.C. World Bank and International Food Policy ResearchInstitute, 2005, `Agriculture and Achieving the MillenniumDevelopment Goals', Report #32729-GLBYWashington DC. World Bank, 2006, `Project Appraisal Document On A Proposed Grant in the Amount of SDR 25.7 Million (US$37.2 Million Equivalent) to the Republic o f Zambia for an Agricultural Development Support Project', Report No: 358O4-ZMyWashington D.C. 23 AppendixA. Calibrationof the DemandFunction Table 1contains the data used to calibrate the demandfunction. Total quantity traded (column 6) andthe observed prices (column 7) are particularly relevant. TableA.1. AnnualMaizeSupplyandPriceEstimates,1994-2006. (1) (2) (3) (4) (5) (6) (7) (8) (9) Year Weather Maize Small- Large- Total Lusaka Urban Adjusted Production scale scale Quantity Wholesale Consumers Total (kMT) Quantity Quantity Traded maize Population Quantity Traded Traded (kMT) urice Index (kMT) (kMT) ($/MT) 1994 good 1,020 357.0 300 657 150 1.oo 657 1995 moderate 737 184.3 200 384 208 1.05 368 1996 excellent 1,409 563.6 350 914 127 1.09 837 1997 moderate 960 240.0 300 540 173 1.14 473 1998 Bad 638 127.6 100 228 183 1.19 191 1999 moderate 822 205.5 300 506 135 1.25 406 2000 moderate 881 220.3 300 520 116 1.30 399 2001 Bad 601 120.2 150 270 192 1.36 199 2002 Bad 620 124.0 150 274 244 1.42 193 2003 good 1,161 406.4 300 706 169 1.49 475 2004 good 1,113 389.6 300 690 150 1.55 444 2005 Bad 866 216.5 300 517 236 1.62 318 2006 excellent 1,400 560.0 350 910 140 1.70 537 Note: Weather is classified according to the maize production, x: Bad: x400. Moderate: 700+4,000. Good: 1,000+4,300. Excellent: x>1,300. Column(7) is the mean o f 12month marketingperiod (May-April). Sources: Columns 3 and 4: Ministry of Agriculture and Cooperatives (MACO) Annual Post-Harvest data. Column 7: Agricultural Market InformationCentre, MACO. Column 8: Central Statistical Office: 1990and 2000 censes. Itwould be unrealistic to expect that all parameters o f the market environment remained constant over the 13-year period for which data was available. First, traded quantities have generally risen over this period due to urbanization. Maize production that is consumed by subsistence farmers would not be recorded in official data. Urban migration therefore increases the traded quantity recorded in table 1 even though the underlying demand remains unchanged. Total quantity (column 7) was therefore adjusted usingan urbanizationindex (column 8) andthe adjusted values (column 9) were usedto calibrate the demand function. Secondly, demand for maize is not independent o f the harvest. Inprinciple, consumer demand is determined by exogenously determined consumer preferences and opportunity cost, which should not be strongly affected by the weather. However, inZambia large quantities o f maize are grown for subsistence. In good weather years small farmers produce for their own consumption and sell their excess quantity to the market. In bad weather years, these small farmers become net buyers o f maize. Thus, demand for maize tends to shift outward inbad weather years and inwardin good weather years. Consequently, the four weather scenarios distinguished intable A.1we considered separately. Since the model i s designed to capture a shortage year, only useddata for bad weather years was used, which leaves four observations only (1998, 2001, 2002, 2005). The data set is further limited by the unusually hightraded quantity in 2005 and this outlier was ignored out o f caution. 24 We then calculated an average price and an average quantity using the three remaining data points, and considered this to be the `representative' outcome for a bad weather year. It is, o f course, not possible to generate a complete demand function from a single data point. To do this, we interpreted the representative observation as the equilibrium outcome o f a Cournot market game with the following assumptions: 1. There are four identical major traders who have jointly supplied 40 percent o f the total quantity.The remaining 60percent comes from non-strategic small traders. 2. The firms' marginal costs (i.e. the producer prices) are 5/6 o f the market price. The empirical gross profit margin o f a trader i s about 20 percent, so this was used as a proxy for the unknown Coumot profits. 3. There is no government intervention. Assumptions (1) and (2) are conceptually dubious as they take as they take a constant variable as an input to estimate something that should be a variable endogenous output. However, these assumptions are unlikely to distort the model outcomes drastically. The third assumption is more critical, since government intervention is typical for shortage years. Unfortunately, reliable data on government supplied quantities were not available. The direction o f this distortion i s also not obvious, since the effect o f government supply on total quantity depends on the strategic reaction bythe traders on expected government behavior. With these inputs one can search numerically for intercept and slope o f the demand function that returns the observed prices and quantities as equilibrium outcomes. For bad weather years the following demand function was obtained: Pc =436 - 0,99(Q+S) (4) where (Q+S, is the total quantity supplied jointly by large and small traders (excluding government supply). This demand function was then used as a basis for calculating trader payoffs inthe experimentalmodel. 25 Appendix B. Instructionsfor the Experiment Thank you for coming to the experiment. Inthis experiment you will make decisions in a market environment. During the session it is not permitted to talk or communicate with the other participants. Ifyou have a question, please raise your handand one o f us will come to your desk to answer it. Duringthe session you will earn money. At the end o f the session a show-up fee o f 5 euros plus the amount you will have earned during the experiment will be paid to you in cash. Payments are confidential, we will not inform any o f the other participants o f the amount you have earned. Inthe following, all amounts of money are denominated intalers, the experimental currency unit. Duringthe experiment you will bepairedwith two other participants.You will bepairedwith the same two other participants throughout the experiment. You will not be informed of the identity o f the person you are pairedwith. The experiment consists o f six separate rounds. Each round follows the same structure described below. There are three active players inthe market: Two Traders and the Government. Two o f the three participants ina group will play the role o f a Trader, the thirdparticipant will play the role o f the Government. Decisions in a round Discretionarytreatment: Each round consists o f three stages. At stage 1the Government announces a quantity he intends to supply at stage 2. At stage 2 the Traders andthe Government choose the quantities they supply. Precommitment treatment: Each round consists of two stages. At stage 1 the Government chooses a quantity it supplies at stage 2. At stage 2 the Traders choose the quantities they supply. Stage 1 Discretionary treatment: At stage 1the Government announces how much o f the commodity he intends to supply to the market at stage 2. It can choose a high quantity or a low quantity. The announcement is not binding, i.e. once stage 2 is reached the Government can choose a quantitydifferent from the one announced. Precommitment treatment: The choice i s binding, i.e. once stage 2 is reached the Government will supply the chosen quantity. The Traders are then informed about the quantitythe Government has chosen. Stage 2 At stage 2 the Traders simultaneously decide how much o f the commodity to supply to the market. Each trader can choose a quantity between 20 and 80, in steps of 20. So the possible choices each Trader can make are 20,40,60 or 80. 26 Discretionarv treatment: At the same time the Government decides how much o f the commodity to supply to the market. This can be the quantity announced at stage 1 or the other quantity. It can choose a highquantity or a low quantity. Precommitment treatment: The above paragraphwas omitted. Payoffs All payoffs are denominated intalers, the fictitious experimental currency. The Traders' and the Government's payoffs are determined by the total quantity supplied by the Traders and the Government. The total quantity is the sum o f the two Traders' quantities plus the Government's quantity. The total quantity determines the sales price for the commodity on sale, and hence, together with a trader's quantity choice, the profit. The Government's payoff represents the extent to which the Government meets its objectives. All payoffs have been calculated on the basis o f a theoretical market model. You need not calculate any payoffs. A Trader's payoffs, for all quantities chosen by the Traders andthe Government are listed inthe Trader's Payoff Tables. There are two payofftables for the Traders. The upper table shows a Trader's payoff for the case that the Government chooses a low quantity. The table below shows a Trader's payoff for the case that the Government chooses a high quantity. The Governor's payoffs, for all possible total quantities of the two Traders are listed inThe Government's PayoffTable. Note that the two traders are identical inthe set o f their options andthe corresponding payoffs. Endof a round After stage 2 has ended, the payoffs for all players are calculated and all participants are informed about the decision made by the other participants intheir group and about their own and the other players' payoffs. Earnings At the start o f the experiment you have a startingcapital o f2000 talers, to which gains are added and losses are subtracted. At the end o fthe session talers are converted into euros at an exchange rate o f one Euro for 500 talers. Inaddition, a show-up fee o f EUR 5 is paidto each participant. 27 Appendix C.Resultsof the Workshop Experiment Table C1. Market 1: DecisionsandPavoffs Discretionary treatment Round Government Trader 1 Trader 2 Government Trader 1 Trader 2 Government Announcement Quantity Quantity Quantity Payoff Payoff Payoff 1 High 20 20 Low 2,763 2,763 913 2 Low 40 80 Low 2,357 4,7 15 1,705 3 Low 60 60 Low 3,536 3,536 1,705 Pre-commitment treatment 4 Not applicable 40 60 Low 3,149 4,724 1,507 5 Not applicable 60 60 Low 3,536 3,536 1,705 Table C2.Market2: DecisionsandPayoffs Discretionary treatment Round Government Trader 1 Trader 2 Government Trader 1 Trader 2 Government Announcement Quantity Quantity Quantity Payoff Payoff Payoff 1 Low 60 40 High -1,722 -1,148 2,122 2 High 20 40 High 218 436 1,726 3 Low 60 40 High -1,722 -1,148 2,122 Pre-commitment treatment 4 Not applicable 20 20 High 614 614 1,528 5 Not applicable 20 20 High 614 614 1,528 28