Determinants of Structural Changes of Food Exports from Developing Countries

Over the past three decades, there has been a rapid expansion of processed food exports in developing countries, replacing traditional agriculture exports such as coffee and tea. However, this development and its policy implications have received little attention in the literature. This paper aims to redress this oversight by providing an overview of key characteristics and growth patterns of processed food exports in developing countries. The determinants of structural changes toward processed food exports in developing countries are examined using panel data econometric analysis. The results suggests that trade policy openness, large domestic market, good macroeconomic management especially in terms of price stability, as well as adequate financial support and infrastructure are the key factors that influence the structural changes toward processed food products.


I. Introduction
There has been a structural change in the composition of food exports from developing countries over the past three decades. Traditional (unprocessed) food exports have continuously declined and have been replaced by processed food exports. Expansion of processed food production and exports has become an important trend for developing countries since such exports have intrinsic characteristics of manufactured goods. The structural change in food exports is believed to generate superior growth performance. Reasons for this belief include employment generation, knowledge spillovers, and terms of trade gains. In particular, processed food production, especially in the final stage, appears to be labor-intensive. An expansion of processed food production could, therefore, generate positive effects on employment.
Processed food exports from developing countries are expected to continuously expand in the future since world demand for processed food continues to grow in response to diet upgrades resulting from rising incomes, growing health consciousness, and urbanization. Factors such as international migration, communication revolutions, and international tourism and trade liberalization initiatives under various rounds of world trade negotiations also contribute to an expansion of demand for processed food. Thus, ability to expand processed food production and exports into the world market of developing countries would provide an important impetus for their economic development.
However, not all developing countries so far have benefited from the growth of world demand in processed food. Why could only some developing countries move from traditional agriculture exports toward processed food exports? This study, therefore, aims to examine factors that explain intercountry differentials in structural changes of food exports in developing countries. To the best of our knowledge so far, this issue has received little attention in the literature on export-led industrialization in developing countries (Jaffee and Gordon 1993, Athukorala and Sen 1998, Athukorala and Jayasuriya 2003, Rae and Josling 2003, Regmi et al. 2005. It is only this study that quantitatively examines factors determining structural changes of food exports in developing countries. In addition, we apply the econometric results with the recent data to formulate a prudential policy in enhancing supply-side capability of developing countries, particularly in the Asia and Pacific countries, to reap benefit from potential world demand expansion in processed food products.
The rest of the paper is organized as follows. Section II presents key characteristics of processed food. Trends and patterns of processed food exports in developing countries over the past three decades are discussed in Section III. Section IV discusses determinants of intercountry differentials in structural changes of food export. Data, variable measurements, and econometric procedure are presented in Section V. The estimation results and implication for developing Asia are discussed in Sections VI and VII, respectively. The final section provides conclusions and policy inferences.

II. Key Characteristics of Processed Food: First Look
Processed food products retain the general characteristics of food products, particularly perishability, bulkiness, and uncertainties related to seasonality of production. Commodity perishability could lead to product loss and value decline during transport and storage. Transaction costs could increase dramatically since these commodities would be required to be repeatedly screened for quality at each level in the commodity system. Commodity perishability could also limit market flexibility of producers and increase their market risk. Because of their bulky and perishable characteristics, physical handling and transportation problems could become issues of concern. Investment in highly specialized and "lumpy" transport and storage facilities and equipment are necessary to lower costs of food production.
Another common set of food production characteristics concerns yield uncertainties and seasonality of production. The production of most food crops and animal products is dependent upon the life cycle of plants and animals. Raw materials of food production also face weather uncertainties and the possible incidence of plant diseases or pests. Adverse natural events could undermine total supply, resulting in farmer losses, underutilized marketing and processing facilities, and unmet consumer demand. These characteristics tend to create problems for cost-efficient utilization of transport and processing facilities. Thus, medium-term planning, including financing, seems to be crucial to ensure supply could meet consumer demand.
However, processed food products have four unique characteristics that differentiate them from traditional food products. First, food commodities exhibit considerable variability in their quality and multiple quality attributes, some of which are difficult to measure and most of which are valued and weighted differently by specific groups and consumers. Consumers, therefore, cannot be expected to have full knowledge about the choices and quality of products available to them. These characteristics tend to create asymmetric information related to product quality, leading to a reduction in an ability to use market prices as a signal for the quality of products. Suppliers of processed food face challenges in creating and preserving the unique characteristics of their products and conveying information about those characteristics to consumers. Because of asymmetric information between producers and consumers, regulations need to be established to protect consumer from potentially harmful products and practices. These are related to certain tests, inspection of products, certain handling or processing procedures, requirements for nutritional information and labeling, and truth in advertising.
Second, processed food products tend to rely on numerous food supply chain. Broadly, the supply chain could be divided into five key processes, which are input supply, primary production, processing manufacturing, wholesalers/retailers/exporters, and consumers. 1 There are three main types of flows relating to the supply chain, namely physical product, financial, and information flows. With regard to the physical product flows, such flows generally are uni-direction, starting with input supply and ending with consumption. By contrast, financial resources tend to move in the opposite direction, with payments going to upstream suppliers as products move downstream toward consumers. In some cases, suppliers who retain ownership of their products as they move through the supply chain may pay fees to downstream firms for production, processing, or distribution services. Information flows move in both directions throughout the supply chain. Suppliers tend to convey information about product attributes and availability to downstream customers, while receiving information about product demand, product inventories in downstream segments, and consumer reactions to product attributes.
It is noteworthy that supply chains in processed food products must be operated in a broad environment characterized by infrastructure and institutions ( Figure 1). 2 Infrastructure is a key element in facilitating food production and supply chains while institutions provide and exercise rules and regulations on both producers and consumers in the marketplace. A wide range of infrastructure is required in the food supply chain, including transportation, telecommunications, and multipurpose technologies for packaging and product preservation. With regard to institutions, they could be established through international organizations, national/local governments, or nongovernmental organizations such as trade associations (Jaffee and Gordon 1993). Such organizations establish laws and regulations that govern commercial practices, food safety and product quality, product packaging, trade, labor practices, and intellectual property.
Note that all five segments are relevant for every supply chain and, in many cases, this process could be further divided into separable processes. 2 Infrastructure quality and institutions can have far-reaching impacts on supply-chain design for processed food products, and cross-country differences can significantly affect supply-chain configuration and the geographic scope of product distribution. Suppliers of branded food products may choose not to enter markets where infrastructure to support advertising is lacking and legal institutions that protect brand trademarks are weak.

Figure 1: Key Element of Processed Food Supply Chain
Source: Regmi et al. (2005).
Third, an established domestic market tends to be a precondition for processed food production and export success. As processed food products are classified as "luxury" consumer goods, i.e., with higher income and price elasticities of demand relative to other primary products, the production for the domestic market must be lucrative enough to enable firms to achieve economies of scale and thus reduce costs to break into foreign markets. Based on evidence provided by Jaffe and Gordon (1993), successful exporters often depend upon prior or parallel development of domestic markets. None of the local commodity systems has developed as an export-oriented enclave and relatively few have relied purely upon export markets for their sales.
Finally, processed food production and exports are likely to create spillover effects in developing countries. Processed food production, especially in the final stage of production, appears to be labor-intensive unlike in the case of further processing of resources such as timber or minerals that tend to rely more on capital equipment. Such characteristics, to some extent, imply that an expansion of processed food production could generate positive employment effects. In addition, processed food production tends to rely more on domestic resource content, unlike conventional manufacturing production that tends to rely more on import content. This naturally implies higher spread effects through input linkages in the domestic economy. While processed food industries tend to be closely related to activities in the rural sector, an expansion of processed food production could generate higher income for the poor and to some extent lead to poverty reduction. On top of that, knowledge spillovers, especially for export firms, could be generated. Such spillovers would arise mainly from learning through interaction with foreign buyers, exposure to foreign technology, and improving quality standards in the face of stringent export competition. Meller (1995) points out that processed food exports might be able to generate knowledge spillovers on a scale to be comparable, or superior to, exports of labor-intensive manufacturing products.
Note that processed food products may be divided into two subcategories according to their specific characteristics, namely food processing (land-based foods) and manufactured food. Food processing refers to products that do not undergo major changes from their basic raw commodities, i.e., the characteristics of the original materials are retained. This type of food product includes not only all raw commodities, such as fruits and vegetables but also preserved products such as preserved (frozen/ canned/dried) fruits, vegetables, meat, and dairy products. The production location of land-based processed food tends to be influenced by product perishability, transportation costs, and geography so that production is typically located near growing areas to minimize spoilage, transaction, and production costs.
Manufactured food refers to products that have significantly lost characteristics of agriculture raw materials from transformation, e.g., confectionery and bakery products. Such a transformation includes not only blending and fermentation but also cooking. Since technology and capital in producing manufactured food products are mobile in the world food market and raw materials of these products, e.g., refined sugar, starches, wheat and grains, are relatively nonperishable and inexpensive to transport, their production location does not need to be tied closely to the presence of natural resources. Production of manufactured food is, therefore, less location-specific than food processing products. To minimize distribution costs, manufactured food products tend to be located close to consumer markets, i.e., demand-oriented location decision so that manufacturing activities tend to be dispersed across the globe.

III. Trends and Patterns of Processed Food: First Look
Over the past three decades, there has been a notable composition shift in world food trade. The relative importance of "classical" food products such as coffee, tea, sugar, and cocoa has been replaced by the processed food trade (see Appendix 1 for a list of processed food products). 3, 4 The share of processed food exports in total world food exports increased from 44% in 1980 to around 63% in 2006. The compositional shift has been contributed to mainly by developing countries, particularly since the early 1990s. While the share of processed food exports in total world food exports tripled in developing countries during 1980-2006, the share was rather stable in developed countries ( Figure 2). 5 The increasing importance of processed food exports has also resulted in a structural shift in world agricultural trade. The share of processed food exports in world agricultural exports increased to 51% in 2006 from only 32% in 1980 ( Figure 3). While international trade in many of these is not entirely "new", their trade has experienced very rapid expansion in recent years, and they are often described as "new food exports" or "nontraditional food exports". To maintain the focus on these new dynamic export items, the traditional beverages (such as tea and coffee) and cereal grains (wheat, maze, rice, etc.) exported in bulk are excluded from our analysis. 4 Generally, the definition of processed food products is based on the International Standard Industry Classification (ISIC). All commodities that belong to ISIC Section are all classified as processed food. However, export data used in our analysis are reported under the Standard International Trade Classification (SITC). Thus, the SITC commodities listing at 5-digit level is used in cross reference to that of the ISIC listing at 4-digit level. See Athukorala and Jayasuriya (2005) for detailed discussion of processed food definition. 5 Developed countries refer to high-income countries according to the World Bank classification. Note that the results are not significantly different when developed countries are defined to include only Australia, Canada, EU5, Japan, New Zealand, United States, and the East Asia tigers. Developing countries refer to low and middle income countries according to the World Bank classification. In the wider context of structural shifts, processed food exports have still been able to retain their importance in world merchandise trade. It is clear that world merchandise trade over the past three decades has been characterized by a dramatic expansion of manufacturing exports and a sharp decline in agricultural exports. The share of manufacturing exports in total world merchandise exports rose by 1.2 times from 59% in 1980 while the share of agricultural exports declined to less than 10% in 2006 from 16% in 1980. The proportionate decline of agricultural exports tended to be sharper for developing countries (from almost 30% in 1980 to 9% in 2006), compared to developed countries (from 15% to 7%). Interestingly, the share of processed food exports in total merchandise exports was relatively stable. In developing countries, the share was around 7-8% during 1980-1996 before slightly declining to around 4-5% after 2000 ( Figure 4). This implies that processed food exports have grown faster than other exports of agricultural products, especially primary food; but growth has still been slower than manufacturing exports ( Table 2). As pointed out in Athukorala and Jayasuriya (2005), the faster growth in manufacturing exports than in processed food is a result of the rapid expansion of parts and components trade underpinned by the broader process of product fragmentation. This led to the high import content of products involved and expansion of manufacturing trade. They argue that if net exports (gross exports -imported inputs) are estimated, the export growth of processed food would turn out to be much higher.
Note that in 2006, the growth rate of processed food tended to be lower than other compositions of agriculture products, agricultural raw materials in particular, so that the average growth rate of the latter during 2001-2006 was higher that that of the former. This could be a result of higher demand for producing biofuel products in response to the oil price hike.   Although processed food exports in developing countries have continuously increased, not all countries have shared in the benefits. In general, countries belonging to uppermiddle and middle-income countries according to the World Bank classification have performed better in expanding processed food exports than low-income countries. Figure 6 shows that more than 90% of total developing countries' processed food exports are contributed by upper-middle and lower-middle-income countries. Processed food is an important component in total food exports in upper-middle and lower-middle-income countries. Processed food exports in these countries accounted for more than half of total food exports. After the late 1980s, the gap between upper-middle and middle-income countries in contributing to processed food exports has narrowed. This resulted from the faster growth of processed food exports in lower-middle-income countries. On average, the annual growth rate of processed food exports in lower-middle-income countries was 10% during 1980-2006, compared to 11.2% in upper-middle-income countries and 7.1% in low-income countries.
Among upper-middle income countries, Chile, Malaysia, Mauritius, Oman, and Seychelles performed better than the other countries (Table 3). Particularly, processed food accounted for almost 100% of total food exports in Oman and Seychelles. The share of processed food exports in agricultural products in Seychelles doubled during 1986-2006 while the share of processed food exports to total exports also noticeably increased. Among middle-income countries, Algeria, Bolivia, People's Republic of China (PRC), Paraguay, Peru, and Syria tended to outperform the other countries. The annual growth rate of these countries was more than 10% and the share of processed food exports to total exports in these countries, except for the PRC and Peru, has increased over the past three decades. The higher share of processed food in total exports was also evident in other middle income countries such as Colombia, El Salvador, Fiji Islands, and Guatemala. The growth rate of processed food in these countries, thus, tended to be higher than other agriculture products (including unprocessed food), and in some countries (e.g., Guyana and Paraguay), the growth rate was even higher than that in the manufacturing sector.
Even though the overall growth rate of processed food exports in low-income countries was lower than in upper-middle and middle income countries, Madagascar is a notable exception. The growth rate of processed food exports in Madagascar was around 10% during 1986-2006. Meanwhile, processed food exports are the key driver of food and agriculture exports in many low-income countries. In Bangladesh, processed food exports accounted for more than 95% of all food exports during 1986-2006 while the share of processed food exports in agriculture exports doubled during this period. These patterns tend to suggest that income per se is not able to explain intercountry differences in export performance of processed food products. In terms of the regions, developing Asia and Latin America tended to perform better than other regions in expanding processed food exports. The shares of processed food exports from Asia and Latin America in total developing countries' processed food exports, over the past three decades, were around 41% and 38%, respectively ( Figure 7). The contribution of processed food exports from Middle East was the lowest, around 2% while Africa and Europe contributed 4% and 14%, respectively. While the contribution from Asia was rather stable over the past three decades, structural change tended to occur within the region ( Figure 8). Processed food exports from Southeast Asia, the highest contribution over the past three decades, have shown a declining trend since the early 1990s, and in 2006, Southeast Asia has been replaced by countries in East Asia. The share of processed food exports in Southeast Asia in total developing countries' processed food exports declined from 30% in 1985 to 25% in 2006 while the share of East Asia increased to 29%, from 14% during the same period. The share of processed food exports in South Asia has been relatively stable over the past decade but it noticeably declined in the early 1980s. The key products that contributed to most processed food exports were processed fish, followed by processed fruits and meat products (Table 4). Processed fish exports were relatively less important in the early 1980s (18% of developing countries' processed food exports), compared to meat products, but their importance has become noticeable since the late 1980s. In 2006, the share of processed fish was 23% of total developing countries' processed food exports, followed by 21% of processed fruits. PRC, Chile, and Thailand were the key exporters of this product. In 2006, the PRC accounted for more than 25% of total developing countries' processed fish exports while Thailand and Chile accounted for 15% and 9%, respectively. However, over the past 5 years, the contribution of processed fish has declined. This may be attributed to the global economic slowdown, resulting in a shift in consumer preference toward meat products exports, which are presumably cheaper than processed fish. 6 Brazil accounted for almost half of total developing countries' meat products exports, followed by the PRC (10%) and Poland (9%). Processed fruits exports have also increased over time, though not as spectacularly as in the case of processed fish. On the other hand, shares of "traditional" items such as coffee extract and animal and vegetable oil have fallen. In 1980, coffee extract accounted for 15% of total processed food exports from developing countries but in 2006, the share was only 4% of total processed food exports.    Developed countries are the key export destinations of processed food exports from developing countries. Among developed countries, the G3 countries (European Union [EU], Japan, United States [US]) accounted for more than 50% of total developing countries' processed food exports. The EU is the most important export destination, followed by the US and Japan, respectively (Table 5). Since the early 1990s, the importance of the G3 markets has declined and developing countries have become more important for developing countries' processed food exports. The share of processed food exports in developing countries doubled during 1990-2006 while the share of G3 markets declined by almost 10 percentage points. Latin America, Asia, and Europe have become the key importers of developing countries' processed food products. These three markets accounted for 25% of total developing countries' processed food exports in 2006. For Africa and the Middle East, the shares slightly increased but were still less than 5% of total developing countries' processed food exports.

IV. Determinants of Structural Changes in Food Exports
This section reviews potential factors that explain intercountry differentials in the structure of food exports in developing countries (PROEX). Specifically, we review potential factors that could alter the share of processed food exports in total agricultural exports. We use total agricultural exports as the denominator instead of total food exports because exports of unprocessed food and other agricultural products tend to behave similarly to each other. The first factor that can affect a structural change is the domestic market (CS).
As mentioned in Part II, processed food products are relatively luxurious, compared to traditional agricultural products. Thus, to enable firms to achieve economies of scale and to reduce costs to break into foreign markets, the domestic market must be large enough to support economies of scale. Thus, an expansion in the size of the domestic market leads to an increase in processed food exports.
Trade policy regime (OPEN) in a country is another variable that can have an impact on structural change in food exports. Trade liberalization provides incentives for producers to export instead of selling in the domestic market. Although tariff and nontariff barriers in the agricultural sector, including processed food, declined after the Uruguay Round of 1986-1994, there are no clear theoretical and empirical studies to show that trade liberalization would affect processed food exports more than other agriculture products. Thus, the direction of the effect of the trade policy regime on structural change of food exports is a priori inconclusive. (See Box 1 for tariffs on processed food products.) A number of trade facilitation measures could also help to support an expansion of processed food more than other agricultural products thereby increasing the share of processed food in total agricultural exports. These include a well-functioning financial market (DC) to provide financial support to all relevant supply chains of processed food industries. Sufficient financial support allows firms to better manage risks and uncertainties, mainly related to transport and storage of raw materials and commodities, and improve production and distribution technologies. Infrastructure (INFRA) is another key variable that can determine export performance of processed food industries. This includes well-developed roads, railways, ports, telephone lines, power systems, terminal markets, storage, and processing facilities. While processed food production involves more processes than other agricultural production, an improvement in infrastructure could provide greater benefits to the former more than the latter.
In addition to trade facilitation, favorable macroeconomic conditions also play a pivotal role in shaping and influencing incentives for investment in production and marketing activities in processed food industries (Jaffee and Gordon 1993). In particular, price instability (PIS) and an overvalued real exchange rate (RER) could induce higher costs of production and lower returns on farm inputs and food products, thereby reducing investment incentives and possible expansion of food exports. However, while these factors could also affect other agricultural products, their effect on the share of processed food exports to total agricultural exports is inconclusive.
Finally, while the behavior of processed food exports resembles labor-intensive manufacturing, foreign direct investment (FDI) could play an important role in this industry. This would be in contrast to traditional agricultural production in which there is no significant involvement of FDI. However, the impact of FDI on processed food industry is ambiguous. On the one hand, an involvement of multinational enterprises (MNEs) could generate positive effects to processed food industries, particularly exporting firms.
MNEs have an international production network so that flows of information on home country and other markets are completed. In addition, they tend to undertake a large proportion of the world's total research and development and are principal bearers of technology across international borders (Borensztein et al. 1998, Lipsey 2000, Vernon 2000. With these advantages, one would expect that MNE affiliates are likely to face lower production and entry costs in export market. However, technology and capital in producing manufactured food products are mobile in the world food market and raw materials of these products are relatively inexpensive to transport. MNEs may, therefore, intend to locate closely to consumer markets to minimize distribution costs so that an increase in FDI could lead to an overall reduction of processed food exports.
All in all, the empirical model of processed food exports could be written as follows:

V. Variable Measurements and Econometric Procedure
The empirical model is estimated based on 79 developing countries 7 during the period 1990-2006. The country coverage is based on data availability. The whole sample is divided into six nonoverlapping 3-year periods (except for the last subgroup for which the data are averaged only from 2-year periods). Three-year periods are applied, instead of annual data, to reduce business cycle fluctuations associated with these data series.
The dependent variable is measured by the share of processed food exports in total agriculture exports to reflect a structural change in agricultural exports (PROEX).
The expansion of the domestic market (CS) is measured by gross domestic product (GDP) per capita in which GDP is measured in real terms at 2000 US$ prices. While there is no unique measure of trade policy regime, this study applies two well-known proxies, namely trade to GDP (OPEN1) and implied tariff rate (OPEN2) (total tariff revenues as a percentage of total trade) in measuring trade policy openness. Tariff revenues are composed of customs and other import duties, taxes on exports, profits of exports or import monopolies, exchange profits, exchange taxes and other taxes on international trade, and transactions collected on a cash basis. In fact there are other proxies for trade policy regime such as binary index, which takes the value 1 for open economies and zero otherwise, originated by Sachs and Warner (1995) 8 and the ratio of merchandise trade to good (agriculture and industry) GDP, excluding nontraded activities. However, because of incomplete and not updated data, 9 this study applies only the former two measures to reflect trade openness.
The ratio of private domestic credit over GDP (DC) is used to proxy the availability of financial support to all relevant supply chains in processed food industries. Price instability is measured by the deviation of agricultural prices (deflator) to their trends (PIS). The trend of agricultural prices is derived from the Hodrick-Prescott filter method. (iv) a socialist economic system; and (v) state monopoly on major exports. 9 The binary index was calculated only during the period 982-994.

Box 1: Tariffs on Processed Food Products
In the Uruguay Round of 1986-1994, the Agreement on Agriculture, which subjects agricultural trade to stronger international disciplines, was adopted. The member countries agreed to convert their agricultural nontariff barriers to tariffs (tariffication) and agreed to reduce agricultural tariff rates. In particular, developed countries agreed to reduce agriculture tariffs, including those resulting from tariffication, by a total of 36%, on simple average, from their base-period rates, with a minimum cut of 15% for each tariff (Regmi et al. 2005). In response to the tariffication process, tariffs in many categories were still so high that it is difficult to see any profitable trade opportunities in such markets (Rae and Josling 2003). The new rounds of agricultural trade negotiations began in March 2001. The use of a tariff-cutting formula, a second round of the 36% average tariff reductions of the Uruguay Round, and the elimination of tariffs on specific commodities were the key topics of discussion. These discussions have continued but progress has been limited, particularly in terms of the formula used to reduce agricultural tariffs. In addition, compared to industrial tariffs, agricultural tariffs remain a major distorting component and global average tariff rates are still higher than manufacturing tariff rates by around 10 times.
The Uruguay Round tariff cuts were by and large confined to bulk agricultural commodities while there was no substantial tariff cut on processed food products. As shown in Table B1, recent tariffs on processed food of the EU, Japan, and US were on average more than 10% while those of bulk agricultural commodities were less than 8%. Higher duties imposed on processed food than bulk commodities result in tariff escalation, i.e., duties imposed by an importing country increase with the level of processing. Such a tariff structure in developed countries would encourage raw materials imports from developing countries, at the expense of processed food exports. Importers in developed countries protect their domestic processing industries in order to capture value added locally. Developing countries seeking to export processed foods have been vocal in supporting efforts to decrease tariff escalation (Rae andJosling 2003, andRegmi et al. 2005). Tariff escalation is also evident in developing countries. The gap between tariffs on imports of unprocessed commodities and on processed commodities in Thailand tends to be higher than other developing countries. In Thailand, while tariffs on processed food were around 27%, they were 19% for the unprocessed commodities. For other countries, the gaps between these two tariffs were around 3.3 percentage points. The tariff escalation structure in developing countries could provide the perception that developing countries are still protecting their processing industries, compared to the unprocessed (primary) commodities. Interestingly, tariff escalation is not evident in India. The unprocessed (primary) food tended to have a higher tariff rate than processed food. This tariff structure may hurt processed food production as the effective rate of protection in the processed food industry may become negative. 1

Box
The existing tariff structure has implications for further tariff cuts in the Doha Round and beyond. In 2003, a World Trade Organization (WTO) proposal recommended that "the rate of tariff reduction for the processed product shall be equivalent to that for the product in its primary form multiplied, at minimum, by a factor of 1.3 percent" (WTO 2003). However, no specific formula has been adopted yet. In 2004, the eventual framework adopted by the WTO General Council pointed out that tariff escalation will be addressed through a formula to be agreed (WTO 2004).
Even though there is no adopted specific formula, the relatively low levels of tariffs on raw materials and traditional (unprocessed) food indicate that any substantial cut in tariffs have to come from those on processed food products. This may provide opportunities for developing countries to obtain wider access into developed countries' markets. However, as pointed out in Athukorala and Jayasuriya (2005) that the net market access gains from such tariff cuts would depend crucially on what happens in the sphere international food safety regulation. Importing countries, particularly developed countries, have ample room to restrict market access by relying on stricter food safety standards while honoring promised tariff cuts. Even if there is genuine market opening, many developing-country exporters may not be in a position to reap benefits because of supply-side constraints on meeting food safety standards.
The effective rate of protection would be useful to reveal the structure of protection of the processed food industry in India.
used as a proxy of price differentials. FDI is measured as net inflows of foreign direct investment as a percentage of GDP. 10 Road density (ROAD), which is measured as total road networks divided by total population, is used as a proxy of infrastructure. 11 Data on processed food exports and agricultural exports are from the United Nations Comtrade database (UNCOMTRADE), Revision 2 (Rev. 2) while real GDP per capita, agricultural products, agricultural price deflator, consumer and producer prices, trade to GDP, tariffs, total import value, total network of roads, private domestic credit, FDI, and total population are compiled from the World Development Indicators (CD-ROM) of the World Bank.
To examine the determinants of structural changes in food exports, an unbalanced panel econometric procedure is applied for six nonoverlapping 3-year periods, 79 developing countries with 262 panel (unbalanced) observations, during 1990-2006. The unit root test for panel data is first performed to ensure that there is no unit root for all dependent and independent variables. Because of stationarity of the data (I(0)), the level of both dependent and independent variables can be used without concerns on spurious regression. Both fixed and random effects are performed in this study. Equation (1) is rewritten in terms of fixed and random effects models as follows: , , , , , where β i is the cross-sectional fixed effect for processed food exports of country i, to control for country specific characteristics, 12 χ t is the time effect to control for timespecific shocks. Inclusion of the latter is to capture some time varying variables that may be missing from the simple specification in equation (2), and ε i,t is the independently and identically distributed error terms across countries and years. 0 Because of data limitations, we opt to use total FDI data to proxy the role of MNEs in food industry. Non-FDI channels that could also perform by MNEs are not included in this paper. See also footnote 6 for non-FDI channels.
Note that other infrastructure variables such as electric power consumption and transport services to total trade are also included but it becomes statistically insignificant and assumes an incorrect sign. This could result from small sample sizes of these variables leading to specification error. 2 Note that world demand would be included in the cross-sectional fixed effect as this variable does not vary significantly across developing countries. Table 6 reports the estimation result of the panel model. Random effect estimation is a preferred technique to fixed effect estimation in this study because the former model performs better in terms of key diagnostic tests, particularly normal distribution and stationarity of the error terms. The estimation results are corrected for serial correlation and heteroskedasticity. A limitation of the random effect estimator, compared to the fixed effect counterpart, is that it can yield inconsistent and biased estimates if the unobserved fixed effects are correlated with the remaining component of the error term. However, this is unlikely to be a serious problem in this study because the number of explanatory variables (N) is larger than the number of "within" observations (T) (Wooldridge 2002, chapter 10). The Hausman test could not provide an appropriate measure in choosing between random and fixed effects in this study because the model tends to violate two key assumptions of applying the Hausman test, namely strict exogeneity and homoskedasticity (Wooldridge 2002, chapter 10). Note that two-stage least squares is applied in this study to redress the possibility of simultaneity problem that could emerge between real GDP per capita and processed food exports. 13 Natural logarithms are applied to all variables. Because of better explanatory power and diagnostic tests, the estimation result reported in Table 6 is based on the model in which trade to GDP and the GDP deflators are used as proxies of trade policy regime and the RER, respectively. 14 The statistically positive significance of real GDP per capita (CS) in the share of processed food exports in total agricultural exports (PROEX) supports the hypothesis that fast growing developing countries are relatively well placed to benefit from emerging trading opportunities in processed food (see Table 6, column B). This is because processed food products are rather luxurious in nature, compared to traditional food products, so that a large domestic market is needed to enable firms to achieve economies of scale and to reduce costs more easily to break into foreign markets. In addition, while production technology in improving and controlling the quality of processed food products tends to be more advanced in high-and middle-income groups, the ability to access world market becomes easier as various costs are reduced. A 1% increase in real GDP per capita would increase the share of processed food exports in total agriculture exports (PROEX) by almost 1%. A relatively elastic coefficient associated with PROEX implies that an improvement in GDP per capita is more important for processed food products than traditional agricultural exports resulting in structural changes in agricultural exports. The positive relationship between processed food exports and real GDP per capita is also found in Jaffee and Gordon (1993) and Athukorala and Sen (1998). In particular, the latter applies to a pooled cross-section procedure, with the first difference, to estimate the determinants of processed food exports for 36 developing countries during 1970-1994. The elasticity associated with the growth of per capita income is 1.2%. The slight differences in the estimation results would come from both different time periods and country coverage and differences in econometric procedures.

VI. Results
The lagged value of real GDP growth per capita is applied as an instrumental variable. 4 Results of other alternatives are available from the authors on request. ; OPEN1 = trade to GDP; DC = domestic credit over GDP; PIS1 = agriculture price instability; RER = real exchange rate (eP*/P); FDI = foreign Direct Investment Inflows as a percentage of GDP; ROAD = road density. Note: (i) Time effects were not included in the fixed effect model because of their statistical insignificance.
(ii) Two-stage least square is performed. The lag value of real GDP per capita is used as an instrument.
The coefficient on OPEN1 is positive and statistically significant. A 1% rise of OPEN1 would lead to an increase in the share of processed food in total agricultural exports by 1.34%, other things being equal. Providing robust statistical support, superior export performance in processed food is closely linked with the openness of the trade policy regime. The relatively high value of the coefficient associated with OPEN1 points out that the nature of trade policy regime is crucial in explaining intercountry differentials in export performance of processed food. Interestingly, the interaction term of OPEN1 with real GDP per capita (CS) becomes statistically significant. This implies that an increase in the degree of trade policy openness would result in less importance of domestic size in explaining the share of processed food in total agricultural exports. Compared to firms in the traditional agriculture sector, the size of the domestic market becomes less important in helping firms in the processed food sector to achieve economies of scale and reduce costs to break into foreign markets. A 1% increase in OPEN1 would lead to a decline in importance of CS by 0.18 percentage point so that the coefficient associated with CS would decline to 0.81.
In terms of trade facilitation, the estimated coefficient associated with road density (ROAD) is statistically significant in determining changes in the structure of agriculture exports (PROEX1). The statistical significance in this equation implies that an improvement in infrastructure, especially transportation, would benefit the processed food industry more than traditional agricultural products since the former tends to involve more numerous members of food supply chains 15 than the latter. The coefficient corresponding to domestic credit (DC) also becomes statistically significant, reflecting the ability to access credits would help to expand processed food production and exports. A 1% expansion in domestic credit leads to 0.06% increase in the share of processed food exports in total agricultural exports.
Favorable macroeconomic conditions, in terms of price stability, are also important in expanding processed food exports. Price instability could eventually lead to a loss in cost competitiveness since risks and uncertainties related to costs of production could increase substantially, thereby reducing investment incentives in the processed food industry.
Interestingly, a statistical insignificance of the coefficient associated with FDI is revealed. This implies that an involvement of MNEs in processed food industries seems unlikely to stimulate processed food exports. The statistical insignificance of FDI in the processed food industry may result from the fact that there are some MNEs that tend to be involved only in the domestic market. This is particularly true for manufacturing food production, such as confectionery and ready-to-eat products. Because technology and capital used to produce these products are mobile in the world food market, and raw materials of these products are relatively nonperishable and inexpensive to transport, MNEs tend to locate their production close to consumers, and sell products mainly in domestic market. Thus, the positive effect from FDI to processed food exports, especially in terms of technology and export spillovers, could be distorted by the nature of processed food industry. 16 We found statistical insignificance in the regional dummy coefficients, which could be used to reflect geographical proximity. Since processed food still contains some characteristics of traditional agricultural products, particularly perishability, geographical proximity becomes statistically insignificant in changing the structure of agricultural exports. However, Jongwanich (2009) found that geographical proximity seems to be one of the important factors in determining the ability to expand volume of processed food exports. In particular, it was found that coefficients associated with Asia, Europe, and Latin America are statistically significant. The significance of these three regional dummy variables could reflect the cost advantages emerging from their shortened distance from the key export destinations of processed food products, which are Europe, Japan, and US. In particular, the coefficient associated with Europe is the highest among these three regional dummy variables. This could be because developed Europe is the largest export destination for processed food products so that developing countries in Europe could get more cost advantages than other developing countries in expanding their processed food exports.

VII. Implications for Developing Asia
As discussed in Section III, Asia has performed better than other regions in exporting processed food products over the past three decades, followed by Latin America and Europe. In Asia, processed food exports were concentrated in some countries of East and Southeast Asia, i.e., PRC, Malaysia, Indonesia, and Thailand. Over the past decade, the structural change of processed food has occurred in Asia. Southeast Asia has become less important and been replaced by countries in East Asia. Although the share of processed food exports in South Asia has been relatively stable, it was significantly lower than that in the early 1980s. This section reviews factors that support and constrain an expansion of processed food exports in Asia by combining the revealed coefficients in the previous section with recent developments in the independent variables in Asia as well as in other regions. Table 7 shows movements of the relevant independent variables in developing Asia and other regions during 1990-2005 while Figure 9 reveals the movements of these variables in each country of developing Asia during 1995-2005. It is found that size of domestic market, trade openness, and price stability tend to support an expansion of processed food exports in Asia, East Asia in particular. However, inadequate infrastructure and inability to overcome food safety standards are constraining an expansion of processed food exports, particularly in South Asian countries.
The level of income per capita in Asian countries, particularly in the PRC and India, increased noticeably during 2001-2005. On average, income per capita in Asia grew by 5.8%, compared to 4.9% and 1.3% in Europe and Latin America, respectively, during this period. Rising incomes have been associated with growing health consciousness and urbanization that contribute to changing dietary patterns of consumers. Although there are differences in the pace and specific features of these dietary changes, there has been a common shift toward increased consumption of processed food products, which have relatively high unit values and high income elasticity of demand, compared to staple food products. The expansion of domestic demand for processed food would help firms to reduce costs and achieve economies of scale to break into world markets.
Trade policy movements toward more liberalization in Asia also helps boost exports of processed food. Even though an index of trade to GDP could roughly measure the trade policy regime in a country, this index might, to some extent, reflect a possibility of exporters to break into the world market. In Asia, the ratio of trade to GDP has increased continuously over the past two decades. The ratio increased to 60% in 2001-2005, from 44% in 1991-1995. The index was higher in Latin America, the second largest exporter of processed food. Exporters of processed food in Southeast Asian countries, Malaysia and Thailand in particular, are likely to benefit most from more trade liberalization. The ratio of trade to GDP, by contrast, was lowest in South Asia. This is consistent with the implied tax rate on international trade and transactions, which shows the highest level in South Asia (7%), compared to an average of 2% in Asia, 3% in Latin America, and 5% in Europe. Box 1 also shows that the average applied tariff rate on processed food in India tends to be higher than that in Southeast and East Asian countries. In particular, higher tariff on unprocessed food in India may distort incentives to produce and export processed food products.
Even though trade liberalization tended to be more pronounced in Southeast Asia, tariff escalation still limits expansion of processed food exports. 17 As mentioned in Box 1, distortions in the agricultural sector, particularly processed food, are still higher than in the manufacturing sector although many rounds of multilateral trade liberalization resulted in a decline in tariff and nontariff barriers. In particular, tariff escalation in the processed food industry provides incentives for producers to produce and sell in the domestic market. Rae and Josling (2003) found positive and high values of the effective rate of protection in the processed food industry in Asia, the Association of Southeast Asian Nations, and the PRC (0.24), compared to 0.014 in the US, and 0.04 in Australia, which are relatively close to their nominal rate of protection. They also found that comprehensive trade policy reforms in developing countries could contribute to growth in processed food exports, i.e., allowing developing countries with a comparative advantage in food production to expand exports and take advantage of increased market access to other developed and developing countries.
Price stability in terms of agricultural prices as well as general producer and consumer prices also help facilitate processed food exports. In particular, the fluctuation of agricultural prices in Asia over the past decade was far lower than that in Latin America and Europe. The low level of agricultural price fluctuations helps to reduce uncertainties of firms relating to costs of production. The relative stability of agricultural prices in Asia may be attributed to the development of contract coordination, which represents an intermediate institutional arrangement between spot market trading and the vertical integration of production and marketing functions (Jaffee and Gordon 1993). 18 Based on historical data, the relatively poor development of infrastructure tends to constrain the expansion of processed food exports in Asia. 19 Road density in Latin America and Europe was more than 6 kilometers/1,000 population during this period; it was only 2 kilometers/1,000 population in Asia. This could have implications for resource allocation to expand processed food production and exports. Land development, improved irrigation system, upgrading production technology, and improved physical infrastructure would result in agricultural and infrastructure development; would help firms to control quality of raw materials; and reduce risks of facing food safety standards imposed by developed countries.

VIII. Conclusion and Policy Inferences
This paper examines the determinants of structural changes in food exports from developing countries using panel data econometric analysis. Over the past three decades, processed food exports have shown greater dynamism compared to primary exports. In the early 2000s, processed food exports accounted for more than 60% of world food exports and around half of world agricultural exports. The dynamism of processed food exports has been contributed mainly by developing countries while the contribution from developed countries has been relatively stable. Rising incomes, growing health consciousness, urbanization, international migration, communication revolution, and international tourism have contributed to diet upgrades. Declines in tariff and nontariff barriers, both in developed and developing countries, have also facilitated the expansion of processed food trade. However, so far, not all developing countries have been able to reap the benefits from increased demand for processed food products.
The empirical model in this study suggests that trade policy openness and large domestic market size are the key factors that influence export success in processed food products. Good macroeconomic management, especially in terms of price stability, as well as adequate financial support and infrastructure could also facilitate an expansion of processed food exports thereby shifting agricultural exports toward processed food exports.
Compared to other regions, a significant expansion of domestic markets, trade policy openness and price stability help to increase the share of processed food in total agricultural exports in developing Asia. These factors were particularly important for East Asia, the PRC in particular, in expanding processed food exports in the world market. However, inadequate infrastructure tends to be a key constraint for an expansion of processed food exports in developing Asia.
9 Note that transport services for exports, which is another indicator of infrastructure, also points out that infrastructure development in Asia is still lower than other developing countries. Transport services as a percentage of GDP in Asia was around 0.7, close to that in Latin America, but far lower than developing countries in Europe (2.% of GDP). The lack of transport services for trade tended to become more serious in East and South Asia than other subregions.
To further develop the processed food industry in developing Asia, both quality and quantity of infrastructure need to be improved. This includes well-developed roads, railways, ports, telephone lines, power systems, terminal markets, storage, and processing facilities. Given the high level of domestic resource content of production, improvement in the agricultural sector would help to stimulate processed food exports in developing Asia. The improvement in the agricultural sector is related to upgrading land quality and irrigation systems as well as the ability to adequately access inputs such as fertilizers. In particular, upgrading production technology is an essential path to improve quality and productivity in the agricultural sector. Improvements in certain technologies would lead to more extended multiple cropping, improved taste and hygiene, and uniform output. Timing (seasonality) of production could be better controlled, thereby reducing risks and enabling producers to diversify their crop/livestock mix. To improve the agricultural sector and processed food industries, the government of developing countries must work to reduce distortions in credit market and to ensure that farmers and firms in processed food industries can adequately and equally access credits.
In addition to providing adequate financial resources, supporting vertical integration, either complete or partial, become important and relevant in the context of processed food industries. Logistic costs associated with the procurement of raw materials and/or the sale of finished products could be reduced. In particular, transport costs can be reduced, especially for bulky and perishable raw materials, as vertical integration involves bringing together in one location formerly distinct operating units. The level of required inventories can be reduced because internal planning allows for a better match of supply and demand in terms of quantity and location. Problems of risks and uncertainties could also be redressed. In particular, variability of supplies and output could be eliminated as more direct control over raw materials can be exercised under completed or even partial vertical integration. The reduction in risks and uncertainties could allow firms to better invest in highly specialized processing and marketing facilities, and to take advantage of potential economies of scale.
Even though tariffs in developing Asia in the agricultural sector have progressively declined, the higher tariff reductions on bulk commodities compared to processed food products still results in tariff escalation. The tariff escalation structure distorts economic incentives to export and provides the perception that a country is still protecting their processing industries, compared to unprocessed (primary) commodities. Trade policy reform that aims to reduce divergence of protection level between stages of production, e.g., a higher uniform cut in tariffs in processed food than unprocessed (primary) commodities would help increase efficiency in the allocation of resources and expanding processed food trade. In some developing Asian regions, South Asia in particular, a significant reduction in tariffs on primary (unprocessed) products would be needed to ensure that the tariff structure does not lead to negative effective protection in the processed food industries.