The Case for Free Trade Agreements: Historical Perspectives and a Projection for China, Japan, and Korea. May Edward Zhu 1

Transcription

1 The Case for Free Trade Agreements: Historical Perspectives and a Projection for China, Japan, and Korea May 2013 Edward Zhu 1 Department of Economics Stanford University Stanford, CA edzhu@stanford.edu Written under the direction of Professor Michael J. Boskin ABSTRACT In recent years, several major Free Trade Agreements (FTAs) have been proposed amongst countries that encompass a lion s share of the world s economic output. Thus, understanding the impact of FTAs on trade and income has become more crucial than ever. This paper sheds light on both the past and future effects of FTAs via a two-step approach. First, I use a gravity model to measure the effect of existing FTAs on the bilateral trade flows of member countries. Second, I project the potential increase in trade and GDP from a China-Japan-Korea (CJK) FTA specifically, through a combination of gravity model analysis and calculations based on various economic growth and tariff reduction scenarios. The results show that historically, FTAs have greatly increased trade amongst member countries. Looking forward, I estimate that a CJK FTA increases intraregional merchandise trade by 21%-46% and services trade by 49%-79%, with commensurately large increases in GDP. This suggests that FTAs might play a major role in boosting world economic growth. Keywords: FTA, gravity model, trade liberalization, tariff, non-tariff barrier 1 I am indebted to Michael Boskin for his guidance, feedback, and mentorship. I am also grateful to Kenneth Arrow, Kyle Bagwell, Gary Hufbauer, and Dean DeRosa for expediting my search for existing research and data. This work is dedicated to my loved ones.

2 Zhu 1 TABLE OF CONTENTS I. Introduction 1. Context 2. Existing Literature II. Preliminaries: The Gravity Model 1. Intuition 2. Model Description and Calibration 3. Theoretical Foundations of the Gravity Model 4. Data III. Gravity Model Analysis 1. Comprehensive FTA Effects 2. Time Series Cross Sectional Analysis 3. Gravity Inferred Trade Increases for a China-Japan-Korea FTA IV. Trade Projections: China, Japan, and Korea 1. Scenario 1: Baseline 2022 Projection 2. Scenario 2: RMB Appreciation and Yen Depreciation 3. Scenario 3: Tariff Elimination 4. Scenario 4: Accounting for Non-Tariff Barriers V. Piecing it Together 1. Estimated Trade Increases for a China-Japan-Korea FTA 2. GDP Gains VI. Conclusion VII. References VIII. Appendix A. CJK Trade Projections: Assumptions and Calculations B. Accompanying Tables

3 Zhu 2 I. INTRODUCTION 1. Context Governments around the world have actively engaged in trade liberalization for several decades. Since the General Agreement on Tariffs and Trade (GATT) was first signed in 1947, average worldwide tariffs on manufactured goods have roughly fallen from 40% to 4%. 2 These tariff reductions have been a major driver of reduced trade costs, which in turn have driven increasing international trade and integration of the world economy. The chart below shows, for recent years, decreasing import tariffs alongside trade as an increasing percentage of worldwide GDP. Worldwide Imports and Import Tariffs Sources: The World Bank Database 2 Import tariffs (applied, weighted mean, all products). The World Bank Database.

4 Zhu 3 Trade liberalization usually occurs in one of three forms: 1. Multilateral negotiations: global trade negotiations involving most nations. This pertains to the GATT and World Trade Organization (WTO), as explained below. Free Trade Agreements (FTAs): In general, a FTA eliminates tariffs, import quotas, and preferences on most (if not all) goods and services traded between member countries. By eliminating tariffs, member countries of an FTA are exempt from the WTO s Most Favored Nation (MFN) rule on tariffs. 3 o 2. Regional FTAs: multilateral agreements that are generally signed among neighboring countries. An example is the North American Free Trade Agreement (NAFTA) amongst Canada, the US, and Mexico. o 3. Bilateral FTAs: FTAs between two sides, where each side could be a country, a trade bloc, or an informal group of countries. Until recently, trade liberalization has primarily been conducted on a multilateral basis. 23 countries initiated the GATT in 1947 with the purpose of substantial reduction of tariffs and other trade barriers and the elimination of preferences, on a reciprocal and mutually advantageous basis. From 1947 to 1995, GATT membership increased to 123 countries and GATT member countries completed eight rounds of tariff and non-tariff barrier reductions, culminating in the Uruguay Round of The Uruguay Round was the biggest negotiating mandate on trade ever agreed upon: member countries extended talks beyond goods into 3 A country that has been accorded MFN status may not be treated less advantageously than any other country with MFN status by the promising country. However, FTAs are allowed as an exception if tariffs are reduced near zero.

5 Zhu 4 services and the sensitive areas of intellectual property, agriculture, and textiles. Furthermore, the World Trade Organization (WTO) was established to replace the GATT and to expand its scope. Since 1995, the WTO has further expanded to 159 member countries. But despite this history of successful multilateral negotiations, the latest round (Doha Round) of trade talks has stalled due to significant differences in objectives between developed nations and developing countries, particularly in agricultural trade. Nearly 12 years since the Doha round was initiated, attempts to restart negotiations have been largely unsuccessful. In the absence of multilateral negotiations, countries have turned towards free trade agreements (FTAs) as an alternative means of further trade liberalization. In recent years, several major FTAs have been proposed. The US is actively pursuing a FTA with the European Union known as the Transatlantic Trade and Investment Partnership (TTIP). The US is also negotiating the terms for a Trans- Pacific Partnership (TPP), a FTA including Australia, Brunei, Chile, Canada, Japan, Malaysia, Mexico, New Zealand, Peru, Singapore, the US, and Vietnam. Notably, neither China nor South Korea is actively involved in TPP negotiations, though the US has welcomed South Korea to join negotiations. Instead, they are active in negotiations for a third major FTA that has been proposed between China, Japan, and South Korea. Unlike multilateral negotiations, these FTAs contain only a handful of countries and involve already integrated regional economies. This bodes well for their political feasibility.

6 Zhu 5 Despite their rising popularity, however, FTAs are not a new phenomenon. As such, the purpose of this paper is twofold: First, I estimate the effect of existing FTAs on the bilateral trade flows of member countries (Section III). Second, I project the potential increases in trade and GDP growth from a China-Japan-Korea (CJK) FTA (Section V). This is done through a combination of historical analysis (Section III) and forward-looking calculations based on various economic growth and tariff scenarios (Section IV). I chose to investigate the CJK FTA due to the potentially large benefits from increased economic integration in the region. Despite the fact China, Japan, and Korea recorded 19.6% of the world s total GDP in 2010, substantial economic integration has not yet evolved in the region. Furthermore, since the CJK FTA shares many commonalities with past FTAs both geographically and in its stage of economic integration, it is a convenient candidate for analysis. I find that existing FTAs greatly increase bilateral trade flows between member countries, with increases in trade ranging from 42% (the EU) to 309% (NAFTA) for major FTAs. I also estimate that a CJK FTA will increase intraregional merchandise trade by 21%-46% and services trade by 49%-79%. This is a large increase by any standard and supports the case for further trade liberalization. However, the reader should keep in mind that I project the effects of a CJK FTA based on the current state of international trade. If the Doha Round negotiations were to be completed beforehand, we might expect the CJK FTA to have a smaller effect due to redundancies in tariff and non-tariff barrier reduction. If the TPP is ratified as it currently stands, China and Korea might be disadvantaged in

7 Zhu 6 trading with TPP member countries and trade more with one another. In this case, we might expect a larger increase in trade from a CJK FTA. However, if Korea follows Japan s lead and joins the TPP, China might be isolated and a CJK FTA might not be formed at all. The likely scenario will become more apparent as negotiations progress. 2. Existing Literature Nearly all economists think international trade should be free (Rose, 2004). 4 This consensus in economic theory has been a driving force behind trade liberalization in recent decades. However, economists have differing opinions on how large past and future gains from trade liberalization might be as measured by increases in trade and GDP. To fully understand how trade liberalization has been interpreted in existing literature, it is useful to understand the literature surrounding the gravity model. The gravity equation is one of the most empirically successful equations in economics (Anderson and van Wincoop, 2003). It relates bilateral trade flows to GDP, distance, and other factors that affect trade barriers. The gravity model has been widely used to infer trade flow effects of a variety of institutions including the GATT/WTO (Rose, 2004; Subramanian and Wei, 2003), Free Trade Agreements (Freund and Ornelas, 2010; Baier and Bergstrand, 2006), bilateral country pairs (Eichengreen and Irwin, 1998; Khan, 2009), exchange rate mechanisms, and international borders (Anderson and van Wincoop, 2003). It has also been used to 4 J. R. Kearl et al. (1979) shows that 97% of economists surveyed in 1976 agreed (generally or with provisions) that tariffs and import quotas reduce general economic welfare.

8 Zhu 7 forecast the growth of future trade, as Hufbauer and DeRosa (2013) have done for US-Chinese bilateral trade. Similarly, Khan (2009) used the gravity model to estimate the potential gains from increased economic integration between India and Pakistan. In Section II, I explain the gravity model in greater detail (II.1 and II.2) and discuss the model s theoretical foundations (II.3). Through gravity model analysis and other econometric techniques, many researchers have found that trade liberalization has led to large trade increases in a variety of scenarios. The paper most relevant to my historical analysis of FTA effects on trade flows is Baier and Bergstrand (2006). By running a variant of the gravity model on panel data of trade flows for 96 countries (the model also incorporates the average treatment effects of FTAs on trade flows), the authors find than an FTA increases two members bilateral trade by 98% on average after 10 years. Baier and Bergstrand s estimates of FTA effects greatly exceed those of the widely cited Bergstrand (1985) and Frankel, Stein, and Wei (1995), which the authors claim are understated because endogeneity is not properly addressed. Similarly, Anderson and van Wincoop (2003) apply the theory of the gravity equation both to estimation and to the general-equilibrium comparative statics of borders. They conclude that national borders reduce trade between industrialized countries by 20%-50%, a finding that has positive ramifications for future FTAs among neighboring countries. Research also suggests that in several country-specific cases, there are sizable benefits from further trade liberalization. Through gravity model estimates and a variety of qualitative arguments, Khan (2009) determines that trade between Pakistan and India should be 5 to 10 times larger than the present $2 billion per

9 Zhu 8 year. It is no surprise that Pakistan-India trade is low due to high political tension between the two countries. Empirically, the mere fact that Pakistan accounted for 1% of India s trade in compared to 63% of Indian trade in suggests that increased trade ties could benefit both nations greatly, improving both their economies and their national security. On the opposite side of the world, the TTIP might increase US-European Union bilateral trade and raise incomes in both regions a free economic boost as both regions are recovering from the 2008 recession. In a recent publication for the European Commission, Francois (2013) estimates via a computable general equilibrium (GCE) model that given the likely scale of trade liberalization 5 under the TTIP, EU exports to the US might increase by 16%-28% while US exports to the EU might increase by 23%-37%. Since average tariff levels between the two regions are already low, this is a surprisingly large increase in trade that demonstrates the potentially large impact of non-tariff barrier reduction. Unlike for regional and bilateral trade, however, research on the effects of multilateral trade liberalization is mixed. In his widely cited and controversial paper, Rose (2002, 2004) finds that once standard gravity model effects have been taken into account, past increases in bilateral trade cannot be dependably linked to membership in the WTO or its predecessor the GATT. However, after slight variations to Rose s data set and methodology, Subramanian and Wei (2003) find that the GATT/WTO has indeed promoted trade, albeit unevenly. According to their calculations, successive rounds of multilateral tariff liberalization through the 5 Less ambitious FTA scenario: 98% of tariffs and 10%-25% of NTBs eliminated. More ambitious FTA scenario: 100% of tariffs and 25%-50% of NTBs eliminated.

10 Zhu 9 GATT/WTO have increased industrial country imports by about 68%, though the effect on developing country imports is low and not significant. Though there is debate about the effectiveness of past multilateral negotiations, the overall consensus amongst trade economists is that regional and bilateral trade liberalization has led to sizable trade increases. Freund and Ornelas (2010) add a proviso, however, that the preferential removal of tariffs in a FTA may lead to trade diversion in which imports shift away from the most efficient supplier to the country receiving preferential treatment. This leads to inefficient world production, which harms countries outside the FTA. It can also hurt members, if the change in consumer prices (and therefore in consumer surplus), is too small to outweigh the costs from the inefficiency. In contrast, if the FTA leads to greater imports from the efficient suppliers within the bloc, consumer gains outweigh the costs from production inefficiency and the agreement necessarily improves members welfare. Thus the degree of trade diversion, in addition to the increase in intra-fta trade, determines the degree to which an FTA is successful. Freund and Ornelas go on to state that FTAs enhance welfare on balance, according to empirical evidence. Magee (2008) further supports this view. By modifying the gravity model (he includes country-pair fixed effects, exporter-year fixed effects, and importeryear fixed effects) to capture the counterfactual more accurately than standard gravity specifications, Magee finds that trade creation dominates trade diversion by approximately one order of magnitude on average. His findings bolster the existing research on FTAs, which suggest sizable increases in trade and GDP resulting from trade liberalization at the regional level.

11 Zhu 10 II. PRELIMINARIES: THE GRAVITY MODEL In recent years, the gravity model has become a workhorse for quantitative studies of international trade. This section explains gravity model analysis, which I rely on to estimate the past effect of FTAs on bilateral trade flows (Section III) and to forecast the growth of China-Japan-Korea bilateral trade to 2022 (Section IV). 1. Intuition The gravity model of trade is analogous to Newton s gravitational model. In its most basic form, the volume of trade ( F ij ) between country i and country j is proportional to the product of GDP i and GDP j and inversely related to the distance between the two countries ( D ij ). Mathematically, T ij G(GDP i i GDP 2 j /D 3 ij ) ij, where G is a constant and represents an error term with expectation equal to 1. To reduce the gravity model to its standard form, we take logs of both sides. ln(t ij ) 0 1 ln(gdp i GDP j ) 2 ln(d ij ) ij Here, the constants G and are transformed to 0 and ij, respectively. To construct the full gravity model, I introduce additional explanatory variables corresponding either to natural causes of trade or to specific FTAs. An example of a natural cause is the language variable Lang ij we expect countries with a common language to trade relatively more. An example of a FTA specific variable is a NAFTA binary variable NAFTA ijt.

12 Zhu 11 Since a group of countries would only form a FTA if they have complementary trade structures, we expect these countries to have higher levels of trade a priori. My empirical strategy is to control for as many natural causes of trade as possible and to search for the effects of FTAs in the residual. Along this line, I augment the gravity equation with numerous natural cause variables including culture (e.g., whether a pair of countries share a common language), geography (e.g., whether two countries share a common border), and history (e.g., whether one country colonized the other). If there were nothing to the notion of FTAs, then the natural cause variables would soak up all the explanatory power. There would be little left to attribute to the indicator variables representing whether two trading partners belong to the same FTA. 2. Model Description and Calibration The exact specification of the gravity model used in later sections is: ln(t ijt ) 0 1 ln(d ij ) 2 ln(y i Y j ) t 3 ln(y i Y j /Pop i Pop j ) t 4 Lang ij 5 Cont ij 6 Landl ij 7 Island ij 8 ln(area i Area j ) 9 ComCol ij 10 CurCol ijt 11 Colony ij t t t 1 EU ijt 2 EFTA ijt 3 NAFTA ijt 4 MER ijt 5 CMAS ijt 6 AFTA ijt 7 SAARC ijt 8 FTA ijt ijt where i and j denote the exporting and importing nation respectively, t denotes time, and the variables 6 are defined as: T ijt denotes the average value of real bilateral trade between i and j at time t, D is the distance 7 between i and j, Y is real GDP, 8 6 I adopt the notation from Rose (2004). 7 Measured as the great circle distance between two country capitals. 8 Y can either refer to market or PPP values. Clarification in Section III.1.

13 Zhu 12 Pop is population, Lang is a binary indicator variable equal to 1 if i and j have a common language and equal to 0 otherwise, Cont is a binary variable equal to 1 if i and j share a land border, Landl is the number of landlocked countries in the country-pair (0, 1, or 2), Island is the number of island nations in the country pair (0, 1, or 2), Area is the country s area in square kilometers, ComCol is a binary variable equal to 1 if i and j were ever colonies after 1945 with the same colonizer, CurCol is a binary variable equal to 1 if i is a colony of j at time t or vice versa, Colony is a binary variable equal to 1 if i ever colonized j or vice versa, { t } is a comprehensive set of time fixed effects and is a vector of extraneous coefficients, The FTA binary variables 9 with corresponding coefficients 1 to 8 are equal to 1 if both i and j are members of the FTA at time t. ij represents the omitted other influences on bilateral trade, assumed to be well behaved. The parameters that interest me are 1 to 8. For instance, if trade beyond natural trade growth is created when both countries are in NAFTA, then 3 should be positive. Although the remaining coefficients ( 1 to 11 ) do not directly influence my estimation of the past effect of FTAS (Section III), they come into play when making projections (Section IV). Furthermore, comparing the natural cause coefficients to intuition and existing research provides validation for my formulation of the gravity model. 9 My gravity model incorporates indicator variables for over 500 FTAs, grouped into eight prominent individual FTAs and groups of FTAs worldwide. The grouping is as follows: European Union (EU); European Free Trade Area (EFTA); North American Free Trade Agreement (NAFTA); Southern Cone Common Market (Mercosur); Chile, Mexico, Australia, and Singapore (CMAS) FTAs; ASEAN Free Trade Area (AFTA); South Asia Preferential Trading Agreement (SAARC); and all other FTAs (FTA).

14 Zhu 13 In subsequent sections, I estimate the gravity model using Two Stage Least Squares (TSLS), computing standard errors that are robust to clustering by countrypairs. TSLS addresses the issue of simultaneous causality between trade flows ( T ijt ) and GDP-related determinants of trade ( ln(y i Y j ) t and ln(y i Y j /Pop i Pop j ) t ). A bias potentially exists because higher joint GDP implies more trade and trade is also a component of GDP. Thus, 2 and 3 may be downwardly biased in an OLS estimation of the gravity model. To address this issue, I employ population ((Pop i Pop j ) t ) and lagged GDP variables ((Y i Y j ) t 1 and (Y i Y j /Pop i Pop j ) t 1 ) as instruments. The population and lagged GDP variables are uncorrelated with at time t, are clearly correlated with the endogenous variables, and are arguably very strong instruments. 10 Analogous to Rose (2004), I also include a comprehensive set of year-specific fixed effects to account for factors such as the value of the dollar, the global business cycle, and commodity price shocks. This specification for the gravity model is the basis for my analysis in subsequent sections, though I vary the setup and application on a case-by-case basis. 3. Theoretical Foundations of the Gravity Model The gravity model has been widely acclaimed in three aspects: (1) its empirical success at predicting trade flows, (2) its recently improved theoretical foundations arising mostly from modern theories of trade in imperfect substitutes, and (3) its ability to treat countries or regions as physically placed at particular locations rather than as disembodied constructs (DeRosa, 2010). 10 For column (1) of Table 3.1, the instruments (Y i Y j ) t 1, (Y i Y j /Pop i Pop j ) t 1, and (Pop i Pop j ) t have first-stage t-statistics of 30.52, 45.24, and 8.74, respectively.

15 Zhu 14 In the theoretical sphere, most international trade economists acknowledge that bilateral trade volumes adhere to a structure resembling a gravity equation. Economists have derived the gravity model from a variety of leading trade theories. For example, Deardorff discovered how to derive the gravity model from Heckscher- Ohlin theory; the standard derivation of the gravity model is from the theory of imperfect substitutes (Deardorff, 1998). In a separate instance, Eaton and Kortum used the gravity model to develop and quantify a Ricardian model of international trade (one based on differences in technology) that also incorporates a role for geography (Eaton and Kortum, 2002). There is no shortage of theoretical foundations for the gravity model. Empirically, the gravity model has long been recognized for its consistent success in explaining trade flows (Bergstrand, 1985). The model s estimated coefficients for natural causes of trade are sensible, economically and statistically significant, and reasonably consistent across studies (Rose, 2004). Furthermore, the gravity model fits the data well in that it explains most of the variation in international trade (Anderson and van Wincoop, 2003) Data My analysis is based on the gravity model data set compiled by the Peterson Institute for International Economics (PIIE) on behalf of Professor Boskin, for both merchandise trade ( ) and trade in services ( ). The dependent variable in most of my analysis is aggregate trade (exports plus imports), 11 Anderson and van Wincoop (2003) discuss the gravity model in greater depth.

16 Zhu 15 in log form, although I distinguish between exports and imports when making trade projections. For both merchandise trade and trade in services, each observation T ij is indexed by exporting country i and importing country j. The data on merchandise and services trade are compiled from the UN Comtrade data set and the OECD database on international trade in services, respectively. Given that import data reported by national authorities are generally more reliable and accurate that export data, the T ij observations are compiled from import data by reporting countries with all individually recorded partners. Where observations are missing, however, export data by reporting countries are used to fill out the data set. The explanatory variables included in the new gravity model data set are drawn from three primary sources. First, the observations on time invariant natural causes variables (such as distance between trading partners, common border, common language, etc.) are drawn from the data set developed by Rose (2004). Second, observations on eight major individual and groups of FTAs are drawn from existing PIIE data sets. Finally, World Bank World Development Indicators data are used to specify GDP and GDP per capita series for all countries. The data set includes trade by 218 countries; 12 a list of the countries is included in Table B1. For 36 years ( ) of merchandise trade data with exports and imports separately identified (x2), we would expect 36 2 C(218, 2) 1,703,016 observations. The data set includes 667,576 observations, with most of the missing values from and from the 12 Not all the trading entities are countries, though I use the word for convenience.

17 Zhu 16 smallest countries. For services, I only use 11 years of data ( ) for my later analysis since data on international trade in services is less complete prior to Trade between non-member countries is excluded, since services trade data is only available for the 20 OECD member countries. Thus, we expect 11 2 (C(218, 2) C(198, 2)) 182,600 observations for services trade. There are 63,519 observations, with most missing values from the smallest countries. In my later analysis, I exclude trade flows below a $10 million cutoff (in 2011 prices) for both merchandise and services trade to prevent insignificant trade flows from throwing off the gravity model. Since most of the missing data points are from trade with small countries and would normally be excluded by this criterion, selection bias in terms of data availability is not a problem given the scope and completeness of the data set for larger economies. Table 2.1 summarizes the data for trade and GDP above various cutoffs for both merchandise and services trade. Cutoff Table 2.1 Summary Statistics for Trade Flows Above Various Cutoffs (Millions of US Dollars at 2011 Prices) Trade Statistics Observations Q1 Median Q3 Mean Mean GDP (Billions) Merchandise Trade ( ) None 667, > $10M 236, ,082 1,037 > $25M 181, ,409 1,182 >$50M 142, ,784 1,312 Services Trade ( ) None 63, > $5Mk 32, ,158 > $10M 28, ,242 > $25M 22, ,388

18 Zhu 17 III. GRAVITY MODEL ANALYSIS This section estimates the effect of existing FTAs on bilateral trade flows. Using gravity model analysis, I demonstrate that two countries in the same FTA have significantly higher trade than two countries otherwise. I then use Time Series Cross Sectional analysis to determine whether the gains in trade (observed by the gravity model) are attributable to the FTAs and are not attributable solely to the tendency for trade to be intraregionally concentrated. Using these gravity model results, I then infer the potential increase in trade from a China-Japan-Korea FTA. The analysis in this section is centered on merchandise and not services trade for two reasons. First, data on services trade is limited prior to Second, interest in services is a recent phenomenon. Historically, barriers to merchandise trade have been the salient issue. However, services will play a larger role in my trade projections since they are expected to be a big source of future growth in trade. 1. Comprehensive FTA Effects To isolate the effects of various FTAs on trade, I estimate the gravity model via Two Stage Least Squares (TSLS), as specified in Section II.2. In my panel regression, I use aggregate merchandise trade data for all countries from , excluding trade flows under $10 million (in 2011 prices) to prevent insignificant trade flows from throwing off the model. By bringing the information from all years in my dataset ( ) to bear at once, we can conduct meaningful tests for smaller FTAs such as NAFTA, which has

19 Zhu 18 only three observations in a given year. However, as explained in Section II.2, pooling observations from different years without controlling for time fixed effects would distort the estimates. Thus, I introduce a comprehensive set of time fixed effects { t } to absorb the effects of global inflation, growth, and other year-specific events. The results are shown in Table 3.1 in five different specifications. The regression in column 1 is my benchmark result. In columns 1 and 2, the GDP explanatory variables are measured by Purchasing Power Parity (PPP) exchange rates. In columns 3 and 4, GDP is measured by market exchange rates. I prefer PPP exchange rates because market exchange rates tend to underestimate the purchasing power of consumers in emerging market and developing countries. For this reason, the Mercosur (Latin America) coefficient is highly significant in columns 1 and 2 even though columns 3 and 4 fail to reveal this relationship. PPP exchange rates are also more stable than market exchange rates over time. Of course, the main drawback of PPP is that PPP exchange rates must be estimated, which can introduce inaccuracies into the measurement. Despite this tradeoff, I use GDP measured by PPP exchange rates 13 for the majority of my analysis and projections. For the regressions in columns 2 and 4, I omit the GDP per capita explanatory variable and drop the corresponding instrumental variable (lagged GDP per capita) from the TSLS. In column 5, I add country-specific fixed effects to the benchmark equation in column 1. These variations in the gravity model have minimal effects on the key results. 13 The PIIE GDP PPP data is taken from World Bank Development Indicators.

20 Zhu 19 Table 3.1 TSLS Gravity Model Estimates for Aggregate Merchandise Trade, PPP Exchange Rates Market Exchange Rates With Country Effects (1) (2) (3) (4) (5) Log Product Real GDP 0.80*** (0.04) 0.82*** (0.02) Log Product Real GDP P/C -0.07** (0.03) Log Product Real GDP PPP 0.89*** (0.04) 0.97*** (0.02) 0.54*** (0.05) Log Product Real GDP PPP P/C 0.08** (0.03) 0.47*** (0.06) Log Distance -0.63*** (0.02) -0.65*** (0.02) -0.67*** (0.02) -0.70*** (0.02) -0.72*** (0.02) Common Language 0.46*** (0.04) 0.49*** (0.04) 0.29*** (0.03) 0.31*** (0.03) 0.38*** (0.04) Land Border 0.64*** (0.06) 0.62*** (0.06) 0.58*** (0.06) 0.59*** (0.06) 0.32*** (0.05) Number Landlocked -0.18*** (0.02) -0.15*** (0.02) -0.12*** (0.02) -0.09*** (0.02) -0.82*** (0.15) Number Islands 0.34*** (0.03) 0.37*** (0.03) 0.24*** (0.03) 0.19*** (0.03) 0.12*** (0.03) Log Product Land Area -0.13*** (0.02) -0.17*** (0.01) -0.07*** (0.02) -0.07*** (0.01) 0.28*** (0.03) Common Colonizer 0.81*** (0.06) 0.80*** (0.07) 0.89*** (0.06) 1.05*** (0.07) 0.94*** (0.07) Currently Colonized 0.84*** (0.18) 0.68*** (0.17) 0.93*** (0.18) 0.82*** (0.18) 0.67*** (0.21) Ever Colony 0.60*** (0.07) 0.59*** (0.08) 0.65*** (0.07) 0.65*** (0.07) 0.62*** (0.08) European Union 0.35*** (0.05) 0.33*** (0.07) 0.28*** (0.05) 0.02 (0.06) 0.27*** (0.05) European Free Trade Area 0.89*** (0.10) 1.00*** (0.09) 0.32*** (0.09) 0.10 (0.08) 0.75*** (0.08) NAFTA 1.41*** (0.15) 1.39*** (0.18) 1.37*** (0.14) 1.37*** (0.14) 1.16*** (0.11) Mercosur 0.83*** (0.19) 0.91*** (0.20) 0.29 (0.21) 0.29 (0.21) 0.65*** (0.16) CMAS FTAs 0.50*** (0.17) 0.54*** (0.18) 0.48*** (0.16) 0.48*** (0.16) 0.38*** (0.13) AFTA 1.07*** (0.15) 0.94*** (0.14) 1.22*** (0.17) 1.22*** (0.17) 0.86*** (0.13) SAARC -1.10*** (0.34) -1.35*** (0.35) -1.03*** (0.33) -1.03*** (0.33) -0.46*** (0.14) Other FTAs 0.57*** (0.06) 0.63*** (0.06) 0.53*** (0.06) 0.53*** (0.06) 0.48*** (0.05) Observations R^2 RMSE 166, , , , , Notes: *p<0.10, **p<0.05, ***p<0.01. Two stage least squares with robust standard errors (clustering by country-pairs) in parentheses. The constant term is omitted.

21 Zhu 20 Shifting our focus to Table 3.1, we see the model works very well. Countries that are farther apart trade less and richer countries trade more. Physically larger countries trade less (since they have more internal trade), as do landlocked countries (since trade is cheaper by sea than by land). A common language and shared colonial history both encourage trade. These effects match our intuition, are highly statistically significant, are in line with estimates from the literature 14, and explain almost two-thirds of the variation in bilateral trade. In this sense, the gravity model effectively explains the relation between natural causes of trade and observed trade flows across over 150,000 observations. In almost all cases, the benchmark gravity model also indicates that membership in the same FTA has a substantial positive effect on trade. For instance, membership in Mercosur is estimated to raise trade over by 129%! (Since the regressand is the natural logarithm of real trade, we can calculate the gains as e %.) For seven of my eight FTA binary variables, the gravity model suggests increases in trade ranging from 42% (European Union) to 309% (NAFTA). The sole exception is the South Asia Preferential Trading Agreement (SAARC), which has a highly significant negative coefficient of Upon closer examination, the seemingly counterintuitive reduction in trade for SAARC member countries exists because India and Pakistan the two largest SAARC economies with a history of 14 Comparing the coefficients in Column 1 with Rose s 2004 benchmark results, Col 1: Log Distance [-0.63]; Log Product GDP [0.89]; Common Language [0.46]; et al. Rose: Log Distance [-1.12]; Log Product GDP [0.92]; Common Language [0.31]; et al. (Transportation costs were higher for the time period of Rose s data, 1948 to 1999, so his coefficient on Log Distance [-1.12] is larger than my coefficient [-0.63].)

22 Zhu 21 conflict trade much less than we would expect given the size of their economies and their proximity. 15 Though my gravity model analysis points to large gains in trade from FTAs, it s important to keep two clarifications in mind. First, I make no argument as to the cause of increased trade for country pairs within the same FTA. Is higher trade due to trade creation and economic growth? Or is higher trade due to trade diversion, with intra-fta trade increasing as existing trade shifts away from countries outside of the FTA? Tariff reductions bring both forces into play. This concern would need to be addressed on a FTA-by-FTA basis, and is outside the scope of my analysis. Second, are the increases in trade from FTAs attributable to changes in the trading system? Or does trade tend to be intraregionally concentrated, which is in turn misidentified by the gravity model as increases in trade from the FTA (an extreme form of the Krugman-Summers claim)? I address this concern in the next section. 2. Time Series Cross Sectional Analysis (TSCS) To determine whether the increases in trade predicted in the previous section are attributable to the FTAs, I modify the FTA binary variables from Section II.2 to be country binary variables. That is, if country i is the US and country j is Canada, NAFTA ijt now equals 1 regardless of the time t. (In the previous section, NAFTA ijt equaled 1 if and only if t 1994, since NAFTA came into force on 15 Fortunately for my analysis, the unique case for SAARC demonstrates that the data can yield negative results. Thus for the other FTA coefficients, we see the data (rather than the methodology) are delivering the positive message!

23 Zhu 22 European Union NAFTA Nations Mercosur Nations CMAS Nations AFTA Nations SAARC Nations China, Japan, and Korea Table 3.2 Estimated Preferential Trade Agreement Coefficients, by Five-Year Intervals Mer Svc Mer Svc Mer Svc Mer Svc Mer Svc Mer Svc Mer Svc 0.16 (0.12) 0.89*** (0.29) 0.69*** (0.25) 2.33*** (0.59) 0.84*** (0.25) (0.46) 0.71** (0.28) Other FTAs 0.95*** (0.19) Observations R-squared RMSE 3, ** (0.12) 1.08*** (0.39) 0.66*** (0.17) 2.52*** (0.30) 1.01*** (0.22) -0.86* (0.47) 0.80** (0.37) 1.02*** (0.22) 3, *** (0.11) 0.60** (0.23) 0.93*** (0.17) 2.86*** (0.28) 1.08*** (0.20) -0.80** (0.39) 0.45* (0.26) 1.23*** (0.17) 4, (0.57) 2.43*** (0.53) 0.49 (0.33) *** (0.08) 1.21*** (0.20) 1.32*** (0.16) 0.61 (0.48) 1.43*** (0.15) -0.74* (0.42) 0.84*** (0.26) 0.73*** (0.09) 6, (0.27) -0.74** (0.28) 2.35*** (0.73) (0.25) 0.33*** (0.08) 1.52*** (0.14) 0.82*** (0.27) 0.65*** (0.11) 1.39*** (0.15) -1.24*** (0.47) 0.76*** (0.29) 0.64*** (0.09) (0.12) (0.19) 2.20*** (0.59) (0.54) (0.15) 0.23*** (0.06) 1.28*** (0.18) 0.65*** (0.21) 0.57*** (0.13) 1.17*** (0.17) -0.91** (0.40) 0.78** (0.32) 0.84*** (0.08) 0.37*** (0.08) 0.48** (0.21) 0.92** (0.31) 0.12 (0.44) 1.07*** (0.20) 0.26*** (0.05) 1.15*** (0.24) 0.74** (0.37) 0.63*** (0.14) 1.02*** (0.18) -1.12*** (0.35) Notes: *p<0.10, **p<0.05, ***p<0.01. Darkly shaded cells indicate the FTA formally exists (e.g. NAFTA for years 1994 and after). Two stage least squares on both merchandise (Mer) and services (Svc) trade using the same specification as the benchmark regression (Column 1) in Table 3.1. All coefficients other than the FTA-coefficients are omitted. Robust standard errors (clustering by country-pairs) are in parentheses. No coefficient estimate is reported when there is insufficient variation in the explanatory variable , , , , ** (0.29) 0.59*** (0.07) 7, *** (0.08) 0.51** (0.24) (0.46) -0.92** (0.37) 0.25* (0.14) 4,

24 Zhu 23 January 1, 1994.) Next, I estimate the gravity model at five-year intervals using the same specification as column 1 of Table 3.1. There is enough data in the cross section dimension that we can estimate each year separately. The results are in Table 3.2. Consider a row in Table 3.2, which represents a particular group of countries. A cell in that row is darkly shaded if an FTA exists for the corresponding year. For instance, NAFTA Nations 1995 is darkly shaded whereas NAFTA Nations 1990 is lightly shaded, since NAFTA was formed in Read each row from left to right and consider the changes in the coefficients at the transitions between lightly and darkly shaded cells. Table 3.3 provides context for all four FTA transitions. NAFTA Nations Mercosur Nations AFTA Nations SAARC Nations Table 3.3 Gravity Model: Inferred Immediate Trade Gains from FTAs Before FTA [Year] After FTA [Year] Change in Coefficient (Est. SE) 1 Predicted Gain in Trade Effective Year of FTA 0.60 [1990] 0.93 [1990] 1.08 [1990] [1980] 1.21 [1995] 1.32 [1995] 1.43 [1995] [1985] +0.61*** (0.21) +0.39** (0.16) +0.35** (0.17) (0.46) 84% % % % 1985 Notes: *p<0.10, **p<0.05, ***p< For each region, I estimate the Standard Error as an average of the corresponding Before and After Standard Errors in Table 3.2. If the null hypothesis that FTAs don t affect trade is true, this approximation is also valid. 16 Since e %. The regressand is the natural logarithm of real trade.

25 Zhu 24 If the observed tendency for trade to be intraregionally concentrated were due solely to the proximity of the countries, we would expect insignificant changes in the gravity estimated coefficients around the years the FTA became effective. Empirically, the opposite is true. In years immediately following FTA establishment, the positive boosts to the estimated coefficients are significant at the 5% level for all three reliable FTA transitions: NAFTA, Mercosur, and AFTA. We have already established that SAARC is an outlier due to the history of sour political and trade relations between India and Pakistan. This supports my initial intuition that the gains in trade observed by the gravity model in Section III.1 are at least partially attributable to the FTAs themselves. 3. Gravity Inferred Trade Increases for a China-Japan-Korea FTA Given that FTAs boost trade, both in the long run (Section III.1) and in years immediately following their establishment (Section III.2), what does the gravity model imply about potential trade increases from a China-Japan-Korea (CJK) FTA? I approach this question in two steps: 1. I estimate the increase in trade from existing FTAs. 2. I estimate the extent to which CJK trade 17 is already developed. Intuitively, if intra-regional trade within CJK is already high, the potential for intra-regional trade growth from a CJK FTA is lower. Next, I discount the second estimate from the first (subtraction since the regressand is the natural log of real trade) to calculate the potential trade increase. 17 CJK trade refers to bilateral trade flows amongst China, Japan, and Korea, but not including other trading partners.

26 Zhu 25 To estimate the increase in trade from existing FTAs, I refer to the benchmark gravity model (column 1) in Table 3.1. In my analysis, I exclude SAARC for reasons mentioned earlier. I also exclude CMAS, since the FTA is too geographically different from CJK to be a meaningful comparison. 18 Of the remaining FTA binary variables, the European Union has the lowest coefficient at 0.35 and NAFTA has the highest coefficient at The mean and median coefficients are 0.91 and 0.89, respectively. Using the median as a proxy for the average effect, two countries in the same FTA have 144% higher trade than two countries otherwise. 19 For the second step, I rely on my estimates from Table 3.2. Note that in the 2010 regression, the coefficient on the CJK country binary variable is 0.62 and significant at the 5% level. I use 0.62 as my estimate for the present stage of development of CJK trade, suggesting that the three countries already have 86% higher trade than we would expect considering their GDP, relative distances, and other natural determinants of trade. This estimate is consistent with the empirical evidence on intra-regional trade. A 2011 joint study by China, Japan, and Korea stated that intra-regional trade patterns have recently evolved towards greater convergence and complementarity in the trade structure of the three nations, though there is still significant potential for a CJK FTA to further boost intraregional trade. For instance, although CJK intra-regional trade is already modestly developed at 21.7%, it still has significant room to develop compared to NAFTA, a mature FTA that has 40.0% intra-regional trade. 18 Distance between countries (Chile, Mexico, Australia, and Singapore) is too large. 19 Since e %. The regressand is the natural logarithm of real trade.

27 Zhu 26 This matches my gravity model inference. Combining the two steps, the difference between coefficients in the median case is Thus, if the CJK FTA has above average success relative to existing FTAs, we can expect trade increases exceeding 31%. In the best case, a CJK FTA could increase trade to the same extent as NAFTA. In this scenario, the difference between coefficients is 0.79 and CJK trade increases by 120%. 21 Such a large increase in trade is not unreasonable, since if CJK trade increased 120% we would expect intra-regional trade to rise from 21.7% to just over 40%, which is the value of NAFTA intra-regional trade. 22 In this sense, the actual trade data corroborates the model prediction. Though my methodology centered on past data does not present complications for my Step 2 estimate of the stage of CJK trade development, any assessment of trade creation potential from a CJK FTA makes assumptions about the future, either implicit or explicit. In the case of Step 1, I implicitly assumed that China, Japan, and Korea have comparable trade potential to countries within existing FTAs. In Section IV, I make more explicit assumptions about CJK natural GDP and population growth, tariff reductions, and non-tariff barrier reductions and examine the effect on intra-regional trade. In Section V.1, I combine my inferences from this section, my Section IV calculations, and additional context to provide more comprehensive estimates of trade increases from a CJK FTA. 20 Median Case: ; e % 21 Best Case: ; e % 22 Joint Study Report for an FTA among China, Japan, and Korea. December 16, 2011.

28 Zhu 27 IV. TRADE PROJECTIONS: CHINA, JAPAN, AND KOREA In this section, I conduct a 10-year-ahead forecast (to 2022) of Chinese, Japanese, and Korean (CJK) bilateral trade growth under four scenarios. 23 Scenario 1 estimates trade growth using the gravity model, under standard assumptions of GDP and population growth. Using partial equilibrium analyses, Scenarios 2-4 augment the Medium GDP growth case from Scenario 1 and estimate changes in trade when accounting for changes in foreign exchange rates, import tariffs, and non-tariff barriers (NTBs). Merchandise and services trade are distinguished, and exports and imports are separately identified. Before I describe and report the four scenarios, the reader should keep in mind that partial equilibrium analyses only consider the effect of tariff reductions ceteris paribus. Thus, this methodology tends to underestimate increases in trade from a new FTA because it does not project and account for changes in world trade patterns. This is in contrast with the gravity model projection from Section III.3, which implicitly interprets all causal factors imbedded in the trade data, including historical changes in world trade patterns. To develop a holistic view of trade increases from a China-Japan-Korea FTA, it is important to refer to both the gravity model projection (which leaves the econometrician to identify the causal factors for increased trade) and partial equilibrium analysis (which is more precise due to the simplifying assumptions in each scenario but is a weaker model of real world economic phenomena). 23 CJK trade refers to bilateral trade flows amongst China, Japan, and Korea, but not including other trading partners.

29 Zhu 28 Table 4.2 summarizes the projections for CJK bilateral trade under Scenarios 1-4. Tables 4.3 and 4.4 show the projected increases in trade for the four scenarios in annualized and absolute terms, respectively. Due to the large number of manual calculations involved in making the projection, I limit my discussion in the following sections to description of the methodology and interpretation of the results. For the more curious reader, I refer to and present subsets of my calculations in Appendix A. To view the assumptions for all four scenarios at once, please refer to Table A18 of the Appendix. Scenario 1: Baseline 2022 Projection Methodology I use the gravity model to estimate CJK trade in Within the gravity model framework, GDP and population levels are important determinants of trade flows. 24 My GDP and population growth assumptions are based on data from the World Bank and Blue Chip Economic Indicators, respectively. Table 4.1 Growth Assumptions, (Percent per annum) Real GDP Population Growth Scenario China Japan Korea China Japan Korea Low Medium High Low Japan Sources: World Bank (GDP), Blue Chip Economic Indicators (Population) 24 As demonstrated in Table 3.1 and Section III.1.

30 Zhu 29 Table 4.2 China-Japan-Korea Trade, and Projections to 2022 China Japan Korea Trade with Japan Trade with Korea Trade with China Trade with Korea Trade with China Trade with Japan Year and Scenario Goods Service Goods Service Goods Service Goods Service Goods Service Goods Service Billions of US Dollars at Current Prices Billions of US Dollars at 2011 Prices Gravity model-based projections Scenario Low Growth High Growth Low Japan Growth Scenario Scenario Scenario 4 1 Sub Comp Full Notes: Export data ( ) from PIIE. The assumptions used to project Scenarios 1-4 are listed in Table A18 of the Appendix. 1 Sub, Comp, and Full correspond to the Substitutable, Complementary, and Full Accommodation cases.

31 Zhu 30 Table 4.3 Trade Growth Rate Scenarios for China-Japan-Korea, China Japan Korea Trade with Japan Trade with Korea Trade with China Trade with Korea Trade with China Trade with Japan Scenarios Goods Service Goods Service Goods Service Goods Service 2 Goods Service Goods Service Compounded Annual Growth Rates (Percent) Actual Trade, Predicted Trade, Scenario Low Growth High Growth Low Japan Growth Scenario Scenario Scenario 4 3 Sub Comp Full Notes: Export Data ( ) from PIIE. Trade growth rate calculations for service exports are based on time periods for actual trade and for predicted trade. 1 Over 11 years, this corresponds to trade expansion of ( ) times. 2 For this column, I use for actual trade and for predicted trade to compensate for the 2010 Japan-Korea services export outlier. 3 Sub, Comp, and Full correspond to the Substitutable, Complementary, and Full Accommodation cases.