WELFARE EFFECT OF CEREAL EXPORT BANS IN THE UNITED REPUBLIC OF TANZANIA

WELFARE EFFECT OF CEREAL EXPORT BANS IN THE UNITED REPUBLIC OF TANZANIA Abstract Mesia Ilomo University of Dar es Salaam Business School, United Republic of Tanzania ilomo5@yahoo.com The aim of this paper is to examine the consumer welfare effect of cereal export bans in mainland Tanzania. The focus of the research is on maize which is the most widely consumed cereal in the country which also accounts for a substantial share of cereal exports. Maize was the main target of cereal export bans over the past two decades. As existing research on the issue has provided mixed results, this paper complements it by looking in greater detail at the impact of the bans on different regions of the country. Regions are categorized into East African Community (EAC) border regions and peripheral regions depending on their level of integration into the EAC which is the main export market for Tanzanian maize. Drawing on data from the 2007 Household Budget Survey of the Tanzania Bureau of Statistics and price data from the Ministry of Industry and Trade and the Food and Agriculture Organization of the United Nations (FAO), the paper runs a series of regressions. The first one aims to determine whether the transmission from international to domestic prices is affected by the level of integration into the EAC market. The regression shows that the level of price transmission is high in regions bordering the EAC and limited in regions far from the EAC border. The subsequent regression at the regional level seeks to establish the effect of the export ban on domestic prices. The findings indicate that all regions are affected by the export ban, but to different degrees and in different directions. The export ban was associated with an increase in prices in most EAC border regions and a decrease in prices in most peripheral regions. Finally, a non-parametric regression estimates the welfare effect of the export ban on maize in Tanzania. At the national level, the ban led to a household welfare loss of approximately 1.5 per cent, with femaleheaded households suffering more than male-headed households. The welfare loss was slightly higher for the poorest households. Analysis at the regional level indicates significant variations of the welfare effect across regions: consumers in most of the EAC border regions lose from the export ban, while those in most peripheral regions experience a welfare gain. These findings partly support the conclusion by Dabalen and Paul (2014) that consumers suffered a welfare loss following the introduction of the export ban in Tanzania. Additionally, they show the heterogeneity of the effect of the export ban in Tanzania, which can be partly explained by the fragmentation of domestic markets. Regional-level analysis indicates no significant difference in the welfare effect attributed to the export ban between femaleheaded and male-headed households. Findings in this paper contribute to the debate about the appropriateness of non-tariff measures in trade policy. In particular, they support a conclusion that export bans are not always beneficial for consumers, which goes against the expectations that policymakers have when they introduce such measures. The paper also provides additional insights, particularly regarding the heterogeneous effects of policy measures, which might encourage targeted interventions that are location-specific. The author would like to thank Cristian Ugarte, Julia Seiermann, and Vlasta Macku from the UNCTAD Virtual Institute for their comments on this paper, and David Einhorn for editing the text. The paper was prepared with financial support from the One UN Fund for Tanzania and the government of Finland. 1

1 Introduction The United Republic of Tanzania is a member of two regional blocs, the East African Community (EAC) and the Southern African Development Cooperation (SADC), both of which aim to promote trade among their member states by, among other measures, eliminating all forms of barriers, including non-tariff barriers. 1 Tanzania has regularly imposed export bans on food, including bans on exports to EAC partner states, in order to ensure the availability of food and stable prices of cereals in the country (Kagira, 2011; Diao et al., 2013). This policy has aimed to protect consumers against increases in prices, but has been contested by producers and some policymakers who believe that a better solution would be to increase production, taking advantage of Tanzania s favourable climatic conditions. The literature shows mixed findings with regard to the effects of export bans on grain consumers in Africa. Some authors (Diao et al., 2013; Kagira, 2011) find that the effect is positive, even though marginal, while others such as Woldie and Siddig (2009) find no evidence that export bans benefit consumers of cereals. Dabalen and Paul (2014) find that consumers suffered a welfare loss following the introduction of the export ban in Tanzania. This paper therefore aims to contribute to this debate while taking into account the geographical characteristics of the domestic market in Tanzania. Statistics indicate that Tanzania mainly exports cereals to East African countries which account for three-quarters of the country s total cereal exports. Most of the exports are transported via roads. Tanzania is a large country with a fragmented domestic market, which makes it difficult to transport produce from production areas to markets. For example, the distance by road from Rukwa (a region in the southern highlands that is a key producer of maize) to Arusha (a region in the north near the border with the EAC) is 1,348 kilometers. Given that cereals are mainly transported by road, the location of the market is likely to affect price transmission and hence the welfare effect of an export ban on consumers. Tanzania has a tropical climate with some regional variations. Temperatures in the highlands range between 10 o C and 20 o C during the cold and hot seasons, respectively. In the rest of the country, temperatures rarely fall below 20 o C. Tanzania is considered an agricultural economy, with 80 per cent of the population in rural areas engaged in agriculture. The agricultural sector is dominated by smallholder peasants, each of whom cultivates an average of 0.9 to 3 hectares, mainly rain-fed and cultivated using hand hoes (Makame, 2013). The sector provides employment for most Tanzanian women, accounting in 2006 for 80 per cent of the country s female employment. 2 Due to climate change, rainfall-dependent agriculture is increasingly confronted with low productivity and variations in production levels. Tanzania has more arable land (approximately 46 per cent of its territory) than other EAC member states, and strategic use of this resource could take advantage of the market potential available in the SADC, EAC and 1 The member countries of the EAC are Burundi, Kenya, Rwanda, United Republic of Tanzania, and Uganda. The member countries of the SADC are Angola, Botswana, Democratic Republic of Congo, Lesotho, Madagascar, Malawi, Mauritius, Mozambique, Namibia, Seychelles, South Africa, Swaziland, United Republic of Tanzania, Zambia, and Zimbabwe. 2 According to World Bank estimates, the share of agriculture in female employment (80 per cent) was higher than the share of the agriculture in total employment (77 per cent) in 2006 (World Bank, 2015b). 2

the Tripartite Framework countries. 3 Regional integration could thus be seen as an opportunity for farmers to gain better access to regional markets. Between February 2002 and June 2014, Tanzania had four rounds of export bans from July 2003 to January 2006, from August to December 2006, from January 2008 to October 2010, and from May to October 2011. Export bans on staple crops, 4 a measure imposed by the Department of National Food Security, were a controversial policy measure used in the Tanzania up to 2012 when they were abolished following the commitment made by Tanzania s President Hon. Jakaya Mrisho Kikwete in bilateral consultation with President Hon. M. Kibaki of Kenya at the G8 Summit in the United States (Makame, 2013). The president s decision was crucial to honouring Tanzania s international commitments, such as the commitment to promote trade among EAC member states and the commitment to trade liberalization, including the pledge to the World Trade Organization (WTO) not to impose any new trade barriers. Export bans on food were introduced in Tanzania when production fell below the level of domestic demand. These measures were introduced on the assumption that unrestricted exports of food lead to a limited supply of food in the domestic market, thus causing an increase in domestic food prices. Consequently, it was assumed that export bans on cereals would increase the availability (supply) of cereals in the domestic market and eventually lead to lower and more affordable prices for consumers (Woldie and Siddig, 2009). Another reason for export bans, particularly in Tanzania, was to address the unethical behaviour of some intermediate traders who buy cereals from producers and particularly from financially constrained households at lower prices before the harvest, depriving producers of the opportunity to get higher prices in export markets (Kagira, 2011). Export bans discouraged traders from buying cereals from farmers because they were denied access to foreign markets where they could attract a higher price. An alternative was to export informally, but that involved additional costs such as bribes (Makame, 2013). In the absence of private traders, producers sold their produce to the National Food Reserve Agency (NFRA), often at a price lower than the price in international markets. The NFRA, a government agency under the Ministry of Agriculture, Food Security and Cooperatives, has the mandate to maintain an optimal level of food reserves in the country in order to address local food shortages and respond to emergency food requirements. The expectation of the government was that nothing on the export ban list would be exported, but the evidence has shown otherwise (MAFAP, 2013), including evidence showing that the export ban was associated with increased informal trade (Makame, 2013). When there was no export ban in 3 The Tripartite Framework was established in 2005 to strengthen and deepen economic integration in southern and eastern Africa through a merger of three regional economic communities the EAC, SADC, and the Common Market for Eastern and Southern Africa (COMESA). These regional communities consist of 26 member countries and account for at least 58 per cent and 57 per cent of African gross domestic product (GDP) and population, respectively (COMESA, EAC and SADC, 2011). The Tripartite Framework is one of the attempts to address overlapping membership through harmonization efforts across the three regional economic communities. 4 The latest export ban was enacted on 16 May 2011, at a time when the demand for food was extremely high in neighbouring countries. The ban was applicable to all staple crops but with a primary focus on maize. Unfortunately, the contribution of maize to inflation is modest or marginal, thus limiting the benefits of the ban to consumers. 3

2007, 2.4 per cent of maize produced in Tanzania was exported; that share was 2.2 per cent when an export ban was in place from 2004 to 2005. 5 Despite lifting the export ban in 2012, the government still controls the export of maize through compulsory export clearance. In addition to the seven export documents listed in the Doing Business Report (World Bank, 2015a), exporters of cereals in Tanzania need to obtain export clearance from the Ministry of Agriculture, Food Security and Cooperatives. Against the backdrop described above, this paper aims to analyse the welfare effect of export bans on cereals in Tanzania on consumers, focusing on consumption at the household level 6. Some studies, such as Diao et al. (2013), examined the welfare effect of the export ban in Tanzania, but this paper also captures the heterogeneity of the country s different regions, which are not uniformly integrated into global markets. Cereals are mainly sold to East African countries; the EAC accounts for 74 percent of Tanzania exports. 7 Given that cereals (and particularly maize, which is a major produced and traded cereal) are mainly exported to East African countries, we distinguish between regions that are more integrated into the EAC market (EAC border regions) and those that are less integrated into this market (peripheral regions). A peripheral region refers to a region that is at least 500 kilometers from the nearest EAC-Tanzania border region. 8 Dabalen and Paul (2014) argue that north and north-west are the only parts of Tanzania that export maize to the EAC, and consequently, the only regions that can be affected by an export ban. Our view, however, is that all regions in Tanzania can be affected by an export ban, and that it is the degree of impact that differs across regions. 9 Border regions are expected to be more sensitive to policy change, as they are more integrated into the regional market than their counterparts. We conduct our analysis using data from the 2007 Household Budget Survey (HBS) (United Republic of Tanzania, 2011), and we focus on maize, which was the main target of the export ban. Maize is among the most important staple foods in Tanzania; according to 2007 statistics from the Food Agriculture Organization, Tanzania was the second largest country in Africa in terms of maize harvested. 2 Production and trade in cereals In 2013, Tanzania exported cereals with a total value of US$40.7 million and imported cereals with a total value of US$400.1 million, resulting in a trade balance deficit of US$359.4 million. Production levels increased from approximately 1 million metric tons in 1961 to 8 million metric tons in 2013 (Figure 1). Production per capita accordingly increased from 0.10 metric tons in 1961 to 0.16 metric tons in 2013. The increase in production is attributed to advancement in 5 The share of exports in total production is computed using data from Barreiro-Hurlé (2012). In any case, the export of maize has never exceeded 7 per cent of the quantity produced in the period from 2000 to 2007. The export share of produced maize was 0.7 per cent in 2006, when export bans were in place. 6 Due to the lack of access to data on cereal prices and household survey data for regions in Zanzibar, our analysis is limited to mainland Tanzania. 7 Author s calculations using data from the International Trade Centre Trade Map, available at http://www.trademap.org/index.aspx (accessed in January 2015). 8 EAC border regions include Arusha, Kilimanjaro, Tanga, Kagera, Mwanza, and Mara; peripheral regions include Lindi, Mtwara, Ruvuma, Iringa, Mbeya, and Rukwa. 9 The level and pace of transmission from international prices to domestic prices varies across regions in Tanzania (Table 4). 4

technology and adherence to recommendable production approaches. Cereal yield in Tanzania is still low and keeps on fluctuating. Cereals that are widely produced include rice, maize, wheat, sorghum, and millet. Top produced cereals include maize and paddy. The country s top maizeproducing regions include Mbeya, Shinyanga, Iringa, Rukwa, Ruvuma, and Morogoro. Maize, which occupies 45 per cent of cultivated area in Tanzania, accounted for approximately 47 per cent of the country s cereal exports and 8.4 per cent of cereals imports in 2012. 10 Maize provides approximately 60 per cent of caloric intake and over 50 per cent of protein content for Tanzanians, and at least 80 per cent of Tanzanians consider maize as their main food item (Saidia et al., 2010). In 2009/2010, annual maize consumption was estimated at 3 million tons, compared with production of approximately 4.7 million tons. 11 Top producing regions for paddy include Arusha, Morogoro, Shinyanga, Rukwa, Mbeya, Mwanza, and Kilimanjaro. Figure 1: Cereal production and yield and land used for cereal production, 1961 2013 Source: World Bank, World Development Indicators 2014. Table 1 indicates that Tanzania s main export destination for cereals is East and Central African countries. Approximately two-thirds of exports in 2013 went to Kenya, Rwanda, and the Democratic Republic of the Congo (DRC). Tanzania is an important source of cereals for the EAC, despite being a net importer of cereals from overseas. Half of cereal imports in 2013 came from Australia, Canada, and the Russian Federation. This implies that Tanzania depends on a limited number of trading partners both in terms of export markets and sources of imports. Imports of cereals to Tanzania from any part of the world except the EAC are subject to a 31.7 per cent tariff, and Tanzanian cereal exports also face tariffs (such as a 20.8 per cent in India, for example). 10 This is established using trade data from the International Trade Centre (accessed in January 2015). 11 Annual production data were obtained from the website of the Ministry of Agriculture, Food Security and Cooperatives (accessed in January 2015). 5

Table 1: Top five export destinations and sources of imports for Tanzanian cereals Maize Rice Wheat Cereals (per cent share in exports/imports) Export destinations Sources of imports Kenya Rwanda Democratic Republic of the Congo Burundi Uganda Zambia Uganda South Africa Italy Malawi Kenya Democratic Republic of the Congo Rwanda Congo Uganda Pakistan Viet Nam India China Egypt Burundi Rwanda Mali Congo Malawi Australia Canada Russian Federation Ukraine United States Kenya (29) Rwanda (21.5) Democratic Republic of the Congo (15.4) India (9.4) Burundi (9.2) Australia (28.6) Canada (16.8) Russian Federation (11.8) Ukraine (8.2) Zambia (6.9) Source: International Trade Centre Trade Map, available at http://www.trademap.org/index.aspx (accessed in January 2015). Food prices are particularly critical for the well-being of poor households, which dedicate a higher share of their budget to food. Figure 2 shows the evolution of domestic and international maize prices. Both prices generally increased between 2002 and 2014. The increase in the international price can be attributed to the increase in the price of oil, population growth, speculation on financial markets, and changing weather patterns (Woldie and Siddig, 2009). Given the nature of Tanzania s domestic market, we are interested in estimating the transmission from international to domestic prices while considering the location of the market. Estimation of the price transmission will allow us to derive some welfare implications of the export bans applied periodically in Tanzania from February 2002 to June 2014. Figure 2: Evolution of maize prices from February 2002 to June 2014 (in Tsh. per 100 kg) Source: Ministry of Industry and Trade; Food and Agriculture Organization of the United Nations, the Food Price Monitoring and Analysis Tool available at http://www.fao.org/giews/pricetool (accessed in January 2015). Note: dp_klm is the price in the Kilimanjaro region, dp_rkw is the price in the Rukwa region, and intp_kg is the international price. Tsh. = Tanzanian shilling. 6

3 Literature survey Export restrictions imposed on products that are consumed in raw form such as maize, rice, and beans are defensive measures pursued by a country to protect consumers. They take different forms, including export bans, export taxes, export quotas, and export-restricting measures of State-run trading enterprises (Kagira, 2011). The export ban on maize in Tanzania was implemented with the assumption that it would increase the quantity of maize available in the domestic market and thus lower consumer prices. Using descriptive statistics, Kagira (2011) shows that the export ban had no effect on the price of maize in the domestic market. Six months after Tanzania put in place its export ban in 2011, the price of 100 kg of maize was virtually unchanged, with a decline of only 0.05 per cent over the period of four months - from Tsh. 42,265 in May 2011 to Tsh. 42,246 in August 2011. The export ban was not sufficient to generate a price reduction because of infrastructural challenges in the Tanzanian market which is highly fragmented and thus causes prices to differ across regions of the country. Woldie and Sidding (2009) use the standard Global Trade Analysis Project (GTAP) model and the GTAP Africa database to simulate the impact of an export ban on all grain products in Ethiopia on food price stabilization and household welfare. Their results show that the price is likely to decline, but that the policy has a negative effect on overall welfare, with a loss of approximately US$148 million. The reason is because the export ban is likely to lead to a reduction in household income and consumption expenditure, and consequently to a decline in GDP. The export ban is therefore found not to be the best policy option to stabilize food prices. Using non-parametric techniques, Calvo (2014) establishes the welfare impact of export restrictions on wheat in Argentina and finds that poor households would be more affected than rich households by the removal of export restrictions and subsidies. She concludes that the welfare gains of export restrictions in Argentina were limited to poor households and that the small gains generated additional costs for the government in the form of subsidies and lost tax revenue. Her results raise doubts as to whether export restrictions were the appropriate instruments to achieve welfare goals in Argentina. In light of this research, it is therefore important to establish the welfare effect of the export ban in Tanzania, as relationships between trade and poverty tend to be country-specific. Using a dynamic general equilibrium model, Diao et al. (2013) find that the maize export ban marginally affected the domestic price in Tanzania. It negatively affected the rural poor, particularly in maize-producing regions, due to reduced return on labour and land. An export ban generally works against long-term development goals, as the profitability of maize production and consequently the incentives to invest in it decline over time. Diao et al. (2013) simulate this export ban using data from the 2007 Household Budget Survey. The same data are also used in this paper but we also consider the location of the domestic market in reference to East Africa to which a large proportion of cereals is exported. Dabalen and Paul (2014) evaluate the price effect of the export ban on maize in 13 markets of Tanzania and Kenya. The six markets in Tanzania are Arusha, Dodoma, Singida, Mwanza, Shinyanga, and Tanga. 12 The authors consider both the wholesale price (a proxy of the producer price) and the retail price (a proxy of the consumer 12 These are the markets for which retail price data were available. 7

price) for the period from January 2006 to February 2014. They find a drop in the wholesale price and a rise in the retail price in Tanzania, indicating a potential welfare loss to both consumers and producers. Their analysis is based on the assumption that the maize exported to Kenya comes from the northern provinces of Tanzania. However, we assume here that the maize exported to East Africa may also come from other regions in Tanzania, including regions in the southern part of the country. The wholesale price is also not a suitable proxy for the producer price because it incorporates costs beyond farm gate prices, such as transportation costs. 4 Methodology and results 4.1 Data This paper mainly uses data from the 2007 Household Budget Survey (HBS) collected by the National Bureau of Statistics and covering the period from 1 January to 31 December 2007 (United Republic of Tanzania, 2011). In total, the analysis includes 10,445 households distributed as follows: 3,457 households in Dar es Salaam, 3,718 households in other urban areas, and 3,270 households in rural areas. Dar es Salaam accounts for approximately 33 per cent of the sample in the 2007 House Budget Survey (HBS) that covered 21 regions in mainland Tanzania. We will consistently use analytical weights in our analysis when applying HBS 2007 data. 13 HBS 2007 provides disaggregated data on consumption expenditure at the household level that include bread and cereals. However, the data do not go into product-specific details that could provide particular information on cereals of interest here wheat (bread), maize, and rice. Wheat is rarely produced and consumed in Tanzania (Maro and Barreiro-Hurle, 2012) and was not the primary target of export bans. Production and price data on wheat in Tanzania are not consistently available to support the intended analysis. Data on other cereals are also problematic. Regional price data for maize are consistently available for 18 of 21 regions in Tanzania, and maize accounted for 47 per cent of cereal exports in 2012. Maize was also the primary target of export bans in Tanzania. This paper will therefore focus on maize which accounts for a large share of cereals. Domestic and international prices of maize are also available. We are not able to disaggregate data on household income, which limits our analysis to the consumption side. Since the export ban aimed to benefit consumers, this paper focuses on the consumption effects of the export ban on maize on household welfare, with particular attention to the gender of the household head and the integration of domestic markets into international markets. 4.2 Income distribution According to Deaton (1989) and supported more recently in a study by Calvo (2014), a wellperforming proxy of household welfare is the logarithm of per capita consumption. The use of consumption as a proxy for household welfare is preferred for two reasons: self-reported 13 Analytical weights were computed as an inverse of each household s selection probability while taking into account the selection of the primary sampling unit and stratification within each primary sampling unit. Adjustments were made to the sample to ensure that the sum of individuals was equal to 2007 project populations (United Republic of Tanzania, 2011). 8

expenditure is more accurate than self-reported income, and households tend to smooth consumption over time. Consumption per capita is computed based on total household consumption and adult equivalent size of households. Both variables are provided in the HBS 2007 dataset. Consumption expenditure is provided on a monthly basis (a span of 28 days) in Tanzanian shillings. In our analysis, we categorize regions into three groups; regions that border the EAC; periphery regions (at least 500 km to the nearest EAC border); and regions that are in between the two groups. Given the nature of Tanzania s domestic market, we expect the degree and speed of the transmission of international prices to domestic prices to vary across markets. Descriptive results are reported for the two groups of interest: EAC border regions and peripheral regions (Table 2). We also grouped households into female-headed and male-headed households. Female-headed households in Tanzania have limited access to productive resources (such as land) and face difficulties in producing food throughout the year (Karatu, Märker, and Mwakolobo, 2011). They were expected to spend more on food than male-headed households. An attempt was also made to group regions based on the ranking in Tanzanian maize production that is, into high-maizeproducing and low-maize-producing regions, with an assumption that demand for maize will be high in regions where less maize is produced. Table 2 provides a summary of household per capita expenditure and the share of total consumption allocated to food and to bread and cereals in 2007 when the survey was conducted. It includes both the share of purchased cereals and the share of total cereals in total household expenditure. Households allocate at least half of their total expenditure to food, and at least 20 per cent of total expenditure to total cereals and 11 per cent to purchased cereals. Urban households are richer than rural households and allocate a smaller share of expenditure to food and cereals. 14 Dar es Salaam household per capita expenditure is higher than the national average and shows the smallest share of expenditure dedicated to total cereals and food. However, the Dar es Salaam household expenditure share of purchased cereals is the highest. EAC border regions are slightly richer and have a higher share of purchased cereals and a lower share of food in total expenditure than peripheral regions. Female-headed households have higher per capita expenditure than male-headed households there is a slight difference between the two groups in terms of the expenditure share of cereals, both purchased and total. Regions that produce less maize have higher per capita expenditure and spend more on purchasing cereals than regions that are the top maize producers. We also decided to group households into five income groups (quintiles) ranging from the poorest to the richest based on total consumption per capita (Table A1 in the Annex). The results indicate that total consumption per capita of the richest households (5th quintile) is 7.7 times total consumption per capita of the poorest households (1st quintile) at the national level. The ratio between consumption per capita of the richest households (5th quintile) and the consumption per capita of the poorest households (1st quintile) is above the national average in Dar es Salaam (8.6 times), urban households (7.9 times), female-headed households (8.1 times), and low-maize-producing regions (8.3 times). On the other hand, the ratio between the 14 We tested whether the four cities other than Dar es Salaam ((Mbeya, Arusha, Tanga, and Mwanza) had any unique feature in income distribution and in cereal and food expenditure. The result, which is not presented here, does not show any significant difference among urban households, as indicated in Table 2. 9

consumption per capita of the richest households and the consumption per capita of the poorest households is below the national average for male-headed households (7.5 times), EAC border regions (7.3 times), peripheral regions (7.2 times), and high-maize-producing regions (7 times). The households in the lowest quintile in Tanzania allocate 76 per cent and 10 per cent of household expenditure to food and to cereals and bread, respectively. The highest quintile allocates 58.5 per cent and 18.9 per cent of household expenditure to food, and to cereals and bread, respectively. Detailed results in Table A1 consistently show that rich households spend relatively less on food and cereals. This confirms the Engel postulate that the share of food in expenditure decreases with increasing income (Calvo, 2014). 10

Table 2: Summary statistics for households by regions and household groups, 2007 Measure National Urban Rural Dar es Salaam region Expenditure per capita (Tsh.) Share of food in total expenditure (per cent) Share of purchased cereals in total expenditure (per cent) Share of total cereals in total expenditure (per cent) Share of cereals in total food expenditure (per cent) Average number of adult equivalents per household EAC border regions Peripheral regions Femaleheaded Maleheaded Highmaizeproducing regions Lowmaizeproducing regions 30,162 38,098 24,479 53,679 29,026 27,986 29,484 32,248 29,007 37,094 70.4 65.6 73.5 58.0 67.6 71.4 69.9 71.8 72.0 66.5 13.4 17.9 11.2 20.1 12.3 11.6 13.5 13.2 11.1 18.8 27.5 23.0 29.7 20.5 21.7 28.7 27.3 27.8 27.2 26.7 38.0 34.1 39.6 34.8 31.4 39.5 38.2 37.6 37.0 39.3 3.8 3.6 4.0 3.1 4.0 3.6 4.0 3.2 4.0 3.3 Number of observations 10,445 3,718 3,270 3,457 2,605 2,082 7,724 2,721 2,961 4,355 Source: Author s calculations, based on data from the 2007 Household Budget Survey. 11

0 Density.2.4.6 0 Density.2.4.6.8 We explore the income distribution of different groups using the kernel density plot of the log of per capita consumption expenditure. Panel (a) in Figure 3 confirms that Dar es Salaam households are better off compared with other areas. Dar es Salaam households are also better off than the national average, as indicated by the shift of the distribution curve to the right. On average, rural households are concentrated in lower quintiles of the income distribution. Panel (b) in Figure 3 shows the income distribution based on the gender of the household head. With the exception of middle-income households, it shows that female-headed households are slightly better off compared with male-headed households. This is consistent with summary statistics reported in Table 2. Panel (a) in Figure 4 presents a comparison of income distribution of regions that border EAC member states and regions that are far from the EAC border. As indicated, households in EAC border regions are slightly better off than households in peripheral regions, but there is no major difference in income distribution between the two geographical areas. Panel (b) in Figure 4 presents the income distribution for areas with different population density. Regions with high population density are slightly better off than regions with low population density. Figure 3: Density estimates of per capita expenditure by area and by gender of the household head (a) All 9 10 11 12 National Urban Dar es Salaam Rural (b) Gender of the household head 9 10 11 12 All Male Female Source: Author s calculations, based on data from the 2007 Household Budget Survey. 12

0 Density.2.4.6.8 0 Density.2.4.6 Figure 4: Density estimates of per capita expenditure by distance to border with EAC countries and by population density (a) Border to EAC 9 10 11 12 All EAC border regions Peripheral regions (b) Population density 9 10 11 12 All High density Low density Source: Author s calculations, based on data from the 2007 Household Budget Survey. For the purpose of this paper, the analysis will be limited to two categories of households: households in regions grouped according to their level of integration into the EAC, and households grouped by the gender of the household head. The analysis of regions with different levels of production of maize and population density is left for future studies. The share of cereals in subsequent analysis in this paper refers to the share of purchased cereals in total expenditure. 4.3 Income level and cereal consumption To assess the relationship between income levels and cereal consumption, we run non-parametric regressions of the share of purchased cereals in per capita household expenditure. Our analysis is 13

6 8 10 12 14 16 limited to household consumption, as data on production of cereals are not available in the 2007 HBS. Non-parametric regressions are useful in this case because they do not require specific assumptions on the data distribution or an econometric specification of the functional form of the relationship between variables (Deaton, 1997; Calvo, 2014). The dependent variable in our case is the expenditure share of cereals and the regressor is the log per capita expenditure. Figures 5 and 6, as well as Figure A1 in the Annex, yield mixed results. While for consumers in Dar es Salaam, urban areas, and EAC border regions the results are consistent with the assumption that poor households spend a higher share on cereals than rich households, poor households in rural and peripheral regions actually spend less on purchasing cereals than rich households. The pattern of cereal consumption at the national level is not so obvious: the poorest and richest households spend less on purchasing cereals than do middle-income households. The difference in consumption between female-headed and male-headed households is generally insignificant, except in rural and peripheral regions where female-headed households consistently spend less on purchasing cereals than male-headed households (see panel (b) in Figure 6 and panel (a) in Figure A1 in the Annex). The Dar es Salam share of purchased cereals is above the national average and all other groupings. The share of purchased cereals for the poorest households is approximately 30 per cent in Dar es Salaam, while it is around 8 per cent at the national level and in rural areas, 7 per cent in peripheral regions, 12 per cent in EAC border regions, and 17 per cent in urban areas. Figure 5: Share of cereals in total household expenditure national and Dar es Salaam, 2007 (a) National 9 10 11 12 14

4 6 8 10 12 14 5 10 15 20 25 30 (b) Dar es Salaam 9 10 11 12 13 Source: Author s calculations, based on data from the 2007 Household Budget Survey. Figure 6: Share of cereals in total household expenditure EAC border regions and peripheral regions, 2007 (a) EAC border regions 8 9 10 11 12 15

5 10 15 (b) Peripheral regions 8 9 10 11 12 Source: Author s calculations, based on data from the 2007 Household Budget Survey. 5 Price transmission and construction of counterfactual scenarios Theoretically, changes in international food prices should be reflected in domestic prices of cereals. However, the extent of this transmission depends on transport and marketing costs, policy measures, local currency valuation, market structures, and the degree of processing of final consumer goods (FAO et al., 2011). Measures such as import duties, export taxes, nontariff barriers, and domestic policies that influence supply or demand of food affect the transmission of international prices to domestic prices. Empirical studies indicate that the degree and speed of price transmission vary across countries and markets within these countries (FAO et al., 2011). The price will be instantly and more fully transmitted in markets that are better integrated into global markets. Thus, in the absence of non-tariff measures such as an export ban on cereals, the movement of the domestic price of cereals in a well-integrated market should follow the pattern of international prices of cereals (Calvo, 2014). At present, cereal exports in Tanzania are free of duties and taxes. 15 As previously indicated, most of Tanzanian exports go to EAC countries. The degree of transmission of international prices to domestic prices is likely to vary across domestic markets, given the fact that those markets are fragmented. The degree of price transmission is expected to be stronger in the EAC border regions than peripheral regions (Dabalen and Paul, 2014). We use equation (1) to assess such potential differences while controlling for yearly and monthly fixed effects that capture yearly macroeconomic shocks and monthly seasonality of prices, respectively. ldp myi = λ 0 + λ 1* lintp my + λ 2* EAC_Border i *lintp my + λ 3* EAC_Border i +δ y + δ m +ε myi, (1) 15 Information available on the Tanzania Revenue Authority website at http://www.tra.go.tz/index.php/exportprocedures. 16

where ldp myi stands for the natural logarithm of the domestic price for month m in year y in region i and lintp my is the natural logarithm of the international price in month m and year y. EAC_Border i is a dummy variable that takes the value 1 in cases where the region borders the EAC and 0 otherwise. 16 We include the interaction term between the natural logarithm of the international price and the dummy for the EAC border. We expect the coefficient of interaction term to be positive, implying that price transmission is expected to be higher in border regions than peripheral regions. The model includes δ y, yearly fixed effects to capture any year-specific factors that could affect price transmission, and monthly fixed effect (δ m ) to capture any seasonal effect across the whole period. It also includes a regional fixed effect to capture any regional specific characteristics. ε myi stands for the statistical error term. We conduct a panel data analysis for 12 regions (six regions bordering the EAC and six regions that are at least 500 kilometers from the nearest EAC border and Dar es Salaam). 17 As suggested by results in Table 3 (columns 3 and 4), the coefficient of the interaction term between the logarithm of the international price and the dummy for the EAC border region is positive, though insignificant. These results provide little evidence regarding the assumption that price transmission is higher in regions bordering EAC countries. 18 Alternatively, we used actual road distance (dist, in km) to the nearest border, which also supports the conclusion that the transmission from international to domestic prices is higher in regions that are closer to the EAC border, even though the relationship is not statistically significant (columns 5 and 6 in Table 3). Table 3: Pass-through from international to domestic prices, 2002 2014 (1) (2) (3) (4) (5) (6) (7) (8) Ldp ldp Ldp Ldp Ldp ldp Ldp ldp Lintp 0.93*** 0.13*** 0.93*** 0.13*** 0.95*** 0.15*** 1.22*** 0.13*** (0.02) (0.02) (0.02) (0.03) (0.02) (0.03) (0.03) (0.03) lintp*eac_border 0.008 (0.03) 0.008 (0.03) lintp*dist -0.0001 (0.0001) -0.0001 (0.0001) lintp*ban -0.44*** (0.05) Ban 4.70*** (0.48) -0.009 (0.03) 0.14 (0.32) 16 EAC border regions refers to geographical regions in Tanzania that physically border any member country of the EAC. Peripheral regions refers to regions that are at least 500 km away from the nearest EAC border region. The definition of these regions is based on the fact that most cereals are transported to Kenya on lorries and trucks (Karugia et al., 2009). 17 Dar es Salaam is the exception in the second category as it is 331 kilometers from the nearest EAC border region. 18 The dummy variable (EAC_border) and the variable dist are omitted for multicollinearity reasons. We had to conduct the Wald test to confirm whether the coefficient of lintp and the EAC-border dummy or distance (dist), or the dummy for the export ban, are jointly and statistically different from zero. The test produced a significance level that is close to zero in all specifications (with EAC-border dummy, distance, and dummy for the export ban). We thus strongly accept that the estimated coefficients are significantly different from zero. 17

Regional fixed effects Yearly fixed effects Monthly fixed effects Constant 0.83*** (0.16) Yes Yes Yes Yes Yes Yes Yes Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes No Yes 9.48*** (0.25) 0.83*** (0.16) 9.48*** (0.26) 0.83*** (0.15) 9.48*** (0.27) -2.29*** (0.35) 9.35*** (0.32) R-squared 0.54 0.81 0.55 0.82 0.52 0.79 0.60 0.81 Observations 2,682 2,682 2,682 2,682 2,682 2,682 2,682 2,682 Source: Author s estimations, based on data from the Ministry of Industry and Trade and the Food and Agriculture Organization of the United Nations. Note: Robust standard errors in parentheses. Dependent variable: logarithm of domestic prices of maize (for a given region/domestic market); *** p < 0.01. We had to establish transmission from international to domestic prices of maize in the presence of the export ban. Theoretically, for well-integrated markets, changes in international prices are expected to match changes in domestic prices in the absence of the export ban. To test this relationship, we run the ordinary least square (OLS) regression of the logarithm of domestic prices on the logarithm of international prices. The model specification is provided in equation (2), where ldp imy is the logarithm of the domestic price of maize in region i expressed in Tanzanian shillings in year y and month m. lintp my is the logarithm of the international price of maize expressed in Tanzanian shillings in year y and month m. 19 It includes the dummy variable for the export ban (ban), which takes a value of 1 if an export ban was observed during the period and 0 otherwise. An interaction term between the logarithm of the international price and the dummy for the export ban is included in the model. We expect the coefficient of the interaction term to be negative, meaning that in the presence of an export ban, a rise in the international price is associated with a decrease in the domestic price. The coefficient on the export ban dummy is also expected to be negative, implying that the export ban is correlated with lower domestic prices and thus beneficial to consumers. The model also includes δ y, which are yearly fixed effects designed to capture any year-specific factors that could affect price transmission, and monthly fixed effects (δ m ), which are designed to capture any seasonal effects affecting a particular market across the year. ε my stands for the statistical error term. This specification is run individually for each region, given previous findings pointing out that price transmission is not uniform in Tanzania. ldp imy = λ 0 + λ 1* lintp my + λ 2 lintp*ban my + λ 3* ban my + δ y + δ m +ε my. (2) 20 Table 4 presents a summary of the average ratio of the domestic price to the international price for selected regions/markets in Tanzania. There are 73 months without an export ban and 76 19 International prices are converted to national currency using the official exchange rate obtained from the Bank of Tanzania. 20 All prices were tested for time series challenges before being used for intended estimation. Data were subjected to a unit root test where all prices were found not stationary but long-term trended. Tests were also done for potential spurious regression. All regressions were found not spurious, except for Mbeya, which is, however, insignificant, and Rukwa where the regression becomes non-spurious with the increase in lags. 18

months with an export ban between February 2002 and June 2014. The data present an inconsistent picture, as in most cases consumers pay more in periods with export bans than what they pay in periods with no export bans. Such inconsistency is common in EAC border regions. The ratio of domestic prices to international prices varies across regions. It appears to be higher in EAC border regions than in peripheral regions, implying that consumers in border regions pay more than their counterparts for the same amount of maize. Frequently, consumers in EAC border regions pay more than the average international price. Table 4: Ratio of domestic to international prices of maize, 2002 2014 EAC border regions Peripheral regions Duration Ban Dar es Kilimanjaro Mwanza Tanga Iringa Mbeya Rukwa Lindi Salaam February 2002 - No 95 77 76 80 83 81 77 88 June 2003 July 2003 - Yes 127 139 139 128 94 89 87 135 January 2006 February - July No 130 137 147 133 119 100 84 113 2006 August - Yes 79 71 108 59 79 68 62 76 December 2006 January - No 65 62 84 62 48 54 53 80 December 2007 January 2008 - Yes 148 145 150 144 124 117 101 156 October 2010 November 2010 - No 119 106 134 125 95 105 87 129 April 2011 May - October Yes 109 114 134 94 76 87 67 83 2011 November 2011 - June 2014 No 142 135 154 127 105 115 98 134 Source: Author s calculations, based on data from the Ministry of Industry and Trade and the Food and Agriculture Organization of the United Nations. Subject to domestic price data availability, our analysis is limited to 18 of 21 regions; price data for three regions (Kigoma, Manyara and Pwani were missing for some years/months. Detailed results for equation (2) are presented in Tables A2 A5 in the Annex. The value of R-squared is higher in cases where fixed effects are considered, which means that year-specific factors and seasonality are relevant control variables in estimating the transmission from international to domestic prices of maize in Tanzania. Table 5 provides a summary of results, particularly the coefficient on the export ban dummy and the coefficient on the interaction between the export ban dummy and the international price. Subsequent discussion dwells on specification (2) which includes yearly and monthly fixed effects; specification (1) does not include yearly and monthly fixed effects. The effect of the export ban on domestic prices is directly captured by the coefficient on the export ban dummy (λ 3 ) and indirectly captured by the coefficient on the interaction between the export ban dummy and the international price (λ 2 ). The signs of these coefficients are consistently opposite, implying that the two effects counteract each other. As expected, results vary significantly across regions, and both coefficients are statistically insignificant in most regions. The two coefficients are only statistically significant at the 5 per cent level in Rukwa, 19

Singida, and Mbeya, and the coefficient on the interaction term is statistically significant at 10 per cent in Mtwara. To obtain the net effect on domestic prices, we multiply the coefficient on the interaction term with the average change in international prices during the months with an export ban (1.53 per cent) and add it to the coefficient on the export ban dummy. The aggregate coefficient (λ 2 *1.53+λ 3 ) shown in Table 5 is computed from specification (2) is positive in Kilimanjaro, Iringa, Lindi, Mwanza, Singida, Tanga, Arusha, Kagera, Dodoma, Morogoro, Tabora, and Shinyanga, and negative in Dar es Salaam, Rukwa, Mbeya, Mara, Ruvuma, and Mtwara. A positive aggregate coefficient indicates that the export ban was correlated with an increase in the domestic price, which suggests a welfare loss to consumers, while a negative aggregate coefficient implies that the export ban was correlated with a decrease in the domestic price, thus suggesting a welfare gain to consumers. Our results do not strongly show that the export ban was beneficial for consumers in Tanzania. The export ban had heterogeneous effects on Tanzanian regions: consumers in six regions benefited from it, while consumers in 12 regions suffered. Consumers in all EAC border regions except Mara suffered from the export ban, while consumers in all peripheral regions except Iringa and Lindi benefited from it. Table 5: Summary results: Coefficient on the export ban dummy (λ 3 ) and coefficient on the interaction between the export ban dummy and the logarithm of the international price change (λ 2 ) Specification (1) (2) Aggregate coefficient Region/Coefficient λ 2 λ 3 λ 2 λ 3 (λ 2 *1.53)+ λ 3 Dar es Salaam -0.35*** 3.74*** 0.02-0.20-0.1694 (0.10) (0.99) (1.18) Kilimanjaro -0.64*** 6.69*** (1.24) -0.07 (0.09) 0.81 (0.94) 0.7029 Rukwa -0.19* (1.00) Iringa -0.09 Lindi -0.57*** Mwanza -0.64*** Singida -0.64*** Tanga -0.49*** (0.14) Mbeya -0.18** (0.09) Arusha -0.54*** Kagera -0.48*** (0.13) Mara -0.52*** Dodoma -0.53*** 2.02** (1.00) 1.07 (1.10) 5.92*** (1.28) 6.71*** (1.23) 6.77*** (1.24) 5.15*** (1.41) 1.91** (0.90) 5.68*** (1.11) 5.08*** (1.39) 5.45*** (1.17) 5.64*** (1.20) 0.21** -0.05 (0.09) -0.05-0.07 (0.08) -0.26** -0.06 (0.13) 0.20** (0.08) -0.08 (0.10) -0.02 0.007 (0.07) -0.099-2.13** (1.05) 0.53 (0.90) 0.53 (1.19) 0.85 (0.77) 2.85** (1.16) 0.63 (1.29) -2.04** (0.78) 0.88 (1.00) 0.21 (1.06) -0.05 (0.68) 1.09 (1.14) -1.8087 0.4535 0.4535 0.7429 2.4522 0.5382-1.734 0.7576 0.1794-0.03929 0.93853 Ruvuma -0.05 0.63 0.18-1.80-1.5246 20