Essays on Product Quality, Trade Costs, and Trade Liberalization DISSERTATION

Transcription

1 Essays on Product Quality, Trade Costs, and Trade Liberalization DISSERTATION Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy in the Graduate School of The Ohio State University By Jihyun Eum Graduate Program in Agricultural, Environmental and Development Economics The Ohio State University 2017 Dissertation Committee: Stanley Thompson, Advisor Ian Sheldon, Co-Advisor Abdoul Sam

2 Copyrighted by Jihyun Eum 2017

3 Abstract This dissertation explores the effects of trade policies on product quality and trade flow. It is composed of three chapters: the implication of food safety standards on traded product quality improvement; the evaluation of the trade determinants using the endogenous quality choice model in food and agricultural trade; and the analysis of asymmetric trade costs and relative productivity between the developed and developing countries using Ricardian trade models. The first chapter analyzes the effect of food safety standards as well as tariffs on efforts to upgrade food product quality with quality. Compliance costs are introduced into a model of competition and innovation where the rate of food product quality upgrading is affected by both changes in tariffs and standards. Using disaggregated data for European food imports from 159 trading partners over the period 1995 to 2003 across 28 industries, the effect of standards enforcement on quality upgrading varies nonmonotonically. In particular, products that already have relatively higher quality are more likely to upgrade but those having relatively lower quality are less likely to upgrade. The second chapter analyzes the determinants of bilateral food and agricultural trade using the heterogeneous-firm trade model with endogenous quality choice. Identifying the effect of trade liberalization with the consideration of product quality and firm productivity is important because productivity plays a role in determining export ii

4 participation. In this chapter, an endogenous quality model provides the framework to estimate the effects of the major determinants of bilateral trade. After identifying the relationship between firm productivity and quality choice the consequences of changes in trade costs are shown. We use the Helpman, Melitz, and Rubinstein approach to show the effects of trade frictions within the context of heterogeneous productivity and product quality. The estimation results with controlling the number of exporting firms are consistent with theoretical expectations. The findings suggest that OECD members who import high-quality products restrict the number of exporting firms by applying relatively higher trade threshold than non-oecd member countries who allow export companies to enter their market easily. The third chapter examines the reason why developing countries trade fewer agricultural goods than developed countries is analyzed. Started from the argument that low trade flow in the agricultural sector is mainly due to high trade costs (Xu, 2015; Reimer and Li, 2010), our analysis investigates how bilateral trade costs in agricultural sector differ among trade partners. Asymmetric trade costs and the variability of technology in the agricultural sector are derived using Ricardian trade models. Systematically asymmetric bilateral trade costs and the variations in productivity level of all observed countries are revealed as main trade barriers for developing countries. Lowincome countries face higher trade costs to export than high-income countries. iii

5 Dedication To my beloved grandfather, Myung-bok Eum, who passed away on December iv

6 Acknowledgments Had it not been for the help of my advisors and committee, I would not have been able to finish my dissertation. I am grateful to my advisor, Stanley Thompson, for supporting my research and for providing encouragement. He has provided invaluable advice regarding academic research and encouraged me when I went through the sorrow while preparing job interviews right after my grandfather passed away. Without his endless support and advice, I would not have got over the loss of a loved one. I am also deeply indebted to my co-advisor Ian Sheldon, who has helped me with in-depth knowledge of trade economics. The opportunity he provided to join his research project has helped me to learn substantial literature as well as how to be an independent economic researcher. I am very much aware how lucky I am to have such great advisors. I would like to express my gratitude to Abdoul Sam and Brian Roe who have helped me to extend my research with great econometrics skills and the in-depth knowledge of agricultural economics. I must also thank all my classmates, faculty members, and staffs in our department. Finally, I would like to thank my parents, sister, brother, brother-in-law, and sister-in-law for endless support, encouragement, and love. Postcards and mail from my precious nieces, Seoyun and Seunga, were the source of the strength when I doubted myself over the past four years. v

7 Vita Hyundai High School B.S. Business, Ewha Womans University M.A. International Studies, Seoul National University M.A. Applied Economics, Ohio University 2013 to present...graduate Teaching Associate, Department of Agricultural, Environmental, and Development Economics, The Ohio State University Fields of Study Major Field: Agricultural, Environmental and Development Economics vi

8 Table of Contents Abstract... ii Dedication... iv Acknowledgments... v Fields of Study... vi Table of Contents... vii List of Tables... viii List of Figures... x Chapter 1: Food Safety Standards and Quality Upgrading through Competition... 1 Chapter 2: Trade liberalization and Endogenous Quality Choice in Food and Agricultural Trade Chapter 3: Measuring Asymmetric Trade Costs in Agricultural Trade References vii

9 List of Tables Table 1.1: Competition, compliance costs and innovation Table 1.2: Descriptive statistics for EU food trade data ( ) Table 1.3: Observed food industries Table 1.4: Food product quality estimation results ( ) Table 1.5: Correlation and OLS results with fraction of exporters products with highest quality Table 1.6: Quality upgrading and competition Table 1.7: Tariffs and standards effect on quality improvement: all observations Table 1.8: Tariff and standards effects on quality improvement: OECD and Table 1.9: Robustness checks - weighted unit value as dependent variable Table Robustness checks - quality changes in low and high HHI market Table 2.1 : List of observed countries and industries Table 2.2: Summary statistics Table 2.3: Top 10 countries with good governance for year Table 2.4: Bottom 10 countries with good governance for year viii

10 Table 2.5: Number of NTM by products for year Table 2.6: Top ten countries with the highest and without NTM for year Table 2.7: Benchmark Gravity and Baseline Results Table 2.8: Gravity estimation: baseline, heterogeneous firm trade model, bias decomposition Table 2.9: North and south trade Table 3.1: Observed countries Table 3.2: Summary statistics Table 3.3: Observed product Table 3.4: Estimation of Si Table 3.5: Estimation of the Effects on Trade Costs Table 3.6: Estimation of Productivity Table 3.7: Asymmetric Trade Costs for selected countries ix

11 List of Figures Figure 1.1: The number of EU standards Figure 1.2: The relation between GDP per capita and fraction of high quality products. 30 Figure 3.1: Destination country effects (S i ) and GDP per capita Figure 3.2: Effects on trade costs and GDP per capita Figure 3.3: Productivity (T i ) and GDP per capita Figure 3.4: Asymmetric Trade Costs Figure 3.5: Asymmetric Trade Costs and GDP per capita x

12 Chapter 1: Food Safety Standards and Quality Upgrading through Competition 1. Introduction Food product quality matters in international trade, especially as regards human, animal and plant health. Food safety concerns have led many countries to adopt non-tariff measures (NTMs) such as standards designed to improve the quality of traded food products. In 1995 the World Trade Organization (WTO) adopted the Agreement on the Application of Sanitary and Phytosanitary Measures (SPS Agreement) which provides the basic trade rules for food safety and animal and plant health standards. It grants member countries the right to set their own standards if based on scientific justification. Although the aim of the SPS Agreement is to share common regulations across WTO member countries in order to improve overall economic welfare, the measures may reduce trade flows, thereby generating negative economic outcomes. 1 The European Union (EU) adopted food safety standards in line with the SPS Agreement, and over time, their enforcement has become stricter. In addition, the number of standards applied to food products by the EU grew from around 150 in 1996 to almost 800 in Food safety standards generally refer to voluntary standards, 2 yet these provide a legal basis for developing or initiating mandatory standards or regulations. All member states and any exporters who want to access its market should conform to the essential requirements of standards enforced by the European Committee for 1

13 Standardization (CEN). 3 Exporters also have an incentive to adopt the standards given that consumers are well aware of the risks associated with food safety (Segerson, 1999). Therefore, standards affect exporter behavior even though adoption may not actually be mandatory. Except for food safety standards, other trade policies such as tariff have been eliminated significantly in recent years. Past literature observes that trade liberalization leads to improvements in firm productivity. Given the positive effect of trade liberalization on productivity, it is natural to link the effect of trade barriers reduction on firm decisions regarding product quality. Product quality is a result of innovations that reduce costs or raise the total factor productivity. Sutton (2008), Khandelwal (2010), and Verhoogen (2008) deliver a mapping between firm productivity (exogenous) and product quality (endogenous). In a similar manner, heterogeneous productivity of firms is incorporated into the vertically differentiated products in the paper of Baldwin and Harrigan (2011) and Johnson(2012). They explain that quality-adjusted price depends on firm productivity since high-quality products are relatively expensive, serve distant markets and have high markups. Accordingly, this essay assumes firm productivity is equivalent to product quality. The objective of this essay is to examine the effect of the EU trade policies on exporter decisions about the quality of the products exported to the EU. More specifically, the effects of tariffs and NTMs on food product quality improvement are estimated using the distance-to-the frontier model originally due to Aghion et al. (2004). This model predicts that increased market competition influences firm innovation and the 2

14 relationship between competition and innovation depends on the closeness of products to the world technology frontier. If firms products are initially close to the technology frontier, the threat of entry encourages firms to innovate as a means of avoiding competition. On the other hand, firms with products far from the frontier anticipate losing market share through competition with new firms and reduce innovation. In this essay, tariffs and NTMs are treated as policy instruments that influence competition where product quality upgrading is a consequence of innovation activities. 4 A few studies have found a direct link between competition and product quality. Using US trade data, Amiti and Khandelwal (2012) examine the effect of tariff reduction on product quality improvement. The results show that reduced tariffs are associated with quality upgrading for products which are equivalent to the frontier, but not so for products far from the frontier in terms of product quality level. The EU food industry has been examined by Curzi et al. (2015). They investigate the impact of changes in tariffs as well as NTMs on food product quality. Their results confirm that the quality of food products close to the frontier is more likely to be upgraded with tariff reduction that results in increasing import competition. Also, according to their result, the introduction of voluntary standards has a positive effect on quality upgrading. While Curzi et al. firstly investigate food product quality in the EU food industry, their results regarding product standards do not follow theoretical expectations. A priori, if some firms do not enter the market because they are unable to satisfy higher standards, quality improvement for products close to the frontier will be negatively affected while those more distant to the frontier will be positively impacted. 3

15 According to the estimation results, increased competition due to tariff reductions in the import market drives leading food manufacturers to improve product quality relative to laggards as predicted by the distance-to-the-frontier model. On the other hand, the effect of standards on quality improvement varies non-monotonically depending on the relative quality level, indicating that products far from the frontier are less likely to undergo quality upgrades, and vice versa for products close to the frontier. In detail, the cost of compliance is the most influential factor on firm s decision to improve product quality for laggard products. Meanwhile, Besides, for leading products, the probability of capturing market share plays an important role in upgrading product quality. The current study is the first to address the augmented competition and innovation model (Aghion et al., 2004) in the context of trade and standards. First, compliance costs are introduced and the conditions under which trade standards are enforced are considered. Second, a way is suggested for improving the empirical estimation. Unlike Curzi et al. (2015), the sample here excludes intra-eu trade data. This exclusion improves the estimation since all EU members share the same external trade policy. Finally, weighted trade standards are used in the analysis. Not all standards have the same importance for trading partners because trading partners do not export all products with standard regulations at the even value. For example, in Nigeria cocoa beans are the highest value food export. Therefore, cocoa bean standards are likely to be of greater importance than other food safety standards. Therefore, standards are weighted according to the products export value as a share of industry sales. Thus, standards data vary substantially across exporters, products, and observation periods. 4

16 The remainder of the essay is structured as follows: in section 2, EU standards in the food and agricultural sector are described, and the distance-to-frontier model that forms the basis of the empirical analysis is derived in section 3. The empirical specification and estimation methodology are outlined in section 4, while the data and estimation results are presented in sections 5 and 6, respectively. Finally in the section 7, the essay is summarized and some conclusions are drawn. 2. EU Standards in the Food and Agricultural Sector A standard is a technical specification indicating requirements for products, production processes, or test-methods. The focus in this essay is on the EU which maintains strict laws and regulations regarding food safety, and where it is expected that standards will have an impact on food product quality improvement. There are three European Standardization Organizations (ESOs), that have ratified European Standards (ENs), CEN, CENELEC, and ETSI. Although the three ESOs are different in terms of their fields of specialization, they cooperate in areas of common interest. EU members have the obligation to implement standards at the national level and so ENs automatically become a national standard. Even though ENs are voluntary technical standards according to EU Regulation 1025/2012, specific EU laws and regulations refer to standards, thereby making compliance with standards compulsory. For example, genetically modified (GM) food labeling requirements laid out in regulation (EC) No 1829/2003 and 1830/2003 are mandatory. Producers have to provide information relating to products that are either GM or contain genetically modified organisms (GMOs). In 5

17 terms of implementation, detailed technical methods of analysis for the detection of GMOs are guided by ENs, e.g., EN ISO 21569, CEN/TS Another example is EU regulation (EC) No 2160/2003, which lays down specific mandatory requirements for non-eu members to have a salmonella control program for food products. In Article 12 of the regulation, participating laboratories in control programs are required to apply a quality assurance system that conforms to ENs. Thus, exporters along with EU member countries must comply with these voluntary standards to enter and remain in the EU market. Panel (A) of Figure 1.1 shows the numbers of EU food safety standards established from 1995 to The numbers of standards have increased significantly over the period, and the numbers of EU standards not conforming to ISO standards are much larger than those that do. In Panel (B) of Figure 1.1, the numbers of total pages of EU standards are shown, implying the strictness of standards. The rigidity of standards has increased along with the number of standards. 3. Theoretical Background Following Aghion et al. (2004; 2005), firms produce each intermediate input in sector v under Bertrand competition. Final good Y t is assumed to be produced in each time period t using a continuum of intermediate inputs: 1 1 t ( t ( ) t ( ) ), 0 Y A v x v dv 0 1 (1) 6

18 where A t (v) is a productivity index indicating the intermediate input quality in sector v at time t, and x t (v) is the intermediate input quantity used in sector v at time t. The parameter v represents both an intermediate sector and an intermediate firm because only one firm at time t exclusively produces and sells each intermediate good. The intermediate-goods firms live for only one period and the follower takes the property rights. Equilibrium profit for each intermediate firm is proportional to the productivity parameter (Acemoglu et al., 2006): 1 1 where indicates 1 2 ( v) A ( v), (2) t (1/1 ). t If a firm v successfully innovates, its technology parameter At () v increases. For example, let the technology level of a frontier firm in the previous time period t 1, be denoted as At. If the frontier firm successfully innovates, then their technology 1 parameter At grows at the exogenous 1 rate in the next time period t: A () v A 1 (3) t t 1 One of three types is assigned to each intermediate firm at the beginning of the period. Type-1 firms run as the frontier at the beginning of the time t, with a productivity level A () t 1 At 1 v. Type-2 firms are located at one step behind the frontier, with A ( v) A ( v) t 1 t 2, and type-3 firms have two steps behind technology denoted as At 1( v) At 3( v). Incumbent firms improve their productivity with the constant rate by 7

19 innovation. However, it is assumed that type-3 firms have no need to invest in innovation because they can innovate automatically as a result of knowledge spillovers (Aghion et al., 2004) Let z denote the probability that firms successfully innovate. This probability also depends on firms willingness to launch and complete innovation, indicating innovation intensity. Accordingly, z becomes a decision variable that represents how much firms invest in innovation based on expected profit and innovation cost. The productivity of incumbent firm improves with probability z and lags behind the new frontier by j-1 steps. A type-j, where j equals one or two, firm at time t must bear innovation cost c i as (Aghion et al., 2004) 2 z j ci ( z j ) ci At j ( v). 2 (4) In addition to innovation cost c i, firms must consider compliance costs of meeting additional standards. Let cc denote compliance costs, i.e., the expenditure required to conform to a product standard. Compliance costs are assumed to be imposed only on a type-2 intermediate firm along with innovation cost, because the technology level of a type-1 firm is assumed greater than the minimum required by the standard. If the required level of technology is At, a type-2 firm incurs compliance costs as well as the 1 innovation cost when they wish to enter the market, but a type-1 firm incurs only innovation cost. Compliance costs have the same functional form as innovation cost 8

20 denoted by 2 z j cc ( z j ) cc At j ( v) 2 for j=2. Therefore, the total cost of a type-j intermediate firm, j 1,2 is: 2 z j c( z j ) ( ci ( j 1) cc ) At j ( v) 2 (5) The current study focuses on competition among firms exporting to the EU market. The existing exporting firms are defined as those currently exporting in the EU market while the potential entrants are defined as other foreign firms who make an effort to enter the EU market and compete with current exporting firms. In each time period there is the threat of entry from outside firms that run with productivity A t, which is equivalent to the frontier productivity at the end of period. Under Bertrand competition an entrant firm seizes the entire market and becomes the incumbent firm if it is more productive. Otherwise the profits of both firms become zero if the entrant has identical productivity. Now, assume that potential entrants are able to observe post-innovation technology. A potential entrant will not pay the entry cost if it cannot operate on the frontier postinnovation because the profits dropped to zero under the Bertrand competition. The incumbent laggard firms no longer invest to innovate because their expected profit is zero at best. Equilibrium Innovation without Compliance Cost Let p define the probability that a new firm enters the market. A type-1 leader retains the market under only two scenarios: either it successfully innovates with probability of z 1, or no firm enters even if it fails to innovate with probability of (1 z1)(1 p). For a type-2 9

21 firm, there are also two scenarios: they successfully innovate with probability of z (1 p) 2 or they fail to innovate under the condition that no firm enters with probability of (1 z2)(1 p). A firm that is initially close to the frontier chooses its investment z 1, and a type-2 incumbent chooses innovation investment z2 to maximize the expected net payoff from innovation as: max [ (1 )(1 ) ] ( / 2) (6a) 2 z z 1 1At z1 p At 1 z1 ci At 1 max [ (1 ) (1 )(1 ) ] ( / 2), (6b) 2 z z 2 2 p At 1 z2 p At 2 z2 ci At 2 with the first-order conditions of (6a) and (6b) yielding: z1 ( 1 p) (7a) c i z2 (1 p)( 1). (7b) c i Equilibrium Innovation with Compliance Costs Firms incur additional costs when their technology does not meet the standards. As assumed above, a type-1 firm does not bear expenses to meet standards, due to their technology satisfying the minimum required standard. In contrast, a type-2 firm has to bear compliance costs under established standards. A type-2 incumbent chooses its innovation intensity z 2 to maximize the expected net payoff from innovation: max [ (1 ) (1 )(1 ) ] ( / 2)( ), (8) 2 z z 2 2 p At 1 z2 p At 2 z2 ci cc At 2 with the first-order condition of (8) yielding: 10

22 z 2 (1 p)( 1). ( c c ) i c (9) Effects of the Competition and Compliance Cost The effect entry on innovation of an increased threat of entry is shown by partial differentiation of (7a) and (7b) with respect to the probability p : z1 / ci 0, p z2 ( 1) / ci 0. p (10a) (10b) A higher p boosts innovation for a type-1 firm. As the likelihood a firm will lose out to entrant increases, the incentive for a frontier firm to escape from the competition increases. On the other hand, higher p lowers the expected payoff from innovation to the type-2 firm, so their effort to innovation is reduced. A firm distant to the frontier expects that it must be left out further if it invests to innovate. The innovation decision of a type-2 firm depends both on the entry threat and compliance costs of meeting standards. The entry threat declines as higher standards are enforced. It is harder for new firms to enter the market when their level of technology fails to meet required standards: z2 ( 1) 0, p c c i c (11a) z2 ( 1) 2 c ( c c ) c i c 0. (11b) 11

23 The impact of tougher standards on innovation activity is ambiguous for a type-2 firm in that entry threats and compliance costs have conflicting effects. Owing to standards enforcement, a type-2 firm is more likely to innovate due to less competition but they are less likely to innovate because of the costs of compliance, (11a) and (11b). If the effect of compliance costs is larger than the effect of competition, a laggard firm will decrease innovation activity. On the other hand, for a type-1 firm the effect of standards enforcement on innovation is also ambiguous. Initially a reduced threat of entry lessens innovation as in (10a). Simultaneously, this reduced level of competition raise the expected profits for the next time period. Existing firms in the market would expect higher profit once they know that new entrants hardly enter the market at the current time. For simplicity, we consider the additional raise in expected profit as the exogenous growth rate of technology, γ 5. The effect of the growth in expected profit can be shown as the partial differentiation of the optimal level of z 1 with respect to γ, which is positive. The theoretical predictions are summarized in Table 1.1. The model of Aghion et al. (2004; 2005) shows that any trade policies that increase competition by reducing entry cost will discourage the laggards from innovation but encourage leading firms to increase innovation expenditures. However, the enforcement of a standard does not always encourage innovation because on the one hand it lessens competition but on the other it increases compliance costs. Therefore, the net effect of product standards establishment on innovation activity is ambiguous. Under the pressure of incurring compliance costs lagging firms may resist innovation activities whereas the high expected profit for firms that satisfy the standards boost innovation activities of leading firms. 12

24 4. Empirical Specification Quality Estimation Typically in the literature, product quality has been estimated through either import or export unit values (Schott, 2004; Hallak, 2006). While this approach is relatively easy to implement, it is problematic in that import or export prices may differ for reasons other than quality, such as exchange rates or labor cost differences. In the current essay, product quality is measured through considering market share information along with unit values (Khandelwal, 2010). Using a nested logit system, this methodology takes into account consumer preference structures as well as the horizontal component of different product varieties, e.g., vanilla-flavored and strawberry-flavored yogurt. The horizontal component is incorporated in the demand estimation to account for horizontally differentiated products having higher market shares. As a result, quality is treated as the unobserved vertically differentiated attributes of products at a given unit price and market share. To measure product quality, the nested logit demand model of Berry (1994) is used. A variety ch is defined as an imported product h from exporting country c. Equation (12) is the reduced form of the demand equation for an imported variety ch at time t with industry subscripts suppressed: ln( S ) ln( S ) P ln( ns ) ( pop ) (12) cht ot 1, ch 2, t 3, cht 1 cht 2 cht 3 ct S is the overall EU s market share of variety ch at time t, defined as S q / MKT, cht cht cht t where q cht is the EU s imported quantity of a variety h from exporting country c at time t 13

25 and MKT t qcht is industry size of the EU. The outside variety S ch o ot represents (1 S ) ot the intra-eu alternative to the imported variety and is defined as S (1 IMPPN ), where IMPPN t = (import quantity) / (import quantity + production quantity - export quantity). The left hand side of (12) expresses consumers indirect utility from choosing the imported variety ch over the intra-eu produced variety. Indirect utility is a function of: a variety s unit value P cht, nested shares ns cht defined as variety ch 's share within product h, at time t, and population pop ct. The unexplained part of indirect utility cht is treated as the measure of product quality, where:, cht 1, ch 2, t 3, cht ot t where 1,ch is the time-invariant valuation of variety ch, is the time-variant common 2,t quality component, and 3,cht is the variety-time deviation from the fixed effect unobservable to consumers. The population of partner countries popct is included in (12) to control for unobserved varieties (Feenstra, 1994; Hallak et al., 2011). For example instance, if China exported a wide variety of green tea products which are not observed in the Harmonized System, observable aggregate data would overestimate the quality of Chinese green tea. The assumption here is that the number of varieties produced increases with a country s population (Krugman, 1980). Two-stage least square (2SLS) is used to estimate (12), due to there being two endogenous variables, price P cht and nest share ns cht. In the case of the endogenous price, 14

26 the identification strategy is to use transportation cost, being correlated with price but not with quality. Due to the limited data on transportation costs, the interactions between oil prices and average distances from partner countries to the EU-15 are used as a proxy. The exchange rate between trading partners is also used as instrument variables given that import prices are influenced by exchange rate changes but quality is not (Amiti and Khandelwal, 2012; Curzi et al. 2015). In the case of nest share ns cht, the identification strategy is to use the number of exporting countries within product h and the number of varieties exported within exporter c. The numbers indicate entry and exit of varieties in the market. The entry and exit decision can be used as instrument variables because entry and exit of other varieties is positively and negatively related with a variety's nested share but not correlated with a variety's quality (Amiti and Khandelwal, 2012; Curzi et al., 2015). Quality Upgrading and Import Competition The non-monotonic relationship between competition and innovation outlined earlier is estimated by the interaction of distance-to-the-frontier DF with tariffs and standards in equation (13) (Curzi et al., 2015): ln(exp( )) DF X ( DF X ), (13a) cht ch, t 3 ch, t 3 ch, t 3 ch, t 3 ht ct cht where: DF exp( ) / max (exp( )), DF [0,1]. (13b) cht cht c ht cht cht In (13a), the dependent variable ln(exp( cht )) is the change in a variety s quality over a period of three years. Distance-to-the-frontier DF is measured according to (13b), where for varieties close to the frontier DF 1, and for varieties far from the frontier 15

27 DF 0. X is a vector of tariff and standards variables lagged by three years, while ch, t 3 ht and ct are product-year and country-year fixed effects respectively. Given quality is estimated by industry, the quality of products should be compared within an industry using product-year fixed effects. Product-year fixed effects also control for systemic shocks that affect all varieties of a specific product at a point in time such as demand shocks. Country-year fixed effects control for country-level shocks such as changes in factor endowments or productivity or national-level technology shocks. The coefficient on tariffs is expected to be positive whereas the coefficient on DFtariff interaction expected to be negative. A reduction in tariffs will increase a variety s quality in the subsequent three years only if products are close to the world quality frontier, DFch, t 3 1, and vice-versa for products far from the frontier DFch, t 3 0, i.e., the escape from the competition and discouragement effects respectively (Aghion et al., 2004; 2005). For standards, two conflicting effects are expected. First, standards enforcement of the EU may limit competition among exporters, positively (negatively) affecting quality upgrading for laggard (leader) firms. Second, standards enforcement may directly affect product quality, exporters toward EU market having to bear the burden of compliance costs. Exporting firms may be reluctant to innovate owing to the high R&D costs associated with product upgrading. Given that the level of compliance cost depends on distance-from-the-frontier, these costs are greater for laggards than leaders. The mostbackward firms innovate only if the compliance costs necessary to satisfy the standards requirement are close to zero. Additional external force to consider is future expected 16

28 profit. For leaders, high profit is expected because lower competition level may facilitates the capturing of market share. Therefore, standards enforcement may have a positive (negative) effect on quality upgrading for leaders (laggards). 5. Data Food product trade data over the period 1995 to 2003, are constructed for the EU-15 as follows: using the concordance table from Eurostat s Reference and Management of Nomenclatures (RAMON) database to identify the industry to which products belong, trade data based on the eight-digit Combined Nomenclature (CN) code from EUROSTAT-Comext are linked to the four-digit NACE industry classification. Domestic production data are taken from Eurostat s Production Communautaire (Prodcom) database using the eight-digit PRC code. Given that the first four digits of the PRC code match the four-digit NACE industry, it is straightforward to link them together. For the second stage of estimation, we use ad valorem EU tariffs applied to all exporting countries from WITS (World Bank) at the HS six-digit level from 1995 to In terms of tariffs, the sample size for estimation of (13a) is constrained according to availability of data. Data on standards are from the European Union Standard database (EUSDB) of the World Bank. These data include product standards for textiles and agriculture issued by CEN over the period (Shepherd, 2006). These count data link into the harmonized system (HS) four-digit level, which is also linked to the PRC code according to the concordance table from EU RAMON. The number of pages 17

29 of standards are used, and while the data refer to voluntary standards, they do have the potential to be adopted as technical regulations as indicated in section 2. Finally, data on exchange rates are from the International Monetary Fund s (IMF) International Financial Statistics (IFS) and oil prices are from Brent. These data are in denominated in US dollars, which are then converted to Euros based on the Euro-US dollar exchange rate from IFS, i.e., exchange rates show the exporter s local currency value per Euro. All prices are deflated using the Consumer Price Index from IFS. (See Table 1.2 for descriptive statistics and Table 1.3 B for a complete list of industries). 6. Estimation Results Quality Estimates Given that the trade data are noisy, observations above and below the 5 th and 95 th percentiles of import unit values are excluded, along with varieties that report zero import quantities. Concerns regarding the sample selection problem are addressed by using the 2SLS estimator with valid instruments. As a result, the estimators using the selected sample become consistent and asymptotically normally distributed (Wooldridge, 2010). The choice between fixed or random effects is based on the Hausman test, with the null hypothesis that random effects are preferable. For three industries in the sample, the null hypothesis is not rejected, random effects being used. In the case of the remaining industries in the sample, the null is rejected, fixed effects are used. Also, three industries in the sample are excluded because their nested shares are perfectly related to the dependent variables. 6 Due to there being few products that belong to the industry, nest 18

30 shares are considered the same as market shares. Overall, twenty-eight separate regression equations are estimated for the sample of industries. The quality estimation results are reported in Table 1.4. The sign of the estimated coefficients other than for population are as expected. The effect of population on import market share is heterogeneous across industries and the magnitudes of the estimated coefficients vary across food industries. The negative price coefficients (mean and median) indicate that an increase in price reduces net market share. For the nested share coefficients the average and median also follow expectations in that net market share increases once a variety achieves a large nested market share. These coefficients are used to predict unobserved product quality. In Table 1.5 the reliability of the quality estimates are evaluated by examining the relationship between exporter productivity and product quality. Specifically, analysis is conducted on the relationship between the proportion of products exported with highest quality and proxies for exporting country productivity: exporter GDP per capita and the structure of their respective labor force by education level. Column (1) reports a pairwise correlation between quality and the proxy productivity variables. Columns (2)-(4) show regression results with the proportion of highest quality products as the dependent variable. Due to highly educated workers having higher productivity and jobs that require more skill (Mincer, 1974; Weiss, 1995), it is expected that a more educated-labor force will have a comparative advantage in implementing new technology, which results in higher product quality (Bartel and Lichtenberg, 1987; Rosenzweig, 1995). The result is column (2) indicates that countries with higher GDP per capita export a larger 19

31 proportion of high quality, while Columns (3) and (4) indicate that exports of higher quality products increase in countries having a greater proportion of the labor force having tertiary education. The plot of quality leaders for a sample of food products is shown in Figure 1.2. All products show a positive relation between the exporter s GDP per capita and the fraction of high quality products. Norway, Japan, USA, and Australia are located above the fitted value line, indicating their export products have relatively higher quality with respect to their income level. Since these countries are highly engaged in international trade, high quality may be due to learning-by-doing. On the other hand, Middle-Eastern countries such as Oman, and Kuwait are located under the fitted value line, showing relatively lower quality goods are being exported with respect to their income level. This is due to a large portion of their GDP per capita being based on crude oil production. In sum, it can be concluded that the product quality estimates are reliable and can be used in the second stage of the estimation. Quality Upgrading and Import Competition In the second stage of estimation the effect of import tariffs and standards of EU on the quality estimates obtained in the first stage is analyzed. Due to noise in the data, changes over three years in both the estimates of product quality and import tariffs are trimmed above and below the 5 th and 95 th percentiles. The estimation results for the second step are shown in Table 1.6. As noted earlier, the interaction terms are used to show the presence or lack of a non-monotonic relationship between competition and innovation. The results for all countries are shown in columns (1)-(3). The results for the OECD and 20

32 non-oecd countries are shown in columns (4) and (5) respectively. In columns (1) and (2) respectively, the effects of tariffs and standards on product quality upgrading is shown. Column (3) reports the estimation results for the full equation (13a and 13b) including all trade policy measures. The signs of the coefficients remain the same in the different specifications, implying that the escape-competition and discouragement effects hold for all model specifications. The distance to the frontier has a negative coefficient, implying that the rate of quality improvement is faster for a variety is located further from the frontier. The tariff variable has a positive coefficient but the coefficient of the interaction term is negative. The former indicates the effect of tariff on a variety far from the technology frontier and the latter indicates the effect on a variety close to the technology frontier. Therefore, the positive tariff coefficient supports the discouragement effect, suggesting laggards are less likely to upgrade product quality when tariffs are reduced. For all observations in the sample, the results reported in Table1.7 show the following: a 10 point reduction in tariffs decreases a variety s upgrading effort at the rate of percentage point if a variety is located far from the technology world frontier (the discouragement effect). On the other hand, the interaction coefficient indicates that a firm with a variety closer to the technology frontier is eager to improve quality, supporting the idea of the escape-competition effect. Statistically, a 10 point reduction in tariffs improves quality upgrading for varieties equivalent to the technology frontier by a 2.71 percentage point (escape from competition). In the case of standards enforcement, the coefficient on the standards variable has a negative sign showing that a laggard 21

33 variety is less likely to undergo quality improvement due to establishment of additional standards. On the other hand, the interaction term has a positive coefficient indicating that a variety close to the technology frontier is more likely to be upgrade in quality. Statistically, additional one page increase in standards enforcement reduces quality growth for varieties far from the frontier by a 23 percentage point On the other hand, equivalent standards enforcement for varieties close to the frontier improves quality growth by a 19.3 percentage point. These results show that the burden of compliance cost is a much stronger factor influencing innovation activities than the effects of competition. The positive coefficient of the interaction term indicates that firms producing leading varieties exhibit a higher level of innovation activity, implying standards enforcement increases quality upgrading efforts. Exporting firms producing leading varieties may make greater efforts to improve product quality once standards enforcement raise the probability of capturing a larger market share by limiting the competition among exporters. The relationship between market share, standard enforcement, and quality upgrading is examined in the robustness tests in the next section. The effect of tariffs on quality upgrading is different between exports from non- OECD and OECD member countries. With tariff reduction, varieties exported by non- OECD member countries show a stronger non-monotonic relationship compared to varieties exported by OECD member countries. As shown in Table 1.8, varieties close to the frontier exported by non-oecd members and OECD members are increased in quality by 4.9 and 2.1 respectively. The discouragement effect on laggard firms is larger 22

34 for varieties exported by non-oecd members at 25 percent compared to 22.1 percentage point for OECD members. These results are consistent with the view that it is difficult to upgrade quality of products exported by low-income countries if varieties are located far from the technology frontier. The effect of standards on quality upgrading is smaller than the effect of tariffs, but the non-monotonic relationship is still valid. Varieties close to the frontier exported by non-oecd members indicate firms increasing efforts to upgrade quality compared to those exported by OECD members, whereas for laggard varieties exported by non-oecd members, firms are less likely to decrease efforts to upgrade quality compared to varieties exported by OECD members. This result can be interpreted as the effect of competition is greater than the effect of compliance costs for laggard varieties exported by non-oecd member countries. Firms producing laggard varieties take advantage of a reduction in competition but an increase in compliance costs has only a minor effect. Therefore, the negative effect of standards establishment on quality upgrading for laggard varieties exported by non-oecd members is less than that for varieties exported by OECD members. Robustness Checks As one robustness check, an alternative dependent variable is used to represent product quality. The model outlined earlier assumes that improved product quality is an outcome of firms innovation activities, product quality being estimated through unit value contingent on market share. Instead of using the nested logit demand model (Berry, 1994), an alternative proxy for product quality is weighted unit value, measured 23

35 according to the share of the products import value in the industry. The results from using this alternative dependent variable are shown in Table 1.9. While the estimated coefficient 1 has the incorrect sign, the signs of the other estimated coefficients remain the same. A second robustness check relates to the effect of standards on leading varieties. Earlier, the positive effect of standards on leading varieties was interpreted as being due to the higher probability of firms capturing a larger market share. Regression results for high and low Herfindal-Hirschman (HHI) EU market concentration indices also support this interpretation. Based on the definition of the market concentration index used by the World Bank, HHI measures the dispersion of imported values across the EU market, and is computed as: HHI ht n ht c 1 2 x cht 1 Xht nht 1 1 n ht, where xcht is the EU s import of product h from country c, at time t and X ht is EU s the total import of product h, at time t, and n ht is the number of exporters of product h time t. HHI represents the products dependency on its trading partner, EU at time t. For example, a higher value of HHI shows that exported products are concentrated in fewer EU markets, whereas a lower value of HHI indicates products are exported to a larger number of EU partners. First, the sample is divided into two groups, the top and bottom percent of HHI, and then (13) is used to evaluate the different effect of standards on products in markets with relatively high and low concentration respectively, the results 24

36 being shown in Table Columns (1)-(3) focus on concentrated markets, the results indicating that additional standard enforcement makes leading varieties more likely to upgrade quality whereas tariff reduction makes laggard varieties less likely to improve quality. Columns (4)-(6) show the results for markets with low concentration, firms producing leading varieties being more likely to upgrade quality due to standard establishment while the magnitude of the coefficient is smaller than that of varieties in highly concentrated markets (column (3) of Table 1.8). These results confirm the earlier findings: leading varieties in highly concentrated markets make a greater effort to upgrade quality as additional standards are established. Due to additional standards pushing leaders to capture the current market by limiting entry of new firms, leading firms can boost their innovation efforts in pursuit of high profits. Therefore, the conclusion concerning the positive effect of standards on leading varieties remains valid. 7. Conclusion In this essay the impact of import tariffs and standards on exporters efforts to upgrade product quality is examined, focusing on EU food imports over the period 1995 to Drawing on Aghion el al. (2004; 2005), a theoretical framework is derived that predicts the direction of the effect of competition depends on the distance of products from the technology frontier, i.e., a non-monotonic relationship between innovation and competition remains valid in a trade setting. Increased competition due to tariff reduction 25

37 drives quality upgrading for leading products, whereas it reduces quality upgrading for the laggard products. The empirical findings support the theoretical predictions that tariff reduction discourages lagging firms from devoting resources on upgrading product quality whereas it encourages leading firms to upgrade product quality. Furthermore, the heavy burden of compliance costs caused by standards enforcement, encourages firms producing leading products to improve quality, but discourages quality improvement for firms with laggard products. For laggard products, the cost of compliance has the most influence on a firm s decision to improve product quality. On the other hand, the probability of capturing market share plays an important role in upgrading product quality for firms with leading products. Therefore, factors other than competition affect quality upgrading when standards limit the entry of new firms. Two major current trends in international trade are tariff reduction and standards designed to promote food safety. The results presented in this essay suggest that these trends have widened the gap between firms at the technology frontier and laggard firms in terms of product quality. Furthermore, the gap is even wider for frontier and laggard firms in developing countries. Therefore, it is important to address the differences in product quality between developed and developing countries in the process of implementing trade policy. One of the ways to mitigate the gap would be to provide a long grace period for developing countries after the effective date of implementation of such trade policies. Future research in this area would focus on evaluating the effect of tariffs and standards under the setting of heterogeneous firm model using firm-level data (Tybout, 26

38 2001). Given the paucity of firm-level data, changes in the composition of firms are neglected in this essay, although heterogeneous quality of varieties is treated as a proxy for firm-level productivity differences in the model. Efforts to distinguish heterogeneous product quality from productivity would be valuable in future research. 27

39 Notes 1. Country-specific trade standards have negative effect on trade flow (Moenious, 2004). EU s Sanitary and Phytosanitary Standards reduces African countries exports (Otsuki et al., 2001). US s Technical Barriers to Trade (TBT) has negative influence on developing countries exports (Essaji, 2008). Maskus et al. (2005) use firm level survey data and explain compliance costs caused by standards have negative influence to developing countries. Chen et al. (2006) find trade standards reduce the probability to export for firms in developing countries. 2. Article 2.4 of the WTO Agreement of Technical Barriers. 3. A trans-national association established by national standard-setting bodies across Europe. 4. Although innovation can include a broad range of activities including patenting, product differentiation, or creating new process, we limit the result from innovation to quality upgrading. 5. Profit is proportional to technology level, and technology level is positively grow with the rate of γ. After new standards establish, the profit changes as t ( v) At ( v) ' At 1 where '. Accordingly, expected profit is a positive function of γ. 6. The three industries are: (manufacture of ice cream), (manufacture of cider), and (manufacture of other non-distilled fermented beverages). 28

40 Figure 1.1: The number of EU standards Panel (A) : Total number of EU Standards Panel (B) : Total page count of EU Standards Source: EU Standard Database 29

41 Figure 1.2: The relation between GDP per capita and fraction of high quality products 30

42 Table 1.1: Competition, compliance costs and innovation Variables Competition External Forces Type Innovation Tariff Decrease Strong - Type1 More - Type2 Less Standard Increase Weak Future expected profit Type1 More or Less Compliance cost Type2 More or Less 31

43 Table 1.2: Descriptive statistics for EU food trade data ( ) Variable Mean Min Max Unit Import value* ,128 Million Euro Import quantity* ,700 Million Kg Export value* ,103 Million Euro Export quantity* Million Kg Product value* 1, ,000 Million Euro Product quantity* 1, ,900 Million Kg Population* ,303 Million people Exchange rate* , Local currency per Euro Distance to EU-15* 6, , Average distances Oil price* Euro Tariff** Tariff rate Standard*** Count Standard pages*** Page count Euro-USD exchange rate* Dollar per one Euro CPI* CPI in Euro *78,012 observations, ** 62,788 observations, and *** 60,183 observations 32

44 Table 1.3: Observed food industries Description (NACE 4 digit REV1) 15.1 Production, processing and preserving of meat and meat products Production and preserving of meat Production and preserving of poultry Production of meat and poultry products 15.2 Processing and preserving of fish and fish products 15.3 Processing and preserving of fruit and vegetables Processing and preserving of potatoes Manufacture of fruit and vegetable juice Processing and preserving of fruit and vegetables nes 15.4 Manufacture of vegetable and animal oils and fats Manufacture of crude oils and fats Manufacture of refined oils and fats Manufacture of margarine and similar edible fats 15.5 Manufacture of dairy products Operation of dairies and cheese making Manufacture of ice cream 15.6 Manufacture of grain mill products, starches and starch products Manufacture of grain mill products Manufacture of starches and starch products 15.8 Manufacture of other food products Manufacture of bread; manufacture of fresh pastry goods and cakes Manufacture of rusks and biscuit Manufacture of sugar Manufacture of cocoa; chocolate and sugar Manufacture of macaroni, noodles, couscous Processing of tea and coffee Manufacture of condiments and seasonings Manufacture of homogenized food preparations Manufacture of other food products n.e.c Manufacture of beverages Manufacture of distilled potable alcoholic beverages Production of ethyl alcohol from fermented materials Manufacture of wines Manufacture of cider and other fruit wines Manufacture of other non-distilled fermented beverages Manufacture of beer Manufacture of malt Production of mineral waters and soft drinks 33

45 34 Table 1.4: Food product quality estimation results ( ) Industry number (1) (2) (3) (4) (5) (6) (7) (8) Industry description Meat Poultry Meat & Fish Potatoes Fruit & Veg Fruit & Crude Oils Variables Mean Median Meat Poultry juice Vegetables & Fats Price * *** ** *** Nest share *** 0.800*** *** 0.755*** 0.403* 1.169*** Population *** ** Number of Varieties , , Observations ,907 1,061 1,419 9, ,435 10,350 2,792 (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) Refined Oils Margarine & Dairy Grain Mills Starches Bread Rusks Sugar Cocoa Macaroni & & Fats Similar Fats & Cheese & Pastry Biscuits Chocolate & Noodles Price ** * * Nest Share 0.611*** 1.005*** 0.765*** *** 1.030*** 0.666*** 0.491* 0.986*** 0.866*** Population ** * Number of Varieties , Observations 2, ,651 4,117 1, ,749 1,706 7,222 1,348 (19) (20) (21) (22) (23) (24) (25) (26) (27) (28) Mineral Tea Condiments Food Other Food Distilled Ethyl Wines Beer Malt Waters Coffee &seasonings Preparations Products Alcohol Alcohol & Soft Drinks Price ** ** ** Nest Share 0.814*** 1.050*** 1.124*** 0.682*** 0.868*** 0.947*** 0.657*** 0.884*** *** Population ** ** Number of Varieties Observations 2,354 2, ,354 2, , ,231 Notes: Country-product and year fixed effects. Robust standard errors are in parenthesis. Significant levels: *** at 1 per cent; ** at 5 per cent and * at 10 percent 34

46 Table 1.5: Correlation and OLS results with fraction of exporters products with highest quality Regressors Pair-wise Correlation OLS OLS OLS (1) (2) (3) (4) Ln GDP per capita *** (0.002) Ln labor force with primary *** education (0.000) Ln labor force with tertiary 0.015*** education (0.000) Fixed effect Yes Yes Yes Adj R-squares Observation 77,009 35,517 35,510 Notes: Product-year and country-year fixed effects. Robust standard errors are clustered by country. Significant levels: *** at 1 per cent; ** at 5 per cent and * at 10 percent. 35

47 Table 1.6: Quality upgrading and competition Regressors All countries Non-OECD OECD (1) (2) (3) (4) (5) Lag DF(β) Lag Tariff(θ 1 ) Lag Trf * DF(η 1 ) Lag Standards(θ 2 ) Lag Stnd * DF(η 2 ) *** (0.142) 1.864*** (0.503) *** (0.904) *** (0.131) *** (0.066) 0.417*** (0.093) *** (0.136) 1.828*** (0.506) ** (0.913) *** (0.066) 0.425*** (0.093) *** (0.219) 2.502*** (0.644) * (1.112) *** (0.066) 0.418*** (0.097) *** (0.398) 2.213*** (0.777) (1.728) * (0.159) 0.423* (0.231) Product-year FEs Yes Yes Yes Yes Yes Country-year FEs Yes Yes Yes Yes Yes Adj R-squared Observations 18,825 18,825 18,825 10,771 7,789 Notes: Dependent variable is change in logged quality and all explanatory variables are with its three-year lag value. Product-year and country-year fixed effects. Robust standard errors are clustered by country. Significant levels: *** at 1 per cent; ** at 5 per cent and * at 10 percent 36

48 Table 1.7: Tariffs and standards effect on quality improvement: all observations Variables Distance to the frontier Quality Improvement Tariffs Decrease by 10 pt Close up by 2.7% Far down by 18.3% Standards Increase by 1 page Close up by 19.3% Far down by 23.2% 37

49 Table 1.8: Tariff and standards effects on quality improvement: OECD and non- OECD Tariff Standard Variables Distance to the frontier Quality Improvement Decrease by 10 pt Increase by 1 page OECD Non OECD OECD Non OECD Close up by 2.1% Far down by 22.1% Close up by 4.9% Far down by 25% Close up by 11.9% Far down by 30.4% Close up by 23.3% Far down by 18.5% 38

50 Table 1.9: Robustness checks - weighted unit value as dependent variable Regressors Δ in unit price (1) (2) (3) Lag DF(β) Lag Tariff(θ 1 ) Lag Trf * DF(η 1 ) Lag Standard(θ 2 ) Lag Stnd * DF(η 2 ) *** (0.126) 2.747*** (0.486) 0.958** (0.484) *** (0.093) *** (0.065) 1.543*** (0.114) ** (0.136) 2.721*** (0.481) (0.485) *** (0.068) 1.528*** (0.115) Product-year FEs Yes Yes Yes Country-year FEs Yes Yes Yes Adj R-squared Observations 17,092 17,092 17,092 Notes: Unit price has been trimmed down for 5% and 95%. Dependent variable is change in logged unit price and all explanatory variables are with its three-year lag value. Product-year and country-year fixed effects. Robust standard errors are clustered by country. Significant levels: *** at 1 per cent; ** at 5 per cent and * at 10 percent. 39

51 Table Robustness checks - quality changes in low and high HHI market Δ in quality (low competition) Δ in quality (High competition) Regressors (1) (2) (3) (4) (5) (6) Lag DF(β) *** *** *** *** (0.3519) (0.2161) (0.3291) (0.5244) (0.2352) (0.5249) Lag Tariff(θ 1 ) (1.365) (1.3431) (1.2132) (1.2157) Lag Trf x PF(η 1 ) ** ** (1.377) (1.1096) (3.043) (3.048) Lag Standard(θ 2 ) (0.1569) (0.1582) (0.1043) (0.1041) Lag Stn x PF(η 2 ) 0.897*** 0.928*** 0.466** 0.474** (0.1969) (0.2013) (0.2022) (0.2091) Product-year FEs Yes Yes Yes Yes Yes Yes Country-year FEs Yes Yes Yes Yes Yes Yes Adj R-squared Observations 1,585 1,585 1,585 2,161 2,161 2,161 Notes: Dependent variable is change in logged quality and all explanatory variables are with its three-year lag value. Product-year and country-year fixed effects. Robust standard errors are clustered by country. Significant levels: *** at 1 per cent; ** at 5 per cent and * at 10 percent. 40

52 Chapter 2: Trade liberalization and Endogenous Quality Choice in Food and Agricultural Trade 1. Introduction Recent trade theory which accounts for heterogeneous productivity across firms helps to clarify the hidden causes and consequences of international trade at the country-level. The heterogeneity in a firm s productivity, size, and factor intensity induce differences in trade participation. Since only high productive firms export and this improves average productivity at the national level, comparing the average productivity without considering firm level activity might cause a problem. Accordingly, in analyzing the effects of trade liberalization it is important to take account of the number of exporters (extensive margin of trade). Although significant trade liberalization has been achieved, concerns over food safety have led many countries to adopt non-tariff measures. It is, therefore, necessary to assess the impact of trade frictions on food and agricultural trade in the context of extensive (i.e., number of exporters) and intensive margins (i.e., export flows). The purpose of this essay is to evaluate the effect of trade frictions on food and agricultural trade by illustrating heterogeneous-firm trade model with endogenous product quality (Baldwin and Harrigan, 2011; Johnson, 2012; Kugler and Verhoogen, 2012). A key assumption is that both consumers and firms can recognize product quality differences when they make decisions. Having identified the link between product 41

53 quality and firm productivity, the correlation between export thresholds and trade costs is derived within the theoretical framework. Next, our model is used to empirically assess the determinants of bilateral trade flows by decomposing margins into intensive and extensive. The two-stage estimation procedure of Helpman et al. (2008) is extended to estimate the extensive margin with country-level data by introducing non-tariff measures as one of the fixed trade costs. In this way, consistent empirical estimates are provided in the context of heterogeneous productivity and quality. Several studies have considered endogenous quality of consumer choices. Amiti and Khandelwal (2012) conclude that lowering tariffs raises the quality upgrading effort for relatively higher-quality products while decreasing quality upgrading effort for relatively lower-quality products. Curzi and Olper (2012) analyze the behavior of Italian food exporting firms and find that more efficient firm exports more qualified goods towards destinations with higher income. These essays derive product quality from consumer demand only which prohibits the possibility of producer quality choice. Another strand of literature develops the theory with quality choice by firms. Fan et al. (2014) show a framework with firms quality choice over imported intermediates. They provide evidence that reduction in tariff generates incumbent firms to increase the price and quality of exported products using Chinese firm data. Kugler and Verhoogen (2012) introduce the model of endogenous quality choices with heterogeneous firms. They conclude that firms having high capability perform well in export markets because they use intermediate inputs with higher-quality to sell them at higher prices. Johnson (2012) also develops the endogenous quality choice model. He argues that price varies across 42

54 countries and finds positive relationship between the unit value price and the hardship of entering the markets. The price rises when the probability of entering into the import markets low. Baldwin and Harrigan (2011) provide a general equilibrium trade model that considers the firm heterogeneity of productivity and quality. According to their result, the export participation decrease as trade costs increases whereas the participation increases as the destination country has high GDP per capita. However, little attention has been given to the effect of the trade costs on productivity or firms quality choices. Our study fills a gap in the existing literature by deriving the heterogeneous-firm trade model with endogenous quality to elaborate the impacts of trade costs on trade flows, by decomposing trade margins into intensive and extensive. It also contributes to the empirical literature by introducing a way to reflect the impact of selection on exporting in the context of product quality. By extending the approach of Helpman et al. (2008), non-tariff measures (NTM) and governance indicators are considered as fixed trade costs and thus determine entry into the export market. In particular, NTM notification data from the World Integrated Trade Solution (WITS) and the Integrated Trade Intelligence Portal (I-TIP) is firstly used in this essay to analyze the impact on trade participation. The rest of the essay is organized as follows. In Section II the theoretical model which provides the basis of the empirical specification is described. In Section Ⅲ the empirical specification and methodology is presented. In Sections IV and V the data are 43

55 described and estimation results discussed. Lastly, in Section VI some concluding remarks are offered. 2. Theoretical Background Here we modify the endogenous quality choice models of Kugler and Verhoogen (2012) and Johnson (2012). Consumers are allowed to observe variety in quality while firms can also choose the level of product quality. Quality is a function of a firm s productivity Preferences Assume that a representative consumer with standard asymmetric constantelasticity-of-substitution utility function over a continuum of differentiated products (Kugler and Verhoogen, 2012); U [ ( q( ) x( )) d ] ( 1)/ ( 1) 1 (1) where ω indicates the final good products and Ω indexes the set of all differentiated products; σ is the elasticity of substitution between products; x(ω) is the quantity demanded, and; q(ω) is the product quality. Quality is referred as the product characteristics that the consumer grants value and is chosen by the producers. Maximization of (1) yields the Equation (2), quantity of product ω demanded; x( ) q( w) p( ) EP 1 1 (2) where p(ω) is the price of the product; P is as an aggregate price index 1 1/(1 ) p( ) P ( ( ) dw), and E is the aggregate consumption of the available qw ( ) products, which is equal to the income of country i. Therefore, E and P represent the 44

56 externally given market demand. Demand increases with improved product quality and decreases with product price Firm Behavior Suppose the world consists of J countries, where j=1,2, J. Each country has a mass of firms N j and goods are produced and sold within a monopolistically competitive market structure. Let define firm in country j has marginal cost cato produce one unit j of final good. The parameter a represents the number of units of input and c j indicates the cost of input per one final good (Helpman et al., 2008). The profit maximization problem for country-j producer, subject to the demand with constant elasticity σ, yields the equilibrium product price which is a constant mark-up over marginal cost. The country-j producer optimally charges the mill price, p j( c, a, q) ( )( c ja). If the product is exported to country i, the producer will 1 bear two additional costs: fixed cost f ij and transport cost ij ( f 0 and 1 for i j ). We assume the fixed costs and transport costs ij ij within the home country equal zero and one, respectively. The producer will now x charge the export price p ij ij p j ( c, a) ( )( c ja ij ) 1 to maximize export profit. The equilibrium export profit of country j towards country i is 1 ca 1 ij j 1 ij ( a) ( ) ( ) Ei fij 1 qpi (3) The productivity level of firms is represented by the inverse of a, 1/a (where a>0). Following Melitz (2003), the productivity of each firm a is determined by a random 45

57 draw from a truncated Pareto distribution g(a) with a cumulative distribution G(a) with range [a L, a H ] where 0< a L < a H is assumed. Productivity 1/a is firm-specific but the distribution function is the same across all countries. Now we define the relationship between quality and productivity. The firm-specific productivity parameter is linked with product quality as in q 1 a 1 ( ) where 1 (4) Quality increases with higher productivity, and the quality elasticity θ-1 indicates the extent to which productivity and quality are related. The rate of change in productivity is expected to differ according to the scope for quality differentiation (θ-1), representing the availability of technology that convert productivity into improved product quality (Kugler and Verhoogen, 2012). Defining quality as a function of productivity simplifies the model and allows firm heterogeneity to be expressed in a single dimension. Accordingly, all firms have the same parameters such as input costs, but have different productivities that randomly assigned. Substituting q from equation (4) into (3), a firm in country j has equilibrium profit from export to country i as follows: ca 1 1 ij j 1 ij ( a) ( ) ( ) Ei fij 1 Pi (5) 2.3. Export Participation A firm determines whether to either remain or exit the export market after learning * about the revealed productivity. This decision yields a cutoff productivity 1/ a where a ij firm makes zero profit (Melitz, 2003). According to equation (5) firms profits are always positive if they sell products domestically because fii 0. Thus, no firm would 46

58 exit from the domestic market. On the other hand profits from exports become positive * * * only when a firm has a a ij, where a is defined by ij ij ( aij ) 0 (Helpman et al., 2008) 1 ca ( ) ( ) 1 P * 1 ij j ij 1 i E i f ij (6) From the zero-profit condition equation we derive, the productivity threshold is * * derived as Equation (7). For simplicity, we denote as the inverse of ij a. ij f c * * ij 1/( 1) ij j 1/ ij 1/ aij [ ( ) ] 1 Ei Pi (7) Note that changes in the importer s expenditure and competitiveness (E i, P i ) and trade frictions (f ij, τ ij ) lead to variation in export thresholds. Destinations having higher expenditures (large E i ) or that are less competitive (higher P i ) allow firms with lower productivity to enter the market. However, higher trade frictions raise the productivity threshold and restrict entrance to firms with high productivity. Some comparative statics results describe how fixed and variable trade costs affect the productivity threshold. c [ ( ) ( )] 0 f E P * ij ( 1) 1 ( 1) ij j 1/ fij ij ( 1) 1 i i (8a) f c [( )( ) ( )] 0 * ij 1 (1/ ) 1 ij 1/( 1) j 1/ ij ij 1 Ei Pi (8b) These partial derivatives indicate a positive relationship between trade costs and productivity thresholds. Since θ and σ are greater than 1, the cutoff productivity decreases with a reduction in variable and fixed trade costs as in equation (8a) and (8b). Increasing the variable costs raises the price; accordingly, the productivity threshold for 47

59 entering the export market becomes higher. Moreover, an increase of fixed trade costs in the destination market hinders export activities; therefore, exporters with lower productivity are not likely to serve the export market. Furthermore, the results indicate that the size of the effect also relies on how much quality is elastic with respect to productivity. The elasticity parameters θ are exogenously given. 3. Empirical Analysis Following Helpman et al., (2008), no firm would find it profitable to export for importer i if a * ijt is less than the minimum value in range (a L ). More explicitly, V ijt is denoted as bilateral trade volume (Helpman et al., 2008): V ijt * aijt (1 ) * a dg( a) for a a ijt al L (9) 0 otherwise With a truncated Pareto distribution V ijt is: k a Vijt ( ) Wijt, k (1 ) k ( 1) ( a a ) where W ijt k (1 ) L k k H L a ijt k (1 ) max{( ) 1,0} a L From the demand function (2) and the output price, the value of product h produced from exporting country j to importer i s market is: M c 1 ijt jt 1 ijt ( ) ( ) Eit N jtvijt 1 Pit (10) Since the productivity threshold cannot be observed directly, trade value is used to infer the relationship between trade costs and cutoff productivity. Since the effect of variable 48

60 costs on trade value is negative (11a), the effect of cutoff productivity on trade flow is also negative because the effect of variable costs on cutoff productivity is known to positive (11b). Trade costs negatively affect trade value: M c E N V ( 1) ( ) 0 ijt 1 jt ijt it jt ijt c jt ijt Pit Pit (11a) M M * ijt ijt ijt 0 * ijt ijt ijt (11b) The bilateral trade value ( M ijt ) can be derived by mapping the expenditure level of importer i s market at time t ( E ), the number of exporting firms at time t ( N ), the it unit input cost for producing products in exporting country at time t ( c jt ), the fixed costs ( f ijt ) and the variable trade costs ( ijt ). Variable trade costs are defined as jt hij 1 uhij De. Unmeasured trade friction u ij ijh is assumed to be stochastic 2 (independent and identically distributed, i.i.d.), uhij ~ N(0, u ), and Dij indicates the geographic distance between countries i and j. Equation (9) can be expressed in loglinear form (Helpman et al., 2008): ln m ln d w u ijt it jt ij ijt ijt where ( 1) ln P ln E and (1 ) c ln N, representing the fixed it it it jt jt jt effects of the time-importer and time exporter, respectively Selection Equation Since no firm would enter the export market if the most productive firm determines not to export, Helpman et al. (2008) suggest using zero trade flows to infer the relative 49

61 productivity threshold, the ratio of firms exporting from country j to country i. The latent variable Z ijt is used to control for the number of exporting firms in country j towards country i at time t. It represents the ratio of the most productive firms export profit to the fixed trade costs. A positive trade flow is observed only when Z ijt >1: Z ijt 1 c ( ) ( ) 1 f 1 ijt jt 1 (1 ) L Pit ijt a E it The firm with cutoff productivity is Z ijt =1, hence the relative threshold a ( ) a * 1 ijt (1 ) Zijt L indicates decreasing profitability for the most capable exporting firm serving market i. Given a fixed a L, the proxy for the unobserved share variable (Z ijt ) relies on productivity threshold, a ijt *. If a ijt * is less than a L, then Z ijt is also less than 1, indicating no exports. Otherwise, if a ijt * > a L, exports will occur with a certain number of exporters. The fraction of exporting firms grows as the most capable exporting firm becomes more profitable. With this relationship, we can denote W ijt as ( k (1 ))/( (1 )) Z ijt 1. Fixed trade costs are assumed stochastic with unmeasured i.i.d. trade frictions (v ijt ), so that the fixed costs are decomposed into exporter-, importer-, and time specific fixed trade costs: f 2 exp( v ), where v ~ N(0, ) ijt j i h ijt ijt ijt v ln z d ijt 0 1i 2 j 3t ij ijt ijt 2 2 where ijt ~ N(0, u v ) The error term ijt is assumed i.i.d, following a normal distribution with a mean of zero and variation Since z ijt is not directly observed, the u 50 v

62 indicator function T ijt is used. T ijt = 1 when j exports to i, and 0 otherwise. The probability that country j exports to country i is represented as ijt, conditional on the observed variables: Pr( T 1 observed var iable) ijt hij * * * * * * * * 0 1i 2 j 3t dij ijt zˆ ijt Xijt ( ) ( ) ( ) (12) where Φ(. ) is the cumulative distribution function of the unit-normal distribution; and starred coefficients index the estimated coefficient divided by σ η Trade Equation Using (12), ˆijt is defined as the predicted probability that country j exports to i * and define the predicted latent variable as z ijt zijt /. A consistent estimate for W ijt is equivalent to max{( z ) ijt 1,0} where ( k ) / ( ). To construct consistent estimates for E[ w., T 1], the latent variable ijt ijt z * ijt conditional * on positive observations ( E[ z., T 1] ) is estimated by Helpman et al. (2008) suggest to replace u v / ijt ijt ijt ijt u v * z ijt ( uijt vijt ) /. with its expectation conditional on Tijt 1, which is inverse Mills ratio ˆ * ijt, to deal with the effect on the sample-selection on the term uijt vijt / u v. Therefore, the consistent estimates for would be wˆ ( ) ln{exp[ ( zˆ ˆ )] 1}. In addition, the trade equation includes * * * ijt ijt ijt the inverse Mills ratio, ˆ * * * ijt zˆ ijt zˆ ijt ( ) / ( ) to correct sample selection error. Therefore, the second stage estimation equation is derived as Equation (13): 51

63 ln m d ln{exp[ ( zˆ ˆ )] 1} ˆ * * * ijt it jt ij ijt ijt u ijt ijt Where corr( u, )( / ) and ε ijt is an i.i.d. error term un ijt ijt u (13) Equation (13) is estimated using maximum likelihood procedure. The additional control * variable z corrects for bias due to the unobserved heterogeneity and potential ˆijt ˆ * selection bias is corrected by the term ijt. 4. Data 4.1. Data Description The data sample consists of bilateral trade flows of agricultural and food products for 159 countries over the period 2010 to There are 77,488 bilateral trade flows observations and zero trade flows take 40.18% (31,140) of the total observation. Bilateral trade data is acquired from the database of the Food and Agriculture Organization of the United Nations (FAO). Trade flows in the food and agricultural sector are observed. (see Table 2.1). Data on variable trade costs such as distance, shared borders, and shared language, come from the Centre d Etudes Prospectives et d Information Internationales (CEPII). To achieve identification in the second stage, fixed trade costs should provide exclusion restrictions. Since fixed trade costs are not directly observable, three sets of governance indicators (quality of regulation, governmental effectiveness, and freedom to trade) and NTMs are used as proxies for trade facilitation. Governance indicators are appropriate proxies of fixed trade costs because the level of governance affects decisions on trade 52

64 but not the actual bilateral trade value. Summary statistics of the data are provided in Table 2.2. Country-level governance indicators are obtained from the World Bank database. The quality of regulations and governmental effectiveness indicators provide information about trade accessibility. These indicators range from -2.5 to 2.5, implying more accessible countries have a higher value, close to 2.5. We, later, define governmental effectiveness and quality of regulations variables as binary indicators, which equal one if the indicators are greater than the median for both the importer and exporter. Accordingly, a country pair which has better governmental effectiveness and quality of regulations has a value of one. We expect a positive effect of governmental effectiveness and quality of regulations on the selection into trade. The freedom to trade internationally index from Fraser Institute is also used to infer country s governance. The index consists of the controls of the movement of capital and human capita, trade regulation, tariffs, exchange rates in black market (Gwartney et al., 2016). This index reflects the impact of trade freedom on the exporter s decision and a positive effect on the decision is expected. Destination market with lower tariffs, lower regulatory barriers, and few controls on movement of capital easily attract exporting firms. A trade standards variable is used as a proxy for fixed trade costs. Data comes from the World Integrated Trade Solution (WITS) of the World Bank and the Integrated Trade Intelligence Portal (I-TIP) of the World Trade Organization (WTO). In the database, NTMs include anti-dumping measures, sanitary and phytosanitary measures, 53

65 and technical barriers to trade. This essay constructs a non-tariff measure (NTM) variable, which varies across countries imposing standards, partners affected, related products, and the effective period, by combining the number of measures applied by importers in the agricultural and food trade. This novel NTM variable allows us to analyze the effect of country-specific NTMs on trade, and to examine whether NTMs act as barriers or catalysts Non-tariff Measures, Governance and Export Participation Fixed trade costs which are associated with access to foreign markets generate selection into export. We assume these market access costs are exogenously given. Costs for advertising, distribution or conforming to foreign regulations or law are considered as fixed trade costs. Previous empirical studies have used governance indicators as proxies for fixed trade costs, since importers with efficient governance (e.g. advanced control of corruption, high governance efficiency) are easy to access and require low entry costs. Helpman et al. (2008) use common language and common religion, while Manova (2013) applies data from Djankov et al.(2002) on the regulation costs of firm entry. Prehn et al. (2015) use governance indicators of regulatory quality. This essay applies governance indicators from the World Bank, the freedom to trade internationally index from the Fraser institute, and NTMs to reflect trade fixed costs. Tables 2.3 and 2.4 show the top and the bottom ten countries with good governance, respectively. In Table 2.3, most of high-income countries are recorded as countries with good governance. Many developed European countries, especially northern European countries such as Denmark, Sweden, Finland, and Norway, are in the list. Singapore and 54

66 Canada are also show improved governance. Although Mauritius and Jordan is middleincome countries, they have high level of internationally trade freedom. This is because Mauritius is recorded with low control of the shift of capital and human capital, low tariffs and no black-market exchange. Also, Jordan has relatively high scores on the overall criteria, including low tariffs, no black-market and well-controlled shift of capital and people (Gwartney et al., 2016). In Table 2.4, most African countries are recorded as countries with low level of governance. Some Latin American countries such as Venezuela and Argentina have low freedom to trade internationally. Some Arab countries, particularly Iran, Syria, and Yemen, also indicate low levels of governance. As expected, the bottom-ranked countries are mainly composed of developing countries in Africa: Zimbabwe, Cameroon, the Central African Republic (CAR), and the Congo. This essay includes NTMs and governance data to recover fixed trade costs because strict and frequent NTMs imposed by importers act as fixed trade costs by raising the difficulty to enter their markets. By including the NTM variable as fixed trade costs, we can assess the effects of fixed trade costs on export participation. Table 2.4 indicates the top products with high NTM numbers when averaged across all trading partners by product category. The number of standards for meat and cereals are significantly higher than those for other products. This is largely due to the frequent occurrence of various animal diseases and food additives to cereals that are subject to maximum residue limits (MRLs). On the other hand, products of milling, sugars, and cocoa preparation have low levels of NTMs, implying the easiness of entry. 55

67 Table 2.6 shows the countries with the highest number of NTMs along with those which do not impose NTMs. Japan has the largest number of standards regarding food safety followed by a number of European countries which have strict agricultural and food standards. On the other hand, large numbers of African and Arabian countries have no food safety standards. These data suggest that high-income countries have a tendency to become sensitive towards the food safety of imported goods whereas low income countries do not show this tendency. 5. Estimation Results 5.1. Specification Estimation of the two-step model controls for the heterogeneity of firm productivity and quality. The first step is to estimate the export participation equation (12) where the dependent variable is a binary indicator of whether trade exists. A positive export flow shows that a country has at least one firm whose productivity is high enough to export. On the other hand, a zero trade flow indicates that no firm is productive enough to enter the export market. The estimation results from the first stage are used to control the extensive margin and sample selection bias in the estimation of trade equation (13). In specifying the first-stage equation, the probability that country j exports to importer i at time t, conditional on the variable and fixed trade costs is estimated. Accordingly, we include an importer-specific variable (Gv it, Reg it ), country-specific variables (distance between countries (DIST ij ), dummies for a shared border (ADJ ij ), common language(lang ij ), and regional trade agreement (RTA ij )), country- and 56

68 product-specific variables (NTM ijt ), and the importer, exporter, and time fixed effects. The specification of selection equation is: Pr( T 1 observed var iable) (14) ijt ijt ( DIST ADJ LANG RTA * * * * * * * * 0 1i 2 j 3t 1 ij 2 ij 3 ij 4 ij Gv Re g NTM ) * * * 1 it 2 it 3 ijt In the second-stage trade equation, the modified gravity model is estimated including a nonlinear function of the Probit index ( wˆ ( ) ln{exp[ ( zˆ ˆ )] 1} ) and the * * * ijt ijt ijt ˆ * ijt inverse Mills ratio ( ). The Probit index term controls the endogenous number of exporters (extensive margin) and the inverse Mills ratio controls the sample selection bias. Fixed trade costs satisfy the exclusion restriction requirement. These costs influence the probability of exporting, but do not directly influence the level of exports. Accordingly, we exclude fixed trade costs in the second stage equation. Therefore, the specification is: ln m ln DIST ln ADJ ln LANG RTA ijt 0 it jt 1 ij 2 ij 3 ij 4 ij ln{exp[ ( zˆ ˆ )] 1} ˆ * * * ijt ijt u ijt ijt (15) 5.2. Results The estimates for the baseline gravity model are presented in the first four columns of Table 2.7. Observations track country pairs that trade, and we use both exporter and importer fixed effects as well as time fixed effects. The first and second columns provide the simple ordinary least squares (OLS) results, and the third and fourth 57

69 columns show the results of Poisson Pseudo Maximum Likelihood (PPML). The OLS estimates of the log-linear model can be biased due to measurement error. Zero trade flows are not defined in logarithm. If zero trade flows are simply missing due to rounding errors or reflects information on trade barriers (e.g., high transportation costs), dropping them in the OLS estimation causes measurement error. Therefore, PPML estimation is proposed to control the problem (Silva and Tenreyro, 2006). Columns (1) and (3) are estimated without fixed cost variables while Columns (2) and (4) are estimated with fixed costs. The log-transformation is applied to the freedom to trade and the NTM variable because only those two have value besides zero and one. NTM is weighted according to the ratio of the number of NTMs and the maximum number of NTMs within the observed time period t (denoted as wntm). The results indicate that country i trades more from country j at time t when the two countries are geographically closer, share a border and an official language, and participate in the same regional trade agreement. We notice that all coefficients from both OLS and PPML methodologies are statistically different from zero at 5% level or less than 5 % level and have the expected signs, although the magnitude of the coefficient in the absolute value is different. Regulation and wntm variables are significant (p-value < 1%) for OLS estimation whereas freedom, governance, and NTMs are significant (p-value < 1%) for PPML. The signs of the coefficients also follow expectations, showing that countries with a higher freedom to trade, better governance and regulation, and fewer NTMs trade more. 58

70 Columns (5) and (6) show the results of the selection equation which provides the predicted probability of trade. The positive coefficients of regulation indicate that any country with better quality regulation has a higher likelihood to participate in trade. The freedom to trade coefficient is also positive, suggesting that a country that has a greater freedom to trade has a higher probability of trade participation. The NTM variable provides evidence that non-tariff barriers negatively influence the likelihood of trade. Importers having larger numbers of NTMs have a tendency to reduce the probability for exporters to enter their markets. The second step estimation results are reported in Table 2.8. The first two columns provide the OLS and PPML benchmark estimations. Although PPML estimator performs better than OLS in that they solve measurement error problem, the coefficients in Column (1) and (2) are confounded with the effects of unobserved firm heterogeneity. Column (3) and (4) are the results of the first and second step estimations of equations of (14) and (15). Using the estimates from the first stage equation, we construct both an inverse Mills ratio for controlling sample selection bias and a measure of firm selection to export for controlling heterogeneous firm productivity. Fixed trade cost variables are included in the first estimation stage as exclusion variables, which affect the probability to trade but not trade flow. With the help of the non-linear coefficient δ for ˆ * w ijt and the coefficient u for ˆ * ijt, the coefficients of variable trade costs are consistent since heterogeneous firm productivity and sample selection error are controlled. 59

71 As expected, both coefficient δ and u are statistically significant. The positive coefficient of ˆ * w ijt suggests that a greater proportion of exporting firms allows higher bilateral trade because term ˆ * w ijt measures firm selection to export implying the extensive margin. In other words, higher trade value does not just come from lower trade costs but also from a larger proportion of exporters. Thus, the coefficients of Column (4) show the consistent estimates taking account of firm heterogeneity. These findings are not sensitive to the distribution assumptions on firm heterogeneity (Silva and Teneyro, 2015). Column (5) of Table 2.8 provides the results with a relaxed assumption on Pareto distribution for firm productivity, and thus the functional form of ˆ * w ijt (Helpman et al., 2008). Instead of constructing a precise estimate ˆ * ijt for w, cubic polynomials in z ijt are used in the second stage of estimation. The joint normality assumption on the unobserved costs is further relaxed. Instead of using variables representing firm heterogeneity and sample selection, we use a non-parametric functional form with directly predicted probabilities in Column (6). The 50 bins of indicator variables are used to approximate the arbitrary functional form of predicted probability ˆijt, and the dummies for each bin are included in OLS second stage estimation. All estimated coefficients are similar to the Column (4), indicating the Pareto distribution and joint normality assumption do not influence heterogeneous firm trade model estimation. To evaluate the relative significance of heterogeneous productivity and firm selection into trade, the bias decomposition is conducted and the results are shown in 60

72 Columns (7) and (8). The bias is decomposed to examine which correction is needed most from the standard gravity model which leads two different types of bias. Accordingly, two different specifications of the second-stage equation are used. The correction for a selection bias only is reported in Column (7), and the correction for an unobserved heterogeneity bias only is reported in Column (8). All the coefficients of Column (7) are higher in absolute value and the estimates are similar to the OLS baseline estimation. On the other hand, the coefficients of Column (8) are similar to MLE estimates, providing insight that most of the biases are driven by unobserved heterogeneity (the proportion of exporting firms). Thus, ignoring firm heterogeneity in the standard gravity model induces significant bias. In the theoretical framework section, the firm s selection into export is determined by the productivity threshold, which consists of firm heterogeneity in quality and productivity. To show changes in the proportion of exporting firms with respect to firm heterogeneity in quality and productivity, the sample is divided into OECD and non- OECD member countries. Importers with OECD membership are assumed to have a higher threshold, and so require high-quality goods to import. On the other hand, importers without OECD membership have a lower threshold, meaning they import relatively low-quality goods. We expect a larger extensive margin for bilateral trade towards non-oecd member countries and a less extensive margin towards OECD countries. In Table 2.9 the estimation results for OECD and non-oecd member countries are provided in Panels A and B, respectively. Column (1) provides the baseline gravity 61

73 model estimation, and Column (2) shows the result after controlling for firm heterogeneity and sample selection. The bias is decomposed into sample selection and firm heterogeneity, as shown in Columns (3) and (4). The dependent variable in Panel A is the trade flow from country j towards OECD member country i at time t. Panel B shows the result using the non-oecd importers sample, so the dependent variable is trade flow from country j towards non-oecd member country i at time t. The effect of trade frictions on bilateral trade flow towards OECD importers becomes stronger when the proportion of exporting firms is considered since the productivity threshold is relatively higher than in other markets. On the other hand, the effect of trade frictions towards non-oecd member countries becomes weaker after consideration of the extensive margin, because exporting firms are able to enter the market relatively easily due to the lower export threshold. Therefore, we conclude that importers requiring high-quality products have a relatively higher productivity threshold restricting exporting firms, whereas importers with a relatively lower productivity threshold allow more firms to access the export market. For the Panel A estimation, trade frictions are biased downward in the case of OLS estimation compared with MLE estimation. The bias decomposition procedure suggests that the bias of the OLS estimation comes from unobserved firm heterogeneity, since the estimated coefficient are closer in absolute value to the MLE estimates in the case of Column (4). In the case of non-oecd importers, the OLS estimates of trade frictions are biased upward compared with MLE estimates, despite the proportion of exporting firms being statistically insignificant. According to the bias decomposition results, the 62

74 correction of the firm heterogeneity is more crucial than the correction of the sample selection approach. 6. Conclusion This essay suggests that selection into exporting should be considered in the evaluation of the effect of trade frictions in food and agricultural trade. This selection is determined by the threshold for exporting, which depends on firm productivity and quality of products. Under the assumption that the most capable firms choose to produce highquality goods, the findings of a negative relationship between threshold and trade costs indicate that only highly productive firms making high-quality goods are able to enter the export market as trade costs increase. Since firm activities are not directly observed from national-level data, we apply a modified version of the Helpman et al. (2008) model. This model allows us to reflect the extensive margin in analyzing the impact of trade frictions by using zero trade flows as indicators for the export threshold. An important contribution of the modified approach is to use freedom to trade, governance indicators, and non-tariff measures as proxies for fixed trade costs in food and agricultural trade. Those three variables serve as exclusion restrictions for the second-stage estimation because the trade facilitation status of importers affects the probability of exporting to other countries but does not influence directly trade flow. Unlike previous studies, this essay considers product quality as well as firm productivity as determinants of the productivity threshold. The estimation results confirm that standard methodologies used to estimate the impact of trade frictions, such 63

75 as OLS, produce upwardly biased estimates. By controlling for extensive margins, the alternative model specification is better fitted to the data and produces statistically unbiased and consistent estimates. Further, based on our estimations with OECD and non-oecd importers, we conclude that importers with relatively higher trade thresholds restrict the number of exporting firms of trade partners whereas importers with relative lower threshold allow more firms of trade partners to enter the export market. 64

76 Table 2.1 : List of observed countries and industries Observed countries Product Description (21) Afghanistan, Albania, Algeria, Antigua and Barbuda, Argentina, Armenia, Australia, Austria, Azerbaijan, Bahamas, Bahrain, Bangladesh, Barbados, Belarus, Belgium, Belize, Benin, Bhutan, Bolivia, Bosnia and Herzegovina, Botswana, Brazil, Brunei Darussalam, Bulgaria, Burkina Faso, Burundi, Cabo Verde, Cambodia, Cameroon, Canada, Central African Republic, Chile, China, Colombia, Comoros, Congo, Costa Rica, Croatia, Cuba, Cyprus, Czech Republic, Côte d'ivoire, Denmark, Djibouti, Dominican Republic, Ecuador, Egypt, El Salvador, Estonia, Ethiopia, Fiji, Finland, France, Gabon, Gambia, Georgia, Germany, Ghana, Greece, Grenada, Guatemala, Guinea, Guyana, Honduras, Hungary, Iceland, India, Indonesia, Iran, Ireland, Israel, Italy, Jamaica, Japan, Jordan, Kazakhstan, Kenya, Kiribati, Kuwait, Kyrgyzstan, Latvia, Lebanon, Libya, Lithuania, Luxembourg, Madagascar, Malawi, Malaysia, Maldives, Mali, Malta, Mauritania, Mauritius, Mexico, Mongolia, Morocco, Namibia, Nepal, Netherlands, New Zealand, Nicaragua, Niger, Nigeria, Norway, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Portugal, Qatar, Republic of Korea, Republic of Moldova, Russian Federation, Rwanda, Saint Kitts and Nevis, Saint Lucia, Saint Vincent, Sao Tome and Principe, Saudi Arabia, Senegal, Seychelles, Sierra Leone, Singapore, Slovakia, Slovenia, Solomon Islands, South Africa, Spain, Sri Lanka, Suriname, Swaziland, Sweden, Switzerland, Syrian Arab Republic, Thailand, The former Yugoslav Republic of Maced.., Togo, Tonga, Trinidad and Tobago, Tunisia, Turkey, Tuvalu, Uganda, Ukraine, United Arab Emirates, United Kingdom, Tanzania, United States of America, Uruguay, Vanuatu, Venezuela, Viet Nam, Yemen, Zambia, Zimbabwe Live animals; animal products Meat and edible meat offal Dairy produce, birds egg Products of animal origin Edible vegetables and certain roos Edible fruit and nuts Coffee, tea, mate Cereals Products of the milling industry Oil seeds and oleaginous fruits Lac; gums, resis and other vegetable saps Vegetable plaiting materials Animal or vegetable gats and oils Preparations of meat, of fish Sugars and sugar confectionery Cocoa and cocoa preparations Preparations of cereals, flour, starch Preparations of vegetables, fruit, nuts Miscellaneous edible reparations Beverages, spirits and vinegar Residues and waste from the food industries 65

77 Table 2.2: Summary statistics Variable Mean Std. Dev. Min Max Unit/ Range Import quantity Thousand Euro Import value Thousand Kg Export quantity Thousand Euro Export value Thousand Kg Quality of regulation ~2.5 Government efficiency ~2.5 Freedom to trade Indicator Non Tariff Measures Count Adjacency Indicator Common language Indicator Distance Indicator Regional Trade Agreement Indicator 66

78 Table 2.3: Top 10 countries with good governance for year 2012 Freedom Governance Regulation Singapore 8.54 Finland 2.23 Singapore 1.97 New Zealand 8.24 Singapore 2.17 Sweden 1.90 Switzerland 8.19 Denmark 1.98 New Zealand 1.85 Mauritius 8.09 Sweden 1.95 Finland 1.83 UAE 8.08 Norway 1.91 Denmark 1.80 Ireland 7.9 Switzerland 1.89 Australia 1.78 Canada 7.9 Netherland 1.81 Luxembourg 1.77 Australia 7.87 New Zealand 1.80 Netherlands 1.76 Jordan 7.86 Canada 1.77 Canada 1.70 UK 7.83 Luxembourg 1.67 Switzerland

79 Table 2.4: Bottom 10 countries with good governance for year 2012 Freedom Governance Regulation Venezuela 3.88 Comoros Zimbabwe Congo 4.54 Libya Libya Zimbabwe 5.06 CAR * Cuba Algeria 5.14 Afghanistan Syria Argentina 5.15 Zimbabwe Venezuela Togo 5.22 Burundi Iran Iran 5.28 Togo Comoros CAR* 5.30 Guinea Congo Burundi 5.32 Yemen Kiribati Gabon 5.53 Sierra Leone Algeria *CAR: Central African Republic 68

80 Table 2.5: Number of NTM by products for year 2012 Product Code (HS2) Product description Live animals; animal products Meat and edible meat offal Dairy produce, birds egg Products of animal origin Edible vegetables and certain roos Edible fruit and nuts Coffee, tea, mate Cereals Products of the milling industry Oil seeds and oleaginous fruits Lac; gums, resis and other vegetable saps Vegetable plaiting materials Animal or vegetable gats and oils Preparations of meat, of fish Sugars and sugar confectionery Cocoa and cocoa preparations Preparations of cereals, flour, starch Preparations of vegetables, fruit, nuts1.68 Miscellaneous edible reparations Beverages, spirits and vinegar Residues and waste from the food industries Average SPS by product

81 Table 2.6: Top ten countries with the highest and without NTM for year 2012 Top 10 countries imposing NTM (# of NTMs) Countries without NTM Japan(297), France(238), Netherland(237), Italy(235), Germany(235), England(232), Spain(223), Belgium(220), Denmark(215), Bulgaria(208) Afghanistan, United Arab Emirates, Armenia, Antigua and Barbuda, Azerbaijan, Burundi, Benin, Burkina Faso, Bangladesh, Bahrain, Bahamas, Bosnia and Herzegovina, Belarus, Belize, Bolivia, Barbados, Brunei, Bhutan, Botswana, Central African Republic, Switzerland, Cote d lvoire, Cameroon, Congo, Congo, Rep., Cook Islands, Comoros, Cape Verde, Cuba, Djibouti, Dominica, Algeria, Ethiopia, Fiji, Gabon, Georgia, Ghana, Guinea, Gambia, Grenada, Guyana, Iran, Israel, Jamaica, Jordan, Kazakhstan, Kyrgyz Republic, Cambodia, Kiribati, St. Kitts and Nevis, Kuwait, Lebanon, Libya, St. Lucia, Moldova, Madagascar, Maldives, Macedonia, Mali, Montenegro, Mongolia, Mauritania, Mauritius, Malawi, Malaysia, Namibia, Niger, Nigeria, Norway, Pakistan, Papua New Guinea, Paraguay, Rwanda, Senegal, Solomon Is., Sierra Leon, Serbia, Sao Tome and Principe, Suriname, Swaziland, Seychelles, Syria, Togo, Tonga, Trinidad and Tobago, Tunisia, Tuvalu, Tanzania, Uganda, St. Vincent and Grenadines, Venezuela, Vanuatu, Yemen, Zambia 70

82 Table 2.7: Benchmark Gravity and Baseline Results (1) (2) (3) (4) (5) (6) OLS OLS PPML PPML PROBIT PROBIT LN DIST *** *** *** *** *** *** (0.0323) (0.0338) (0.0501) (0.0504) (0.0185) (0.0199) ADJ 0.816*** 0.880*** 0.573*** 0.580*** (0.126) (0.129) (0.138) (0.137) (0.108) (0.114) LANG 0.874*** 0.876*** 0.272** 0.265** 0.370*** 0.395*** (0.0617) (0.0629) (0.121) (0.120) (0.0304) (0.0319) RTA 0.732*** 0.668*** 0.679*** 0.640*** 0.400*** 0.361*** (0.0676) (0.0706) (0.111) (0.119) (0.0468) (0.0488) Ln Freedom *** 1.933*** (0.407) (0.615) (0.261) Governance *** (0.0687) (0.118) (0.0385) Regulation 0.262*** *** (0.0804) (0.146) (0.0427) wntm *** *** *** (0.0483) (0.0282) (0.0443) N Adj R-sq Importer, exporter, and year fixed effects are applied. Robust standard errors clustered at countrypair level are in parentheses. *p <10%, ** p <5%, *** p<1% 71

83 Table 2.8: Gravity estimation: baseline, heterogeneous firm trade model, bias decomposition LN DIST Baseline Heterogeneous firms model Bias Decomposition OLS PPML Probit MLE Poly- 50 bins Sample selection Firm nomial Heckman Heteroge neity (1) (2) (3) (4) (5) (6) (7) (8) *** *** *** *** *** *** *** *** (0.0338) (0.0504) (0.0199) (0.0946) (0.0937) (0.0523) (0.0366) (0.0948) ADJ 0.880*** 0.580*** *** 1.031*** 0.882*** 0.883*** 0.776*** (0.129) (0.137) (0.114) (0.137) (0.121) (0.114) (0.133) (0.133) LANG 0.876*** 0.265** 0.395*** 0.642*** 0.630*** 0.449*** 0.941*** 0.616*** (0.0629) (0.120) (0.0319) (0.0899) (0.0878) (0.0694) (0.0650) (0.0897) RTA 0.668*** 0.640*** 0.361*** 0.428*** 0.410*** 0.288*** 0.715*** 0.461*** IMR (λ ijt ) (0.0706) (0.119) (0.0488) (0.0949) (0.0942) (0.0766) (0.0686) (0.0942) w ijt 0.765*** * 1.822*** 0.408*** (0.173) (0.367) (0.0865) (0.165) z ijt 1.537*** 0.655*** 2 zijt 3 zijt (0.262) (0.163) (0.044) *** (0.003) N Adj R-sq Importer, exporter, and year fixed effects are applied. Robust standard errors clustered at countrypair level are in parentheses. *p <10%, ** p <5%, *** p<1% 72

84 Table 2.9: North and south trade Baseline Heterogeneous firms model Bias Decomposition OLS MLE Sample selection Firm Heckman Heterogeneity (1) (2) (3) (4) Panel A. Bilateral trade value towards OECD member countries Ln dist *** *** *** *** (0.0550) (0.279) (0.0702) (0.280) ADJ 0.566** 1.036** 0.908*** 0.935* (0.223) (0.480) (0.313) (0.477) LANG 0.601*** 0.672*** 0.720*** 0.639*** (0.0990) (0.176) (0.112) (0.176) RTA * * (0.111) (0.160) (0.153) (0.160) IMR (λ ijt ) 0.609* 0.714*** (0.318) (0.135) w ijt (0.302) zijt (0.301) N Adj R-sq Panel B. Bilateral trade value towards non-oecd member countries Ln dist *** *** *** *** (0.0375) (0.117) (0.0458) (0.117) ADJ 0.744*** 0.662*** 0.869*** 0.704*** (0.140) (0.160) (0.151) (0.157) LANG 0.910*** 0.745*** 1.020*** 0.730*** (0.0693) (0.105) (0.0750) (0.104) RTA 1.016*** 0.766*** 1.076*** 0.794*** (0.0873) (0.129) (0.0905) (0.128) IMR (λ ijt ) *** (0.230) (0.123) w ijt 0.785*** (0.221) zijt 0.534** (0.216) N Adj R-sq Importer, exporter, and year fixed effects are applied. Robust standard errors clustered at countrypair level are in parentheses. *p <10%, ** p <5%, *** p<1% 73

85 Chapter 3: Asymmetric Trade Costs in Agricultural Trade among Developing and Industrialized Countries 1. Introduction Trade value of agricultural goods (less than US$ 2 trillion in 2013) is significantly less than that of manufacturing goods (about US$ 13 trillion in 2013). Agricultural trade mostly originates from developed countries. Over 60 percent of trade of vegetable and food products largely flows from developed to developed countries (North - North) or from developed to developing (North - South) (UNCTAD, 2014). The main causes of the low agricultural trade flows from developing countries are considered to be high levels of relative productivity differences and high trade costs (Tombe 2015; Xu 2015). Productivity variations across countries are more significant in the agricultural sector than in the non-agricultural sector (Caselli 2005; Restuccia, Yang, and Zhu 2008; Lagakos and Waugh 2013). Gollin, Lagakos, and Waugh (2013) attribute the relative lower productivity in agricultural sector, the so-called agricultural productivity gap, to the misallocation of labor across sectors. This gap is even greater in developing countries. Lagakos and Waugh (2013) find that self-selection of heterogeneous workers is a major contributor to cross-sector and cross-country productivity differences. They observe that, in developing countries, where a large percent of the workforce is engaged in the agricultural sector, the level of productivity in 74

86 the agricultural sector is lower than that in manufacturing. Conversely, in industrialized countries, they find the opposite relationship to be present. Furthermore, Gollin and Rogerson (2014) and Adamopoulos (2015) suggest that high transport frictions also affect low labor productivity in agriculture and distort labor allocations across sectors. Alleviation of transportation costs is expected to improve agricultural the productivity as well as welfare of the economy. In this essay, the differences in the trade of agricultural goods between developing and industrialized countries are examined using a neo-ricardian trade model. The multicountry model consists of individual countries specializing in continuum goods according to their comparative advantages. Countries demonstrate a variety of productivity levels where productivity is randomly drawn from a country-specific distribution (Eaton and Kortum, 2002; Waugh 2010; Reimer and Li, 2010). Bilateral trade flow in this model is explained by relative unit costs of production, bilateral trade costs, and productivity differences. In this essay, the value of trade elasticity in the agricultural sector is estimated. The estimated low value of trade elasticity reflects more variety in productivity across countries, implying that the degree of comparative advantage is strong against trade barriers. Secondly, asymmetric trade cost is noted as a main cause of bilateral trade share differences between the industrialized (North) and developing (South) countries. In particular, developing countries face relatively higher trade costs to export their products to the North than what industrialized countries pay to export their goods to the South. 75

87 Reimer and Li (2010) investigate the gains from agricultural trade liberalization by estimating the elasticity of trade. They conclude that the gain is not distributed equally because of differences in openness and productivity. Xu (2015) finds the causes of low trade intensity in the agricultural sector, compared to manufacturing trade, to be high trade costs and the large variation in agricultural productivity. However, neither paper addresses systematically asymmetric trade costs between developing and developed countries. In this essay, the systematically asymmetric trade costs, as developed by Waugh (2010), is used to analyze why agricultural products are not traded from South to North to the same degree that they are from North to South. The remainder of the essay is structured as follows. In Section 2, the theoretical model is derived. The data are described in Section 3. The empirical specification, the estimation methodology and the results are presented in Section 4. Finally, in Section 5, the essay is summarized and conclusions are drawn. 2. Model Following Reimer and Li (2012), each country i is assumed to have a tradable agricultural product sector. There is a continuum of agricultural products indexed by j [0,1] (Dornbusch, Fischer, and Samuelson, 1977). Countries differ in their production efficiency z i (j). In terms of producing agricultural goods in country i, land (L i ) with land rental rate ( r i ) is used with productivity z i (j). With a constant return to scale, the cost of production is r / z. i i 76

88 Productivity is assigned by a random draw from a country-specific Frechet probability distribution (Eaton and Kortum, 2002) (henceforth EK). This probabilistic structure allows each country to have some possibility of producing at a lower cost than the others, thus assigning a comparative advantage. Fi ( z) exp{ Ti z i } (1) The location of the distribution is controlled by the parameter T i, implying average productivity in country i. A more productive country has a higher T i. The parameter θ, which is common across countries, governs the distribution of the yield. A lower θ implies great fluctuation in productivity levels across products and countries, indicating that comparative advantage exercises more power than trade costs on trade patterns. Assume that country i is the exporter and country n is the importer. The delivery of one unit of an agricultural good requires ni units produced in country i. Home trade indicates ii 1 when i=n. Assuming that the market is perfectly competitive, the price that country n pays for the imported product j from country i is: p ni niri ( j) (2) z ( j) i The consumer price in n for good j is the lowest price across all trading partners: p ( j) min{ p ( j), p ( j), p ( j),..., p ( j)} n n1 n2 n3 nn A representative consumer has the following constant elasticity of substitution (CES) utility function: 1 U [ q( j) dj] 0 ( 1)/ /( 1) 77

89 where q( j ) indicates the quantity purchased by consumers and σ is the elasticity of substitution across products. The utility maximization is subject to aggregated (across all buyers in country n) budget constraint X n, total spending in country n. The possibility that country i exports a good to country n is the probability that the price of country i will be the lowest. Using the distribution of efficiency (see Equation (1)), EK shows the probability that country i deliver its good at the lowest price in country n is: Pr[ P ( j) P l i] T( r ) i i ni ni nl N i 1 T( r ) i i ni Equation (3) indicates that country i s probability of exporting to country n decreases with the land rental rate and the distance between the trade partners increases with high average yields. Equilibrium 1. Price Index: At the country level, each country n has an aggregated price index. The moment-generating function for the extreme value distribution generates a price index (EK, 2002). (3) where 1 P T r where (4) N n [ ( 1/(1 ) 1/ )] [ i ( i ni ) ] i [ ( )] 1/(1 ) is the Gamma function. The aggregate price index is expressed as a function of the technology level (T i ), the land rental rate ( r i ), and trade costs ( ni ). Equilibrium 2. Trade shares: Denote X ni as total n s the spending on imports from i and Xn as n s total spending. The share of n s expenditure on imported products from i 78

90 among total n s expenditure is equal to the probability that country i exports to n at the minimum price. Therefore, the probability that country i exports to country n at the minimum price can be written with the trade share at the aggregate level. X X T ( r ) ni i i ni N n Ti ( ri ni ) i 1 Equation (5) shows that trade shares consist of productivity parameters (T i ), bilateral trade costs ( ni ), and the land rental rate (r i ). Using the equation above, trade share is normalized by home trade of importers. This is composed of the relative technology level, land rental rate, and trade costs. X ni / X n Ti ri ( ) ( ) ni (6) X / X T r nn n n n (5) Equilibrium 3. Allocation of land resource and land rental rate: Equation (7.1) indicates trade balance requirement. Total exports are equal to total imports. Equation (7.2) shows that the total domestic product equals the sum of country i s export towards all trading partners including itself. The optimal land allocation derived from the first order condition from the producer s problem yields: r L i i ni n 1 l X X Export = Import: in Xni (7.1) i n i n l Y X r L (7.2) i ni i i n 1 79

91 3. Data Balanced trade flow data for products among 128 countries were obtained for the year The total number of observations is 9,709. The countries and descriptive statistics are shown in Table 3.1 and 3.2, respectively. Zero trade flows are revised to 1/ in order not to lose a substantial number of observations. 1 Trade and production data were obtained from the Food and Agriculture Organization of the United Nation (FAO) database. The observed values of mainland China, Macao, Taiwan, and Hong Kong have aggregated as China. The observed value at the country- level is aggregated trade and production values for the agricultural products. The list of observed products is presented in Table 3.3. Trade cost data were obtained from the Centre d Etudes Prospectives et d Information Internationales (CEPII) gravity dataset. The geographic distances between two countries, common border, common language, and common regional trade agreements were used as proxies for trade impediments. Distance variables consist of six dummies, representing the intervals of the circular distance between the capitals. The criteria for dividing the intervals ([0,375);[375,750);[750, 1500);[1500,3000);[3000,6000); and [6000, maximum]) is taken from EK (2002). Arable land data are obtained from the World Bank s World Development Indicators. 4. Empirical Analysis 4.1. Estimation of trade elasticity The value of the elasticity is critical to estimate the effect of trade policies on trade (Simonovska and Waugh, 2014) because it influences the measurement of trade 80

92 frictions, the fluctuation of trade flows, and the welfare effects (see Equation (6)) In this essay, trade elasticity estimates follow the approach used by EK (2002), who suggest using the second highest price difference among trade partners to measure bilateral trade costs with product-level price data. where X ni / X n Pi ni ( ) ( ) X / X P ii i n Pi ni max 2{ln Pn ( j) ln Pi ( j)} ln( ) P 1 J n (ln P ( ) ln ( )) j 1 n j Pi j J (8.1) (8.2) Equation (8.1) indicates that the trade share of country i in country n relative to i s share at home can be expressed with relative prices and trade costs. If the relative price in market i with respect to n falls or the distance between country i and n increases, then country i s normalized share in n declines. In the theoretical model, a lower θ indicates more variation in productivity, reflecting strong comparative advantage (see Equation (1)). As θ becomes small, the left-hand side of the equation, representing normalized import share, is less elastic to the relative price and trade costs ni (EK, 2002). Therefore low trade elasticity (θ), implying greater fluctuation in productivity levels across products and countries, enhances the effect of relative price differences and trade costs on trade share. EK (2002) suggest using the second highest relative price as a proxy for trade costs. By converting Equation (8.1) into logarithmic form and substituting the right-hand variable into Equation (8.2), the value of trade elasticity θ can be recovered by using simple ordinary least squares (OLS) estimation. The product-level price data come 81

93 from the FAO price statistics database for each observed country in the year A simple OLS estimation yields a value of This value is similar to the value used in Reimer and Li (2010): 2.83 and 2.52 from the generalized method of moments (GMM) and maximum likelihood estimation (MLE) techniques, respectively, for crop products among twenty-three countries in the year Originally, EK (2002) used a simple method-of-moments technique for the manufacturing sector using 19 OECD countries in 1990 and suggested 8.28 as the θ value. Simonovska and Waugh (2014) propose a value of 2.79 to 4.46 based on results from the simulated method-of-moments estimations for all sectors with 123 countries in the year The estimates of the latter studies (Simonovska and Waugh 2014; EK 2002) suggest larger value than the estimates of the current essay because the observed sectors are different. The value of θ in this study is determined to be 2.5 because the main observed products are limited to agriculture Estimation of Si Equation (6) shows that a normalized trade share by home trade consists of the relative technology level, land rental rate, and trade costs. Taking logs yields a structural gravity equation: Xni / Xn ln( ) Si Sn ln ni X / X nn n where ln b l RTA d ex ni ni ni ni rni i ni r (9) Following EK (2002), Waugh (2010), and Heerman and Sheldon (2016), trade costs ( ij ) are estimated with a common border (b ni ), common language (l ni ), common 82

94 regional trade agreement (RTA ni ), distance between the two countries ( d r ni ), and exporter fixed effects ( ex ). The distance variable is constructed as in the k th distance i intervals. S i has the same value in the parameter vector S, which is the combination of the state of technology and the land rental rate ( S ln( T r ) assumed to be the sum of the two components: i 1 2 ni ni i i ). The error term ( ni ) is. 2 The first component ( ) ni indicates an unobserved one-way direction (with variance ). The second is a countrypair specific component affecting, the two-way direction, so that (with 2 2 ni in 2 variance 2 ). Accordingly, the error term has a variance-covariance matrix with the diagonal elements of E( ) and the off-diagonal elements of ni in E( ) (EK, 2002). The error term, overall, controls the potential reciprocity ni ni 2 2 in the geographic barriers (Reimer and Li, 2010). 3 Xni / Xn ln( ) Sˆ ˆ ˆ ˆ i Sn ni ni Si Sn ( bni l ni RTAni dr ) ni ni X / X nn n where S Sˆ ex ˆ i i i r (10) The exporter fixed effects ( ex ) measure the additional trade costs for a specific i exporter i, which enables identification of the difference between high export costs and S i. Including the exporter fixed effects in the trade cost equation helps identify the importer and exporter effects separately (Simonovska and Waugh, 2014). As Equation (10) shows, the two separate effects, destination country fixed effects ( S ˆn ) and source country fixed effects ( S i ), are estimated with dummy variables. Since S ˆi is a common 83

95 component for countries that are both exporters and importers, the exporter-specific component of trade costs is recovered as the deviation in the importer and exporter fixed effects ( Sˆ S Sˆ ( Sˆ ex ) ex ). Accordingly, Equation (10) is estimated using i i i i i i generalized least squares (GLS) with the diagonal elements ( ) of the variancecovariance matrix (Eaton and Korum 2003; Reimer and Li 2010; Simonovska and Waugh 2014). To avoid the dummy variable trap, two constraints ( Sn 0, exi 0 ) are imposed (Reimer and Li 2010; Simonovska and Waugh 2014). Table 3.4 shows the estimation results of Equation (10) using 9,709 observations for 128 countries. Most of the coefficients are statistically significant with adjusted R 2 as Panel A indicates the estimated coefficients of the geographic barriers and Panel B presents the estimated S i terms and recovered exporter effects. The coefficients for the geographic barriers imply that the trade share increases in common border, common language, and common regional trade agreements. The coefficients are positive and statistically significant at the 1% level. At the same time, the normalized trade share by home trade decreases in distance between the countries. In detail, the coefficient on the first distance dummy is and this is the smallest in the magnitude of any further distance dummies. The magnitudes of all distance variables in absolute values are larger than that of any other variables, suggesting that the transportation costs are the main impediments to international trade. The estimated destination country effects (S i ) and the exporter effects (θex i ) are reported in Panel B. S i, which is equivalent to ln( Tr ), is interpreted as the adjusted i i

96 average productivity level by unit production cost of country i. In other word, S i is a decreasing function of a unit cost for a producer with the average technology level. The estimated S i implies that the production unit costs with the average productivity level do not significantly vary with GDP per capita, implying that South and North are similar in term of unit production costs, as shown in Figure 3.1 (Waugh, 2010). 4 By using the exporter fixed effects, the model precisely reflect the same level of aggregate price of tradable goods across countries in the data. In the next section, we examine the effects on trade costs and heterogeneous technology in agricultural trade between the North and South Effects on trade costs and state of technology With the estimated θ, Table 3.5 illustrates the implied effects on the trade cost and the state of the technology. The implied effects on cost is estimated by ( 1/ )* b ( e 1) with θ= 2.5. In Panel A, the effects of the geographic barriers on trade share are estimated. While common border, common language, and common regional trade agreement reduce trade costs, the distance variables increase trade costs. The size of the geographic distance influence is much larger for the distance than that of the shared border, shared language, and shared regional trade agreement. A distance of less than 375 miles requires at least an additional units of agricultural goods to be imported. Other geographic barriers (common border, common language, and common regional agreement) reduce trade costs by at least an additional 0.28~0.73 units of traded agricultural goods. 85

97 Agricultural goods exported from the United States (US) are cheaper by an additional 1 unit than products exported from the average country. Similarly, it costs less to export from Argentina, China, Chile, and Brazil than to the average country (about 0.97 units). On the other hand, a product exported from Nigeria costs about units more than the average. Goods exported from Mali, Mongolia, Guinea and Surinam cost more than an additional 50 units than the average country. Therefore, it costs less for the relatively open and industrialized countries to export, as Figure 3.2 shows. As noted in the previous section, the unit costs of a producer with the average productivity level are equivalent among countries (Waugh, 2010). The differences in S i are assumed to be caused by the differences in productivity. The average status of the technology is recovered using the definition of S i : lnt Sˆ ln r i i i where r i indicates land rental rate, which is estimated using the exporter s agricultural output per hectare of arable land (Heerman and Sheldon, 2016). In this way, the country s average technology level (T i ) can be separated from its competitiveness (S i ). As Figure 3.3 shows, more productive countries reveal higher income. The relationship between the log of estimated technology level (T i ) and the log of GDP per capita is positive. The North and South differ in terms of technology level. Table 3.6 shows the normalized technology level by calculating the value relative to the US value, Ti 1/ ( ). The US, T us China, Argentina, Brazil, and Chile are recorded as the top five high-technology countries in the agricultural sector whereas Gambia, Botswana, Benin, Guyana, and 86

98 Zimbabwe are recorded as the bottom five countries. Besides, the normalized technology level is interpreted as the adjusted technology level of a country by land rental rate. For instance, Australia (8.243) is more competitive than France (8.01) and Germany (7.61), but it is ranked below France and Germany. It is assumed that the competitive edge is due to the lower land rental rate than the advanced state of technology. Similarly, low estimate for the competitiveness of Belgium (ranked 24 th ) is the consequence of the high land rental rate (ranked 19 th ) Recovering Asymmetric Trade Costs, ij Using the estimates from the previous section, bilateral trade costs from the structural model are estimated. We use Equation (9) to derive asymmetric trade costs. exp( ˆ / )*exp( ˆ / )*exp( ˆ / )*exp( ˆ / )*exp( ˆ ni bni lni rtani dr ex / ) ni i r Trade costs for selected countries are presented in Table 3.7. The rows indicate exporters and the columns indicate the destination markets. Trade costs to export ( ni ) follow the standard iceberg assumption, in that they refer to transportation costs or costs to overcome geographic barriers. It also includes unobserved related barriers, which are the asymmetric components.. For rich countries, e.g., China, France, Japan, the UK, the US, and so on, trade costs to the South, which are represented in the upper diagonal, are less than the trade costs for the South towards the North markets, as represented in the lower diagonal. For example, the trade cost for the US to Zimbabwe (6) is smaller than that of Zimbabwe to the US market (31,672). In addition, the trade cost of Ethiopia to France is more than twice the 87

99 cost of exporting agricultural products to Ethiopia from France. Accordingly, the asymmetric trade costs imply that the South trading with the North faces relatively more difficulties in exporting its goods than importing goods from the North. Figure 3.4 shows the relationship between τ in and τ ni (where n is trading partner and i is the US). Trade cost from the US towards n country is relatively smaller than that of country n's trade cost towards the US market. Also, developing countries are located in the upper part of the graph. They indicate that they have a relatively higher trade cost than that of the US. Figure 3.5 shows the relationship between asymmetric trade costs and the GDP per capita. Most of the observed countries have the positive deviation of trade costs, meaning their trade costs towards the US market are higher than the US trade costs towards their markets. The relationship between GDP per capita and the deviation is negative. Thus, countries with the higher deviation of trade cost towards the US have a lower GDP per capita. An important conclusion is that the South, lowincome countries, pays relatively higher trade costs than what the US pays for exporting its goods. 5. Conclusion Trade flows in the agricultural sector are significantly less than those involved in manufacturing trade. This essay investigates the degree of relative productivity differences as well as trade costs in the agricultural sector to explain the low trade flows. Based on a Ricardian model, trade shares are expressed with relative productivity, relative land rental rate, and bilateral trade costs. Using the trade data for 88

100 128 countries for the year 2013, the current essay estimates the value of trade elasticity. The estimated value of trade elasticity is relatively low than the value for the manufacturing sector. The low trade elasticity reveals that there is large heterogeneity in productivity in the agricultural trade sector, implying that the role of comparative advantage is strong. Furthermore, large trade frictions restrict agricultural trade flow. In particular, asymmetric trade costs support the low agricultural trade of developing countries in that the South faces relatively higher trade costs than the North. Based on the estimation results, the trade costs levied on the South are much higher than what the North pays while the domestic unit costs and price of tradable goods are equivalent between the North and the South. In conclusion, the relatively higher trade costs, as well as differences in productivity are suggested as the main causes why the South trades fewer agricultural goods in this essay. 89

101 Notes 1. If we drop zero trade flows, the number of observations decreases to 4,928 with 116 countries 2. Simonovska and Waugh (2014) use both specifications with the error term in Equation (9.1) and (9.2) as well as in trade costs; and interpret the error term as a measurement error and structural shock to trade barriers, respectively. According to their results, the estimates are nearly identical. 3. It is possible that the error term related to the trade from n to i is correlated with the disturbance concerning trade from i to n 4. The interpretation of Si is different from EK using importer fixed effects. The model with the importer fixed effect allows for a larger import share as a result of the lower unit cost of production. If two countries import a similar share of goods, then the model considers an increase in trade costs as a cause of a similar trade share. 90

102 Albania Burkina Faso Table 3.1: Observed countries Algeria Burundi Fiji Jordan Ethiopia Japan Netherlands Saudi Arabia USA New Zealand Senegal Uruguay Antigua & C?te Barbuda d'ivoire Finland Kazakhstan Nicaragua Seychelles Vanuatu Argentina Cabo Verde France Kenya Niger Singapore Venezuela Armenia Cambodia Gambia Kyrgyzstan Nigeria Slovakia Viet Nam Australia Cameroon Georgia Latvia Norway Slovenia Yemen Austria Canada Germany Lebanon Oman* South Africa Zambia Azerbaijan Chile Ghana Lithuania Pakistan Spain Zimbabwe Bangladesh China, mainland Greece Luxembourg Panama Sri Lanka Barbados Colombia Guinea Madagascar Paraguay Suriname Belarus Congo Guyana Malawi Peru Sweden Belgium Honduras Malaysia Philippines Switzerland Belize Costa Rica Hungary Maldives Poland Thailand Benin Croatia Iceland Mali Portugal Bolivia FYR Macedoni Cyprus India Malta Togo Czech Indonesia Mauritius South Korea Republic Trinidad Tobago Bosnia &Herzegovina Denmark Iran Mexico Moldova Tunisia Botswana Ecuador Ireland Mongolia Russia Turkey Brazil Egypt Israel Morocco Rwanda Ukraine Brunei El United Italy Namibia Saint Lucia Darussalam Salvador Kingdom Bulgaria Estonia Jamaica Nepal St Vincent & Grenadines Tanzania 91

103 Table 3.2: Summary statistics Variable Obs Mean Std. Min Max Unit Import value ij 9, Million US$ Export value ij 9, Million US$ RTA ij 9, Common border ij 9, Common lang ij 9, Distance ij 9, mile dist1 ij 9, dist2 ij 9, dist3 ij 9, dist4 ij 9, dist5 ij 9, dist6 ij 9, Total imports i 9, Million US$ Total exports i 9, Million Total prod i 9, Xn ij 9, Pini ij 9, Dep ij 9, ln dep ij 9, , US$ Million US$ 92

104 Table 3.3: Observed product Wheat Rapeseed Tangerines, mandarins Mata Barley Sesame seed Lemons and limes Hops Maize Mustard seed Grapefruit Pepper (piper spp.) Rye Poppy seed Apples Chillies and peppers Oats Cottonseed Pears Vanilla Millet Linseed Quinces Cinnamon (canella) Sorghum Oilseeds nes Apricots Nutmeg, mace and cardamoms Buckwheat Cabbages and other brassicas Cherries, sour Anise, badian, fennel, coriander Triticale Artichokes Cherries Ginger Canary seed Asparagus Peaches and nectarines Rubber, natural Grain, mixed Lettuce and chicory Plums and sloes Meat, cattle Potatoes Spinach Strawberries Milk, whole fresh cow Sweet potatoes Tomatoes Gooseberries Meat, sheep Roots and tubers, nes Cauliflowers and broccoli Currants Meat, goat Sugar beet Pumpkins, squash and gourds Blueberries Meat, pig Beans, dry Cucumbers and gherkins Cranberries Meat, chicken Broad beans, horse beans, dry Eggplants (aubergines) Grapes Eggs, hen, in shell Chillies and peppers, Peas, dry green Watermelons Meat, duck Melons, other Chick peas Onions, shallots, green (inc.cantaloupes) Lentils Onions, dry Figs Meat, turkey Mangoes, mangosteens, Cashew nuts, with shell Garlic guavas Meat, horse 93 Meat, goose and guinea fowl Leeks, other alliaceous vegetables Avocados Meat, rabbit Chestnut Walnuts, with shell Beans, green Pineapples Meat, game Pistachios Peas, green Dates Honey, natural Kola nuts Carrots and turnips Persimmons Nuts, nes Maize, green Kiwi fruit Soybeans Mushrooms and truffles Papayas Coconuts Vegetables, fresh nes Fruit, fresh nes Oil, palm Ba as Coffee, green Olives Plantains Cocoa, beans Sunflower seed Oranges Tea

105 Table 3.4: Estimation of Si Panel A Dist1 (-θd1) *** (0.437) Dist2 (-θd2) *** (0.299) Dist3 (-θd3) *** (0.208) Dist4 (-θd4) *** (0.161) Dist5 (-θd5) *** (0.106) Dist6 (-θd6) *** (0.153) Border (-θb) 1.74 *** (0.456) Lang (-θl) *** (0.215) RTA (-θrta) *** (0.225) Panel B Destination country (S n ) Source country (θex i ) Destination country (S n ) Source country (θex i ) Coeff SE Coeff SE Coeff SE Coeff SE Armenia (0.75) (0.52) Lebanon (0.75) (0.65) Albania (0.38) (0.63) Lithuania (0.44) (0.84) Algeria (0.49) (0.58) Madagascar (0.59) (0.84) Antigua and Barbuda (1.08) (0.38) Malawi (0.97) (0.79) Argentina (0.30) (0.52) Malaysia (0.69) (0.58) Australia (0.35) (0.67) Mali (0.66) (0.52) Austria (0.29) (0.52) Malta (0.53) (0.67) Barbados (0.80) (0.55) Mauritius (0.62) (0.56) Bangladesh (0.61) (0.64) Mexico (0.67) (0.68) Bolivia (0.64) (0.79) Mongolia (1.18) (0.48) Botswana (2.24) (0.89) Morocco (0.48) (0.62) Brazil (0.53) (0.65) Moldova (0.51) (0.80) Belize (0.76) (0.44) Namibia (0.71) (0.66) Brunei Darussalam (1.04) (0.64) Nepal (1.02) (0.40) Bulgaria (0.31) (0.55) Netherlands (0.44) (0.45) Continued 94

106 Table 3.4 continued Panel B Destination country (S n ) Source country (θex i ) 95 Destination country (S n ) Source country (θex i ) Coeff SE Coeff SE Coeff SE Coeff SE Burundi (1.34) (0.74) Macedoni (0.51) (0.72) Cameroon (0.82) (0.69) Vanuatu (1.16) (1.61) Canada (0.56) (0.56) New Zealand (0.55) (0.66) Cabo Verde (0.87) (0.49) Nicaragua (0.95) (0.67) Sri Lanka (0.44) (0.81) Niger (1.18) (0.57) Chile (0.35) (0.54) Nigeria (0.96) (0.60) China (0.79) (0.52) Norway (0.46) (0.53) Colombia (0.55) (0.50) Pakistan (0.47) (0.53) Congo (0.92) (0.29) Panama (0.92) (0.51) Czech Costa Rica (0.69) (0.68) Republic (0.35) (0.67) Cyprus (0.45) (0.56) Paraguay (0.68) (0.93) Azerbaijan (0.72) (0.48) Peru (0.62) (0.69) Benin (1.24) (0.59) Philippines (0.44) (0.63) Denmark (0.30) (0.57) Poland (0.33) (0.57) Belarus (0.62) (0.74) Portugal (0.43) (0.56) Ecuador (0.67) (0.63) Zimbabwe (0.92) (0.47) Egypt (0.50) (0.56) Rwanda (0.95) (0.72) Russian El Salvador (1.74) (0.56) Federation (0.73) (0.77) Estonia (0.56) (0.75) Saint Lucia (0.71) (0.57) Saint Fiji (1.04) (0.83) Vincent (1.16) (0.42) Finland (0.47) (0.61) Saudi Arabia (0.60) (0.56) France (0.42) (0.48) Senegal (0.78) (0.73) Georgia (0.55) (0.57) Seychelles (1.22) (0.83) Gambia (1.68) (0.54) Slovenia (0.37) (0.68) Germany (0.39) (0.40) Slovakia (0.49) (0.74) Bosnia and Herzegovina (0.45) (0.79) Singapore (0.51) (0.52) Ghana (0.67) (0.58) South Africa (0.38) (0.61) Greece ( (0.50 Spain (0.58) (0.45) - Guinea (0.65) (0.35) Suriname (1.06) (0.45) Guyana (1.00) (0.40) Sweden (0.35) (0.55) Continued

107 Table 3.4 continued Panel B Destination country (S n ) Source country (θex i ) Destination country (S n ) Source country (θex i ) Coeff SE Coeff SE Coeff SE Coeff SE Honduras (0.75) (0.59) Switzerland (0.41) (0.59) Hungary (0.33) (0.48) Tanzania (0.58) (0.78) Croatia (0.36) (0.69) Thailand (0.44) (0.57) Iceland (0.70) (0.68) Togo (0.95) (0.51) Trinidad and India (0.67) (0.61) Tobago (0.59) (0.56) Indonesia (0.67) (0.62) Tunisia (0.63) (0.67) Iran (0.75) (0.54) Turkey (0.74) (0.53) United Ireland (0.46) (0.64) Kingdom (0.47) (0.46) Israel (0.44) (0.67) Ukraine (0.42) (0.64) Italy (0.39) (0.45) USA (0.76) (0.51) Cote d'ivoire (0.63) (0.56) Burkina Faso (1.05) (0.80) Kazakhstan (0.61) (0.90) Uruguay (0.40) (0.71) Jamaica (0.90) (0.40) Venezuela (0.70) (0.46) Japan (0.83) (0.49) Viet Nam (0.69) (0.64) Jordan (0.51) (0.57) Ethiopia (0.53) (0.70) Kyrgyzstan (1.06) (0.99) Yemen (0.69) (0.38) Kenya (0.49) (0.66) Zambia (1.03) (1.22) Cambodia (0.98) (1.38) Belgium (0.36) (0.46) South Korea (0.54) (0.51) Luxembourg (0.96) (0.80) Latvia (0.55) (0.76) Num of obs 9,709 Num of groups 129 countries F stat R squared Adj R squared Notes: Estimated by generalized least squares. The specification is given in equation (9). Standard errors are in parentheses. * p <10%, ** p <5%, *** p<1% 96

108 Figure 3.1: Destination country effects (S i ) and GDP per capita 97

109 Table 3.5: Estimation of the Effects on Trade Costs Panel A effect on cost θ =2.5 Dist1 (-θd1) *** Dist2 (-θd2) *** Dist3 (-θd3) *** Dist4 (-θd4) *** Dist5 (-θd5) *** Dist6 (-θd6) *** Border (-θb) 1.74 *** Lang (-θl) *** RTA (-θrta) *** Panel B θex i effect on 98 ex i effect on cost cost Coeff θ=2.5 Coeff θ =2.5 Armenia Lebanon Albania Lithuania Algeria Madagascar Antigua and Malawi Barbuda Argentina Malaysia Australia Mali Austria Malta Barbados Mauritius Bangladesh Mexico Bolivia Mongolia Botswana Morocco Brazil Moldova Belize Namibia Brunei Nepal Darussalam Bulgaria Netherlands Burundi Macedoni Cameroon Vanuatu Canada New Zealand Cabo Verde Nicaragua Sri Lanka Niger Chile Nigeria China Norway Colombia Pakistan Congo Panama Costa Rica Czech Republic Cyprus Paraguay Azerbaijan Peru Continued

110 Table 3.5 continued Panel B θex i effect on effect on ex cost i cost Coeff θ=2.5 Coeff θ =2.5 Benin Philippines Denmark Poland Belarus Portugal Ecuador Zimbabwe Egypt Rwanda El Salvador Russian Federation Estonia Saint Lucia Fiji Saint Vincent Finland Saudi Arabia France Senegal Georgia Seychelles Gambia Slovenia Germany Slovakia Bosnia and Herzegovina Singapore Ghana South Africa Greece Spain Guinea Suriname Guyana Sweden Honduras Switzerland Hungary Tanzania Croatia Thailand Iceland Togo India Trinidad and Tobago Indonesia Tunisia Iran Turkey Ireland United Kingdom Israel Ukraine Italy USA Cote d'ivoire Burkina Faso Kazakhstan Uruguay Jamaica Venezuela Japan Viet Nam Jordan Ethiopia Kyrgyzstan Yemen Kenya Zambia Cambodia Belgium South Korea Luxembourg Latvia

111 Figure 3.2: Effects on trade costs and GDP per capita 100

112 Figure 3.3: Productivity (T i ) and GDP per capita 101

113 Table 3.6: Estimation of Productivity Country Ti ( ) T us 1/ Country Ti ( ) T United States of America Switzerland China, mainland Slovakia Argentina Albania Brazil Jordan Chile Nigeria Canada Malawi France Croatia Germany Zambia Australia Estonia New Zealand Singapore Netherlands Nicaragua Spain Cote d'ivoire Italy Latvia South Africa Cambodia India Venezuela Peru Cameroon United Kingdom Jamaica Belgium Ghana Turkey Senegal Ukraine Rwanda Uruguay Namibia Thailand Panama Russian Federation El Salvador Egypt Tunisia Mexico Bosnia and Herzegovina Hungary Fiji Denmark Yemen Paraguay United Republic of Tanzania Continued us 1/ 102

114 Table 3.6 continued Country Ti ( ) T us 1/ Country Ti ( ) T Israel Seychelles Ethiopia Trinidad and Tobago Sri Lanka Burundi Viet Nam Nepal Poland Vanuatu Austria Belize Malaysia Mongolia Japan Georgia Bulgaria Mauritius Indonesia Armenia Republic of Korea Saint Vincent and the Grenadines Greece Azerbaijan Portugal Iceland Kyrgyzstan Suriname Costa Rica Cabo Verde Philippines Antigua and Barbuda Kenya Luxembourg Iran Saint Lucia Madagascar Barbados Republic of Moldova Mali Macedonia Burkina Faso Ireland Norway Czech Republic Brunei Darussalam Ecuador Congo Colombia Belarus Kazakhstan Algeria Bolivia Togo Honduras Niger Lithuania Malta Slovenia Guinea Pakistan Zimbabwe Morocco Guyana Cyprus Benin Continued us 1/ 103

115 Table 3.6 continued Country Ti ( ) T us 1/ Country Ti ( ) T Sweden Botswana Saudi Arabia Gambia Bangladesh Finland Lebanon us 1/ 104

116 105 Table 3.7: Asymmetric Trade Costs for selected countries Ex\Im Arge n-tina China Franc e Guyana Italy Japan Nicar a-gua Peru Morocco Zimbabw e South Africa Spain UK USA Argentina China France Guyana Italy Japan Morocco Nicaragua Peru Zimbabw 3483 e South Africa Spain UK US Viet Nam Ethiopia Viet- Nam Ethiopia 105

117 Figure 3.4: Asymmetric Trade Costs 106

118 Figure 3.5: Asymmetric Trade Costs and GDP per capita 107