This paper investigates the trade creation effects of Japan's free trade agreements (FTAs) using aggregate trade data for the years 1996–2015. We estimate various specifications of a gravity model. Our main finding is that the effects of Japan's FTAs are not clearly observed when the gravity model is specified with three types of fixed effects (i.e., exporter-year fixed effects, importer-year fixed effects, and country-pair fixed effects). In fact, the effects of FTAs vary substantially among trade partners and around half of the FTAs increase Japan's trade values. Our results also suggest that FTAs with small trade partners tend to have large effects on Japan as well as other countries. Recently enforced FTAs, however, increase Japan's import values more rapidly.

Free trade agreements (FTAs) are currently the dominant form of commercial policy. The pattern of trade policies in the last two decades has been characterized mainly by the proliferation of FTAs. According to the Regional Trade Agreements Information System in the World Trade Organization (WTO), the cumulative number of physical regional trade agreements increased from 24 in 1992 to 286 in 2017.1

Japan started to seek trade liberalization through FTAs around 2000 and established the first FTA with Singapore in 2001. Since then, 15 FTAs with 17 countries have been established by August 2018.2 In addition, Japan recently signed two important multilateral trade agreements, the Comprehensive and Progressive Agreement for Trans-Pacific Partnership and the economic partnership agreement with the EU. Although East Asian countries including Japan began forming FTA networks only recently, they are catching up with Western countries and are expected to play a vital role against protectionism.

This paper investigates the trade creation effects of Japan's FTAs. We focus on Japan for three reasons. First, despite increasing interest among policymakers, the ex post evaluation of FTAs is extremely limited. No papers apply recently developed methods of analysis to Japan's FTAs. The gravity model, which is commonly used in ex post studies, is developing rapidly in the academic literature and is now estimated differently from the way it was ten years ago. Furthermore, some recent studies show that different FTAs have very different effects, and trade creation effects are not clearly observed in some cases. Even under FTAs, Japan has not eliminated trade barriers for agricultural goods, which are heavily protected by most favored nation (MFN) tariff rates. In addition, MFN tariff rates on manufacturing goods are already low and there is little scope to reduce them further through FTAs. It is therefore unclear whether Japan's FTAs have increased its own trade values. Trade creation is crucial for demonstrating the positive welfare effects of FTAs; if trade creation does not occur, a reconsideration of Japan's trade and commercial policies would be called for.

The second reason is that Japan provides a suitable case study for exploring the heterogeneous effects of FTAs in the context of the surge in regionalism and the proliferation of bilateral trade agreements since the 1990s. European countries trade mainly among themselves and they established trade agreements with each other soon after World War II. The United States also established trade agreements with its major trading partners, Canada and Mexico, around 1990 by signing the Canada–U.S. Free Trade Agreement and the North American Free Trade Agreement (NAFTA). In contrast, Japan, the world's fourth largest exporter and third largest economy, began negotiations of bilateral trade agreements with the countries of ASEAN after 2000.

Finally, Japan's FTA partners vary substantially in terms of economic development, ranging from Cambodia, Laos, and Myanmar to Australia, Singapore, and Switzerland. The differences in the trade creation effects across different FTAs can be attributed to the characteristics of the partner countries.

The objective of this paper is to evaluate Japan's FTAs. To this end, we first estimate the effects of Japan's and other countries’ FTAs using a state-of-the-art gravity model.3 We include three types of fixed effects and estimate using Poisson pseudo maximum likelihood (PPML).4 We estimate the gravity model using various specifications to compare the coefficients for the FTA dummies. In addition, the effects of each of the FTAs are separately identified to take heterogeneity into account. Then the estimated crude effects of individual FTAs are regressed on some variables to explore which FTAs have larger effects. This paper therefore contributes to the literature by examining the determinants of successful FTAs.

The structure of the paper is as follows. We briefly review the literature on the impacts of FTAs in the next section. In Section 3, we describe the source of the data and provide a descriptive analysis. Section 4 discusses the econometric methodology for estimating the gravity model. Results of the estimation are presented in Section 5, followed by conclusions in Section 6.

FTAs are major instruments for promoting international trade in the 21st century. The ex post effects of FTAs are usually estimated using a gravity model; Baier and Bergstrand (2007) recommend the use of panel data to remove all time-invariant bilateral factors not controlled for in the traditional specification. A similar specification is applied in Magee (2008), although the effects of FTAs weaken when the gravity model is estimated with fixed effects. Cipollina and Salvatici (2010) conduct a meta-analysis and robustly reject the hypothesis that FTAs have no effects. Large effects are also confirmed in Eicher, Henn, and Papageorgiou (2012).

Although the effects of FTAs have been established in many papers, Kohl (2014) notes that the trade creation effects are heterogeneous and only about one-quarter of agreements are actually trade-promoting.5 Heterogeneity is also studied in Baier, Yotov, and Zylkin (2019) and Baier, Bergstrand, and Clance (2018). Zylkin (2016) examines the heterogeneous effects of FTAs, using the case of NAFTA. The differences between these papers and ours are three-fold. First, we focus on the heterogeneity of directional effects rather than agreement-specific effects or pair-specific effects. This is because Japan's FTAs are bilateral except for the agreements with ASEAN (ASEAN–Japan Comprehensive Economic Partnership, AJCEP), initially applied in December 2008 between Japan and four countries in ASEAN. Second, we compare the coefficients estimated in various specifications to consider what is important for the evaluation. This analysis provides a good benchmark for future studies because our state-of-the-art specification is computationally burdensome. Finally, our sample period, 1996–2015, includes recent agreements. Because Kohl (2014) finds that FTAs signed after 1990 have smaller effects, extending the sample period is not a trivial point.

Some papers examine FTAs in Japan or East Asia.6 Ando and Urata (2011) investigate the impact of the Japan–Mexico economic partnership agreement and find large trade-creation effects for some products. Ando and Urata (2015) conduct a similar analysis for three of Japan's FTAs with Malaysia, Thailand, and Indonesia. Yamanouchi (2017) examines the effects of Japan's FTAs by using the Trade Statistics of Japan published by the Ministry of Finance and Japan Customs. Although the effects of individual FTAs are estimated in these papers, only Japan's trade data are used. The trade values of Japan's FTA partners with third countries are not considered. Vietnam, for example, is undertaking rapid liberalization, including WTO accession in 2007. It is therefore useful to separately identify the effects of Japan's FTAs with Vietnam from the effects of Vietnam's unilateral trade liberalization. Okabe (2015) explores FTAs formed by ASEAN countries and their trade partners, so AJCEP is studied in that paper. She concludes that the impact of AJCEP is unclear. Furthermore, while the effects of each of Japan's FTAs are estimated, a country-pair dummy is not included in the specification.

In this paper, we use world trade data to estimate the effects of Japan's FTAs. We estimate various specifications of a gravity model and place emphasis on the importance of estimating the effects of FTAs using the correct specification. In addition, we discuss the characteristics of the partners with which FTAs are working well.

The trade data used in this paper are obtained from UN Comtrade. The sample period extends from 1996 to 2015. We first construct a large data set of 156 countries to interpolate missing trade values. Then the sample to estimate the gravity model is limited to 69 countries.

We include a country in the data set if its import data are available for more than 11 years during the period 1996–2015. All of Japan's FTA partners are then added regardless of data availability. Many countries have some missing import data. We interpolate the missing import values using the export data reported by exporters. Because import values are reported as cost, insurance, and freight, and export values are reported as free on board, the gap must be estimated.7 We regress import values on a quartic of export values, a quartic of bilateral distance (population weighted), a quartic of log of importer GDP, a quartic of log of importer population, other gravity variables (a contingency dummy, a common colonizer dummy, and a common language dummy), variables related to trade policies (a FTA dummy, a customs union dummy, a partial scope agreement dummy, a common currency dummy, an importer EU dummy, and an importer WTO dummy), and an exporter-year fixed effect. We obtained these variables, other than some trade policy variables, from the CEPII Web site, constructed by Head, Mayer, and Ries (2010) and Head and Mayer (2014).8 The information on the FTA, customs union, and partial scope agreement dummies is obtained from the Mario Larch Regional Trade Agreements Database from Egger and Larch (2008).9

The actual and predicted values have a high correlation coefficient of 0.92, suggesting that the estimation is valid and the missing import values are well approximated by the corresponding export values.

The sample we use in the main analysis is smaller because of computational difficulty. We then select the countries by ranking trade values averaged over 20 years. The sample includes Japan's FTA partners and the countries with a ranking of export or import values higher than 60. Our sample includes 69 countries.10

Table 1 shows the evolution of Japan's exports to all countries and FTA partners. The trade flows under FTAs are shaded. Japan's exports to all countries were around US$ 410 billion in 1996. The total export value has increased rapidly over the period 2003–08. Although Japan's total exports collapsed during the 2008–09 global financial crisis, they soon recovered before decreasing slightly. From Table 1, it is difficult to identify the effects of Japan's FTAs because most of them were enforced during the expansion period, although the export values increased after the enforcement. In addition, although export values to developing countries such as Myanmar and Vietnam grew most rapidly, we must account for the impacts of the deepening integration of the world economy.

Table 1.
Evolution of Japan's export values to FTA partners
Export value in 1996 = 100
FTA partnerExport 1996 (million US dollars)1997199819992000200120022003200420052006200720082009201020112012201320142015
Singapore 23,842 97 71 77 97 67 61 64 80 80 83 90 108 78 102 110 99 85 84 77 
Mexico 3,837 116 118 132 168 210 243 197 275 340 398 425 424 297 391 429 460 445 457 452 
Malaysia 19,225 90 59 70 89 73 74 74 88 86 90 98 101 80 107 111 104 93 87 71 
Chile 950 111 104 66 73 57 56 74 104 133 154 208 338 167 357 308 273 260 248 221 
Thailand 20,445 78 49 59 74 67 72 89 109 127 126 146 163 122 185 206 242 200 174 152 
Indonesia 8,504 97 50 34 63 55 51 49 71 81 64 76 177 115 199 228 267 226 199 155 
Brunei 139 169 72 61 68 51 147 89 100 124 154 127 242 192 176 201 192 150 104 171 
Laos 49 80 54 72 70 45 62 54 58 78 90 149 263 297 67 89 147 226 185 148 
Myanmar 106 90 84 93 273 367 171 213 191 186 221 359 514 549 206 377 835 1605 1544 1447 
Vietnam 1,208 113 120 152 190 180 207 246 294 337 389 512 682 618 746 861 960 957 1064 1174 
Philippines 7,578 104 84 86 91 92 104 109 106 111 101 95 93 76 96 92 91 73 73 89 
Switzerland 2,112 100 100 108 110 95 83 100 116 111 117 138 183 156 167 221 238 190 188 166 
Cambodia 81 104 86 109 72 75 78 96 102 123 159 173 140 146 192 305 294 214 324 520 
India 2,186 97 112 116 101 81 97 106 135 168 216 266 356 305 378 513 565 479 455 440 
Peru 440 109 121 108 110 97 93 83 81 101 128 178 290 210 311 298 341 326 251 244 
Australia 7,859 106 106 110 119 106 115 141 163 175 174 202 229 171 223 236 250 228 196 188 
World (156 countries) 410,174 102 94 101 117 103 106 119 142 151 163 180 198 149 188 206 206 189 182 164 
Export value in 1996 = 100
FTA partnerExport 1996 (million US dollars)1997199819992000200120022003200420052006200720082009201020112012201320142015
Singapore 23,842 97 71 77 97 67 61 64 80 80 83 90 108 78 102 110 99 85 84 77 
Mexico 3,837 116 118 132 168 210 243 197 275 340 398 425 424 297 391 429 460 445 457 452 
Malaysia 19,225 90 59 70 89 73 74 74 88 86 90 98 101 80 107 111 104 93 87 71 
Chile 950 111 104 66 73 57 56 74 104 133 154 208 338 167 357 308 273 260 248 221 
Thailand 20,445 78 49 59 74 67 72 89 109 127 126 146 163 122 185 206 242 200 174 152 
Indonesia 8,504 97 50 34 63 55 51 49 71 81 64 76 177 115 199 228 267 226 199 155 
Brunei 139 169 72 61 68 51 147 89 100 124 154 127 242 192 176 201 192 150 104 171 
Laos 49 80 54 72 70 45 62 54 58 78 90 149 263 297 67 89 147 226 185 148 
Myanmar 106 90 84 93 273 367 171 213 191 186 221 359 514 549 206 377 835 1605 1544 1447 
Vietnam 1,208 113 120 152 190 180 207 246 294 337 389 512 682 618 746 861 960 957 1064 1174 
Philippines 7,578 104 84 86 91 92 104 109 106 111 101 95 93 76 96 92 91 73 73 89 
Switzerland 2,112 100 100 108 110 95 83 100 116 111 117 138 183 156 167 221 238 190 188 166 
Cambodia 81 104 86 109 72 75 78 96 102 123 159 173 140 146 192 305 294 214 324 520 
India 2,186 97 112 116 101 81 97 106 135 168 216 266 356 305 378 513 565 479 455 440 
Peru 440 109 121 108 110 97 93 83 81 101 128 178 290 210 311 298 341 326 251 244 
Australia 7,859 106 106 110 119 106 115 141 163 175 174 202 229 171 223 236 250 228 196 188 
World (156 countries) 410,174 102 94 101 117 103 106 119 142 151 163 180 198 149 188 206 206 189 182 164 

Source: UN Comtrade. Trade values are basically reported by importers, and missing values are interpolated from corresponding export values.

The evolution of Japan's import values is like that of its export values. As shown in Table 2, total imports almost doubled over the period 2003–08. After the collapse in 2009, import values recovered quickly but then decreased. As for each of the FTA partners, import values have grown more rapidly than export values. For example, imports from Cambodia have increased by 147 times in the last two decades.

Table 2.
Evolution of Japan's import values to FTA partners
Import value in 1996 = 100
FTA partnerImport 1996 (million US dollars)1997199819992000200120022003200420052006200720082009201020112012201320142015
Singapore 7,323 80 64 74 87 73 68 74 85 91 102 96 107 83 111 118 119 101 107 107 
Mexico 1,890 85 65 87 126 106 95 94 114 134 149 167 201 148 184 210 233 223 226 251 
Malaysia 11,750 96 73 92 123 109 95 107 120 124 131 148 197 142 193 259 279 253 248 182 
Chile 2,763 107 86 91 102 88 77 95 151 185 262 295 286 192 280 355 337 290 294 217 
Thailand 10,212 93 80 86 103 101 102 116 138 152 165 179 203 156 205 240 231 215 212 200 
Indonesia 15,194 96 71 82 107 97 93 108 123 137 158 174 214 143 186 224 212 190 168 130 
Brunei 1,392 100 74 75 118 121 109 131 135 164 167 179 326 239 295 409 430 340 288 168 
Laos 23 90 85 58 51 29 28 31 34 34 52 51 77 115 161 415 527 458 491 416 
Myanmar 103 96 87 98 116 99 106 135 174 198 239 287 306 331 375 573 653 738 837 840 
Vietnam 2,018 108 86 97 130 129 125 153 191 225 262 303 450 344 405 572 747 705 764 750 
Philippines 4,522 110 98 117 159 141 144 155 182 170 176 193 186 141 175 197 206 204 225 196 
Switzerland 3,563 96 84 94 92 92 92 108 135 141 143 146 180 176 190 220 230 204 202 207 
Cambodia 200 245 526 795 1005 1141 1362 1519 1606 1831 2117 1843 2172 3165 4691 6158 8882 11761 14755 
India 2,843 93 76 79 92 78 73 76 91 112 142 146 184 131 200 240 246 248 245 171 
Peru 427 127 67 68 82 99 100 101 159 164 309 523 495 389 510 548 656 619 411 290 
Australia 14,229 102 91 90 104 101 98 105 136 172 196 219 334 244 316 398 396 358 338 244 
World (156 countries) 327,328 98 81 90 110 102 98 112 133 151 169 182 223 162 203 251 261 245 239 182 
Import value in 1996 = 100
FTA partnerImport 1996 (million US dollars)1997199819992000200120022003200420052006200720082009201020112012201320142015
Singapore 7,323 80 64 74 87 73 68 74 85 91 102 96 107 83 111 118 119 101 107 107 
Mexico 1,890 85 65 87 126 106 95 94 114 134 149 167 201 148 184 210 233 223 226 251 
Malaysia 11,750 96 73 92 123 109 95 107 120 124 131 148 197 142 193 259 279 253 248 182 
Chile 2,763 107 86 91 102 88 77 95 151 185 262 295 286 192 280 355 337 290 294 217 
Thailand 10,212 93 80 86 103 101 102 116 138 152 165 179 203 156 205 240 231 215 212 200 
Indonesia 15,194 96 71 82 107 97 93 108 123 137 158 174 214 143 186 224 212 190 168 130 
Brunei 1,392 100 74 75 118 121 109 131 135 164 167 179 326 239 295 409 430 340 288 168 
Laos 23 90 85 58 51 29 28 31 34 34 52 51 77 115 161 415 527 458 491 416 
Myanmar 103 96 87 98 116 99 106 135 174 198 239 287 306 331 375 573 653 738 837 840 
Vietnam 2,018 108 86 97 130 129 125 153 191 225 262 303 450 344 405 572 747 705 764 750 
Philippines 4,522 110 98 117 159 141 144 155 182 170 176 193 186 141 175 197 206 204 225 196 
Switzerland 3,563 96 84 94 92 92 92 108 135 141 143 146 180 176 190 220 230 204 202 207 
Cambodia 200 245 526 795 1005 1141 1362 1519 1606 1831 2117 1843 2172 3165 4691 6158 8882 11761 14755 
India 2,843 93 76 79 92 78 73 76 91 112 142 146 184 131 200 240 246 248 245 171 
Peru 427 127 67 68 82 99 100 101 159 164 309 523 495 389 510 548 656 619 411 290 
Australia 14,229 102 91 90 104 101 98 105 136 172 196 219 334 244 316 398 396 358 338 244 
World (156 countries) 327,328 98 81 90 110 102 98 112 133 151 169 182 223 162 203 251 261 245 239 182 

Source: UN Comtrade. Trade values are basically reported by importers, and missing values are interpolated from corresponding export values.

Overall, we cannot conclude from this simple analysis that FTAs have had significant effects on Japan's international trade, although trade with some countries increased rapidly after an FTA was enforced. Instead, we turn to explore the contribution of FTAs using a correctly specified gravity model with three types of fixed effects.

In this paper, we use the standard gravity framework established by Anderson and van Wincoop (2003), in which the bilateral trade value from country i to country j, pijxij, takes the form
(1)
where Yi, Ej, Y, tij, Pj, Πi, and σ denote the total sales of country i, the total expenditure in country j, the sum of the sales all over the world, bilateral iceberg trade costs from country i to country j, an inward multilateral resistance term, an outward multilateral resistance term, and the elasticity of substitution, respectively.
We estimate the following equation, based on Baier and Bergstrand (2007) and Yotov et al. (2016), corresponding to equation (1):
(2)
where Tradeijt is the aggregate trade value from country i to country j at year t. FTAijtW is a dummy variable and equal to 1 if the trading countries are included together in an FTA and both are not Japan. FTAijtJX and FTAijtJM are FTA dummies and equal to 1 if the trading countries are included together in an FTA and the exporter or importer is Japan. The FTA dummies are separated by partners in the estimating equation when we focus on the heterogeneity of FTAs across partners. We interpret positive coefficients for these FTA dummies as evidence of trade creation.11CUijt is equal to 1 if both countries are included in the same customs union. Similarly, PSAijt is a partial scope agreement dummy and CommonCurrencyijt is a common currency dummy. δijB is a country-pair fixed effect and reflects all time-invariant factors that affect the bilateral trade values, such as distance, language, and the historical relationship between two countries. δitX is an exporter-year fixed effect and reflects the production capacity of the exporter, outward multilateral resistance, and unilateral trade policies such as WTO accession. Finally, δjtM is an importer-year fixed effect and reflects the total expenditure of the importer, inward multilateral resistance, and unilateral trade policies such as the reduction of MFN tariff rates. In addition, the combination of these two country-year fixed effects controls for the log of the levels of the bilateral exchange rates. In all estimations, standard errors are clustered by country pair.

Equation (2) is estimated initially by ordinary least squares (OLS). Recently, however, the bias of the OLS estimator has been pointed out. Santos Silva and Tenreyro (2006) show that when a log-linearized model such as the gravity model is estimated by OLS, heteroskedasticity affects both consistency and efficiency. They recommend specifying the conditional variance as proportional to the conditional mean and estimating the log-linearized model by PPML. Our baseline specification is therefore the estimation with three types of fixed effects by PPML.12 We can deal with the zero trade flows problem by PPML. In addition, the use of PPML is supported by the need to satisfy the adding up constraint (Arvis and Shepherd 2013; Fally 2015).

To explore the role of the country-pair fixed effects, we also estimate the gravity equation as follows:
(3)
In this specification, log of distance, a contingency dummy, a common language dummy, and a common colonizer dummy are added instead of the country-pair fixed effect. Compared with equation (2), this specification ignores the effects of unobservable factors related to the level of bilateral trade flows. The endogeneity bias of trade policies becomes severe if those factors are closely related to the determinants of trade policies.
We also estimate a panel version of the naïve gravity equation (Head and Mayer 2014). In this specification, country-year fixed effects are dropped as follows:
(4)
where Remotenessit=jiGDPjtdistij-1 is the inverse of the market potential function and measures the degree of isolation from the rest of the world. Remotenessjt is defined in the same way and it is the inverse of supply potential. Other additional explanatory variables are log of GDP, log of population, a WTO dummy, and a year fixed effect. In this specification, multilateral resistance terms are not included. Although the remoteness indices are used instead, Anderson and van Wincoop (2003) and Head and Mayer (2014) criticize their use because the indices do not have a solid theoretical foundation. Furthermore, the coefficients for the FTA dummies are biased if FTAs enter into force simultaneously with unilateral trade liberalization other than WTO accession.

In this section, we provide the estimation results of equations (2)–(4). We first estimate the average treatment effects of all FTAs. Japan's FTAs are then separated from the other FTAs. We further decompose the effects of Japan's FTAs by partner countries. Individual FTAs all over the world are also investigated to explore the determinants of successful FTAs.

5.1  Trade creation effects of all FTAs

Before estimating the effects of Japan's FTAs, we first estimate the effects of all FTAs. We start with the results of the traditional gravity specification (equation [4]) by OLS. As reported in column (1) of Table 3, the coefficients for the FTA dummy and the customs union dummy are both positive and statistically significant. The coefficients for the other gravity variables are also consistent with the standard gravity model estimates in previous studies.

Table 3.
Estimation results for overall FTAs
Variables(1) OLS(2) OLS(3) OLS(4) PPML(5) PPML(6) PPML(7) PPML w/o zero
FTA 0.367*** 0.202*** 0.118*** 0.374*** 0.104** −0.00703 −0.00786 
 (5.856) (6.216) (3.266) (5.655) (2.051) (−0.127) (−0.142) 
CU 0.238** 0.561*** 0.292*** 0.656*** 0.448*** 0.0606 0.0535 
 (2.363) (9.950) (4.390) (6.536) (6.739) (0.786) (0.688) 
PSA 0.0181 0.0768 0.0663 −0.154 0.0993 −0.000895 −0.00176 
 (0.231) (0.989) (0.837) (−1.431) (1.631) (−0.0207) (−0.0407) 
Common currency −0.584*** −0.139*** −0.00833 −0.122 0.0302 −0.0386 −0.0396 
 (−6.059) (−4.521) (−0.209) (−1.590) (0.907) (−1.251) (−1.284) 
ln(Distance) −1.108***   −0.669***    
 (−27.46)   (−17.48)    
Contiguity 0.612***   0.353**    
 (4.019)   (2.257)    
Common colonizer 0.635***   0.407***    
 (5.358)   (6.170)    
Common language 0.509***   0.122*    
 (6.936)   (1.938)    
Exporter ln(GDP)  0.0710***   0.268***   
  (7.134)   (3.145)   
Importer ln(GDP)  0.218***   0.740***   
  (16.61)   (23.22)   
Exporter ln(Population)  1.448***   0.698***   
  (12.86)   (5.788)   
Importer ln(Population)  0.555***   0.0680   
  (7.393)   (0.611)   
Exporter ln(Remoteness)  0.233**   0.746***   
  (2.035)   (5.415)   
Importer ln(Remoteness)  −0.0708   −0.119   
  (−0.601)   (−0.898)   
Exporter WTO  0.594***   0.366***   
  (12.65)   (5.632)   
Importer WTO  0.465***   0.226***   
  (10.33)   (4.945)   
Observations 91,267 91,262 91,262 93,840 93,760 93,760 91,262 
Exporter-year fixed effects Yes No Yes Yes No Yes Yes 
Importer-year fixed effects Yes No Yes Yes No Yes Yes 
Exporter-importer fixed effects No Yes Yes No Yes Yes Yes 
Variables(1) OLS(2) OLS(3) OLS(4) PPML(5) PPML(6) PPML(7) PPML w/o zero
FTA 0.367*** 0.202*** 0.118*** 0.374*** 0.104** −0.00703 −0.00786 
 (5.856) (6.216) (3.266) (5.655) (2.051) (−0.127) (−0.142) 
CU 0.238** 0.561*** 0.292*** 0.656*** 0.448*** 0.0606 0.0535 
 (2.363) (9.950) (4.390) (6.536) (6.739) (0.786) (0.688) 
PSA 0.0181 0.0768 0.0663 −0.154 0.0993 −0.000895 −0.00176 
 (0.231) (0.989) (0.837) (−1.431) (1.631) (−0.0207) (−0.0407) 
Common currency −0.584*** −0.139*** −0.00833 −0.122 0.0302 −0.0386 −0.0396 
 (−6.059) (−4.521) (−0.209) (−1.590) (0.907) (−1.251) (−1.284) 
ln(Distance) −1.108***   −0.669***    
 (−27.46)   (−17.48)    
Contiguity 0.612***   0.353**    
 (4.019)   (2.257)    
Common colonizer 0.635***   0.407***    
 (5.358)   (6.170)    
Common language 0.509***   0.122*    
 (6.936)   (1.938)    
Exporter ln(GDP)  0.0710***   0.268***   
  (7.134)   (3.145)   
Importer ln(GDP)  0.218***   0.740***   
  (16.61)   (23.22)   
Exporter ln(Population)  1.448***   0.698***   
  (12.86)   (5.788)   
Importer ln(Population)  0.555***   0.0680   
  (7.393)   (0.611)   
Exporter ln(Remoteness)  0.233**   0.746***   
  (2.035)   (5.415)   
Importer ln(Remoteness)  −0.0708   −0.119   
  (−0.601)   (−0.898)   
Exporter WTO  0.594***   0.366***   
  (12.65)   (5.632)   
Importer WTO  0.465***   0.226***   
  (10.33)   (4.945)   
Observations 91,267 91,262 91,262 93,840 93,760 93,760 91,262 
Exporter-year fixed effects Yes No Yes Yes No Yes Yes 
Importer-year fixed effects Yes No Yes Yes No Yes Yes 
Exporter-importer fixed effects No Yes Yes No Yes Yes Yes 

Source: Author's calculations.

Note: Robust t-statistics in parentheses. CU = customs union dummy; FTA = free trade agreement dummy; PSA = partial scope agreement dummy. ***Statistically significant at the 1 percent level; **statistically significant at the 5 percent level; *statistically significant at the 10 percent level.

We also estimate equation (3) and report the results in column (2). Although the coefficient for the FTA dummy is significantly positive, it has halved in value. This implies that FTAs are more likely to be signed between country pairs with high ex ante trade values, conditional on gravity variables. This is the opposite result to that of Baier and Bergstrand (2007), but it is qualitatively consistent with Magee (2008). The coefficient for the customs union dummy increases. Overall, the coefficients for the other country-year variables are positive and statistically significant as expected.

Column (3) of Table 3 shows the estimation results with all three types of fixed effects. Again, although the coefficient for the FTA dummy is positive and statistically significant, it has decreased compared with columns (1) and (2). This result implies that the role of trade policies is overestimated when the gravity model is specified with all three types of fixed effects.

The results of the estimations by PPML are almost the same as those by OLS. As reported in columns (4)–(6), the coefficients for the FTA dummies and the customs union dummies are positive and statistically significant. One notable difference between the results of OLS and PPML is that when three types of fixed effects are included, the coefficients for the FTA dummies and the customs union dummies decrease to almost zero, which means that these trade policies have no trade creation effects on average.

A potential reason for the lack of significant effects is our choice of the sample period. Baier and Bergstrand (2007) and Magee (2008), for example, use the sample periods: 1960–2000 and 1980–98, respectively. The coefficient for the FTA dummy in the present paper reflects the impacts of only recently signed FTAs because the trade creation effects of FTAs enforced before 1996 are absorbed into the country-pair fixed effect. Kohl (2014) points out that FTAs enforced after the 1990s performed poorly.

5.2  Trade creation effects of Japan's FTAs

We next turn to the trade creation effects of Japan's FTAs. Table 4 shows the results of estimation, in which the effects of Japan's FTAs are separated from those of others. Other covariates are included in the estimation, but not reported in the table to save space. As reported in column (1) of Table 4, the coefficients for Japan's FTA dummies are positive and statistically significant for exports and positive and slightly insignificant for imports when the country-year fixed effects are included but the country-pair fixed effects are not. Although the results change when the country-pair fixed effects are included in the regression, the point estimates are still large in the estimation with three types of fixed effects. In addition, this result holds when the gravity model is estimated by PPML. As reported in column (6), the point estimates of the coefficients for Japan's FTA dummies are 0.067 for exports and 0.086 for imports, but they are not statistically significant.

Table 4.
Estimation results for Japan's FTAs and others
Variables(1) OLS(2) OLS(3) OLS(4) PPML(5) PPML(6) PPML(7) PPML w/o zero
Japan's FTA (export) 0.672*** −0.191* 0.178 0.504*** −0.00362 0.0672 0.0667 
 (3.636) (−1.758) (1.583) (3.245) (−0.0249) (0.781) (0.775) 
Japan's FTA (import) 0.760 −0.0525 0.106 0.423* 0.173** 0.0864 0.0855 
 (1.615) (−0.513) (0.808) (1.821) (2.516) (1.312) (1.296) 
Other FTA 0.397*** 0.154*** 0.0517 0.296*** 0.0738 −0.0439 −0.0460 
 (6.164) (4.789) (1.437) (4.267) (1.430) (−0.684) (−0.716) 
Observations 91,267 91,262 91,262 93,840 93,760 93,760 91,262 
Exporter-year fixed effects Yes No Yes Yes No Yes Yes 
Importer-year fixed effects Yes No Yes Yes No Yes Yes 
Exporter-importer fixed effects No Yes Yes No Yes Yes Yes 
Variables(1) OLS(2) OLS(3) OLS(4) PPML(5) PPML(6) PPML(7) PPML w/o zero
Japan's FTA (export) 0.672*** −0.191* 0.178 0.504*** −0.00362 0.0672 0.0667 
 (3.636) (−1.758) (1.583) (3.245) (−0.0249) (0.781) (0.775) 
Japan's FTA (import) 0.760 −0.0525 0.106 0.423* 0.173** 0.0864 0.0855 
 (1.615) (−0.513) (0.808) (1.821) (2.516) (1.312) (1.296) 
Other FTA 0.397*** 0.154*** 0.0517 0.296*** 0.0738 −0.0439 −0.0460 
 (6.164) (4.789) (1.437) (4.267) (1.430) (−0.684) (−0.716) 
Observations 91,267 91,262 91,262 93,840 93,760 93,760 91,262 
Exporter-year fixed effects Yes No Yes Yes No Yes Yes 
Importer-year fixed effects Yes No Yes Yes No Yes Yes 
Exporter-importer fixed effects No Yes Yes No Yes Yes Yes 

Source: Author's calculations.

Note: Robust t-statistics in parentheses. OLS = ordinary least squares; PPML = Poisson pseudo maximum likelihood. ***Statistically significant at the 1 percent level; **statistically significant at the 5 percent level; *statistically significant at the 10 percent level.

Overall, Japan's FTAs do not appear to have positive effects on trade values when the model is correctly specified. Although the coefficients are statistically insignificant, the point estimates are not small. These results weakly suggest that the effects of Japan's FTAs are heterogeneous.

5.3  Trade creation effects of Japan's individual FTAs

The results of the previous subsection suggest that some of Japan's FTAs are working well, but others are not. To examine this point further, we decompose the effects of Japan's FTAs by partner countries.

Table 5 presents the coefficients for Japan's FTA dummies. We regard column (4), which includes three types of fixed effects and estimates by PPML, as the most reliable result.13 We also add the results from other studies, Ando and Urata (2015) and Yamanouchi (2017), in columns (5) and (6). Based on column (4), the export values from Japan to Australia, Chile, India, Indonesia, Mexico, Myanmar, Thailand, and Vietnam are positively affected by the FTAs with these countries. The FTA with Myanmar (AJCEP) has the largest effect and it increased Japan's exports to Myanmar by exp(0.517)-1=67.7percent. This result is surprising because Myanmar's tariff rates were not lowered in the sample period under AJCEP. Therefore, this implies that removing nontariff barriers is crucial for trade creation. In contrast, FTAs with Brunei, Cambodia, Laos, Peru, the Philippines, and Switzerland have no significant effects on Japan's exports. The coefficients for FTAs with Malaysia and Singapore are negative and statistically significant.

Table 5.
Estimation results for Japan's individual FTAs
(1) OLS(2) PPML(3) PPML(4) PPML(5) Ando and Urata (2015) OLS(6) Yamanouchi (2017) PPML
VariablesExportImportExportImportExportImportExportImportExportImportExportImport
FTA (Australia) 0.285** 0.350*** 0.312** 1.469*** 0.0401 0.139*** 0.153** 0.190**     
 (2.202) (3.352) (2.415) (3.658) (0.896) (3.442) (2.135) (2.008)     
FTA (Brunei) 0.132 0.803*** 0.188 2.473*** 0.0685 0.557*** 0.108 0.627***   −0.0429 0.370*** 
 (1.495) (3.597) (0.951) (6.187) (1.600) (16.12) (1.119) (3.025)     
FTA (Cambodia) 0.125 −0.245 −0.491** −0.0673 0.134*** 1.245*** 0.0570 0.610***   0.114*** 0.413*** 
 (0.988) (−1.595) (−2.226) (−0.192) (2.809) (31.42) (0.470) (3.252)     
FTA (Chile) 0.744*** 0.132 0.147 1.350*** 0.462*** 0.369*** 0.397*** 0.155 0.593*** 0.0533 0.595*** 0.109** 
 (5.023) (1.228) (1.094) (4.393) (10.85) (9.481) (2.728) (1.603) (4.556) (0.578)   
FTA (India) −0.175 −0.252*** −0.631*** −0.639*** 0.469*** 0.330*** 0.170** −0.0246     
 (−1.156) (−3.389) (−3.347) (−2.680) (11.24) (9.168) (2.511) (−0.323)     
FTA (Indonesia) 0.271** −0.0343 0.585*** 0.993*** 0.302*** −0.000329 0.265** −0.0129 −0.106 −0.303*** 0.0122 −0.201** 
 (2.508) (−0.458) (4.385) (5.322) (6.463) (−0.00699) (2.561) (−0.213) (−0.710) (−2.856)   
FTA (Laos) 0.853*** 1.041*** −0.634 −0.730** −0.180*** 1.161*** −0.151 0.594***   0.560*** 1.108*** 
 (5.839) (5.470) (−1.192) (−1.994) (−3.274) (22.66) (−1.097) (2.650)     
FTA (Malaysia) −0.520*** 0.0652 0.796*** 0.711*** −0.452*** 0.231*** −0.178** 0.304*** −0.148 −0.0262 −0.220*** −0.0515 
 (−3.965) (0.853) (6.387) (3.270) (−8.978) (6.386) (−2.490) (3.599) (−1.201) (−0.301)   
FTA (Mexico) 0.192* −0.223** 0.361*** −0.799*** 0.534*** 0.214*** 0.428*** 0.00810 0.628*** 0.264*** 0.498*** 0.141** 
 (1.795) (−2.340) (3.012) (−3.572) (12.84) (3.680) (4.567) (0.128) (4.800) (2.844)   
FTA (Myanmar) 0.864*** 0.874*** 0.515** 0.211 0.0588 0.530*** 0.517*** 0.679***     
 (6.553) (4.820) (2.334) (0.547) (1.074) (5.468) (3.776) (2.734)     
FTA (Peru) 0.148 0.238* 0.0522 0.632** 0.185*** 0.341*** 0.0251 −0.0108     
 (1.263) (1.929) (0.356) (2.182) (4.430) (9.035) (0.225) (−0.0902)     
FTA (Philippines) −0.0955 −0.199** 0.267 0.433* −0.716*** −0.158*** −0.144 0.199* −0.216 −0.400*** −0.181*** −0.369*** 
 (−0.871) (−2.169) (1.544) (1.958) (−15.71) (−4.556) (−1.583) (1.841) (−1.464) (−3.822)   
FTA (Singapore) −0.316*** −0.113 0.397** −0.348* −0.486*** −0.202*** −0.307*** −0.138 −0.0863 −0.118 −0.485*** −0.407*** 
 (−3.012) (−1.357) (2.563) (−1.705) (−9.341) (−4.993) (−4.272) (−0.884) (−0.445) (−0.857)   
FTA (Switzerland) 0.0576 0.184** −0.0444 0.471* 0.251*** 0.137*** 0.0990 −0.00131 0.914*** 0.0844 0.391*** 0.128** 
 (0.544) (2.246) (−0.325) (1.945) (6.393) (2.615) (0.919) (−0.0139) (4.704) (0.612)   
FTA (Thailand) 0.343*** −0.281*** 1.203*** 0.506*** 0.208*** 0.295*** 0.211*** 0.0366 0.0366 −0.0761 0.0664 −0.0248 
 (3.687) (−3.308) (9.873) (2.762) (4.124) (7.980) (3.393) (0.444) (0.281) (−0.824)   
FTA (Vietnam) −0.0664 −0.448*** 0.575*** 0.598** 0.321*** 0.551*** 0.209* −0.153 0.239 0.0845 0.182*** 0.175*** 
 (−0.520) (−3.978) (3.814) (2.436) (5.802) (10.54) (1.856) (−1.177) (1.599) (0.799)   
Observations 91,262 93,840 93,760 93,760 360 360 1,908 1,908 
Exporter-year fixed effects Yes Yes No Yes Yes Yes Yes Yes 
Importer-year fixed effects Yes Yes No Yes No No No No 
Exporter-importer fixed effects Yes No Yes Yes Yes Yes Yes Yes 
(1) OLS(2) PPML(3) PPML(4) PPML(5) Ando and Urata (2015) OLS(6) Yamanouchi (2017) PPML
VariablesExportImportExportImportExportImportExportImportExportImportExportImport
FTA (Australia) 0.285** 0.350*** 0.312** 1.469*** 0.0401 0.139*** 0.153** 0.190**     
 (2.202) (3.352) (2.415) (3.658) (0.896) (3.442) (2.135) (2.008)     
FTA (Brunei) 0.132 0.803*** 0.188 2.473*** 0.0685 0.557*** 0.108 0.627***   −0.0429 0.370*** 
 (1.495) (3.597) (0.951) (6.187) (1.600) (16.12) (1.119) (3.025)     
FTA (Cambodia) 0.125 −0.245 −0.491** −0.0673 0.134*** 1.245*** 0.0570 0.610***   0.114*** 0.413*** 
 (0.988) (−1.595) (−2.226) (−0.192) (2.809) (31.42) (0.470) (3.252)     
FTA (Chile) 0.744*** 0.132 0.147 1.350*** 0.462*** 0.369*** 0.397*** 0.155 0.593*** 0.0533 0.595*** 0.109** 
 (5.023) (1.228) (1.094) (4.393) (10.85) (9.481) (2.728) (1.603) (4.556) (0.578)   
FTA (India) −0.175 −0.252*** −0.631*** −0.639*** 0.469*** 0.330*** 0.170** −0.0246     
 (−1.156) (−3.389) (−3.347) (−2.680) (11.24) (9.168) (2.511) (−0.323)     
FTA (Indonesia) 0.271** −0.0343 0.585*** 0.993*** 0.302*** −0.000329 0.265** −0.0129 −0.106 −0.303*** 0.0122 −0.201** 
 (2.508) (−0.458) (4.385) (5.322) (6.463) (−0.00699) (2.561) (−0.213) (−0.710) (−2.856)   
FTA (Laos) 0.853*** 1.041*** −0.634 −0.730** −0.180*** 1.161*** −0.151 0.594***   0.560*** 1.108*** 
 (5.839) (5.470) (−1.192) (−1.994) (−3.274) (22.66) (−1.097) (2.650)     
FTA (Malaysia) −0.520*** 0.0652 0.796*** 0.711*** −0.452*** 0.231*** −0.178** 0.304*** −0.148 −0.0262 −0.220*** −0.0515 
 (−3.965) (0.853) (6.387) (3.270) (−8.978) (6.386) (−2.490) (3.599) (−1.201) (−0.301)   
FTA (Mexico) 0.192* −0.223** 0.361*** −0.799*** 0.534*** 0.214*** 0.428*** 0.00810 0.628*** 0.264*** 0.498*** 0.141** 
 (1.795) (−2.340) (3.012) (−3.572) (12.84) (3.680) (4.567) (0.128) (4.800) (2.844)   
FTA (Myanmar) 0.864*** 0.874*** 0.515** 0.211 0.0588 0.530*** 0.517*** 0.679***     
 (6.553) (4.820) (2.334) (0.547) (1.074) (5.468) (3.776) (2.734)     
FTA (Peru) 0.148 0.238* 0.0522 0.632** 0.185*** 0.341*** 0.0251 −0.0108     
 (1.263) (1.929) (0.356) (2.182) (4.430) (9.035) (0.225) (−0.0902)     
FTA (Philippines) −0.0955 −0.199** 0.267 0.433* −0.716*** −0.158*** −0.144 0.199* −0.216 −0.400*** −0.181*** −0.369*** 
 (−0.871) (−2.169) (1.544) (1.958) (−15.71) (−4.556) (−1.583) (1.841) (−1.464) (−3.822)   
FTA (Singapore) −0.316*** −0.113 0.397** −0.348* −0.486*** −0.202*** −0.307*** −0.138 −0.0863 −0.118 −0.485*** −0.407*** 
 (−3.012) (−1.357) (2.563) (−1.705) (−9.341) (−4.993) (−4.272) (−0.884) (−0.445) (−0.857)   
FTA (Switzerland) 0.0576 0.184** −0.0444 0.471* 0.251*** 0.137*** 0.0990 −0.00131 0.914*** 0.0844 0.391*** 0.128** 
 (0.544) (2.246) (−0.325) (1.945) (6.393) (2.615) (0.919) (−0.0139) (4.704) (0.612)   
FTA (Thailand) 0.343*** −0.281*** 1.203*** 0.506*** 0.208*** 0.295*** 0.211*** 0.0366 0.0366 −0.0761 0.0664 −0.0248 
 (3.687) (−3.308) (9.873) (2.762) (4.124) (7.980) (3.393) (0.444) (0.281) (−0.824)   
FTA (Vietnam) −0.0664 −0.448*** 0.575*** 0.598** 0.321*** 0.551*** 0.209* −0.153 0.239 0.0845 0.182*** 0.175*** 
 (−0.520) (−3.978) (3.814) (2.436) (5.802) (10.54) (1.856) (−1.177) (1.599) (0.799)   
Observations 91,262 93,840 93,760 93,760 360 360 1,908 1,908 
Exporter-year fixed effects Yes Yes No Yes Yes Yes Yes Yes 
Importer-year fixed effects Yes Yes No Yes No No No No 
Exporter-importer fixed effects Yes No Yes Yes Yes Yes Yes Yes 

Source: Columns (1)–(4) are calculated by the author. Column (5) is taken from Ando and Urata (2015, Table 10). Column (6) is taken from Yamanouchi (2017, Tables 43 and 47).

Note: Robust t-statistics in parentheses. OLS = ordinary least squares; PPML = Poisson pseudo maximum likelihood. ***Statistically significant at the 1 percent level; **statistically significant at the 5 percent level; *statistically significant at the 10 percent level.

Import values, however, increased significantly because of FTAs with Australia, Brunei, Cambodia, Laos, Malaysia, Myanmar, and the Philippines. The coefficients are statistically insignificant for the other FTAs. The largest effect on imports is also observed for the FTA with Myanmar.

We also check the symmetry of the trade creation effects using a joint test of the hypothesis that all pairs of coefficients are equal for exports and imports. The chi-square statistic is 103.1 and the hypothesis is clearly rejected.

Tables 6 and 7 present the matrices of the root mean square differences between the coefficients for the FTA dummies across specifications. Table 6 shows the root mean square differences of the coefficients for Japan's exports to its FTA partners. The coefficients in our baseline specification (column [6]) differ little from those estimated without country-year dummies. The choice of estimator does not matter much. However, the specifications without the country-pair dummies (columns [1] and [4]) show very different coefficients. Table 7 shows the same matrix for Japan's imports. The results are like the case of exports, and misspecification is problematic if the country-pair fixed effects are excluded from the estimation. If the country-pair fixed effects are not included, the trade creation effects of the FTAs are overestimated because of endogeneity.

Table 6.
Matrix of root mean square differences between coefficients for Japan's FTAs (Exports)
Export(1)(2)(3)(4)(5)(6)
(1) OLS, no pair dummies 0.00      
(2) OLS, no country-year dummies 1.20 0.00     
(3) OLS, three types 0.68 0.60 0.00    
(4) PPML, no pair dummies 0.89 0.79 0.66 0.00   
(5) PPML, no country-year dummies 0.97 0.29 0.43 0.65 0.00  
(6) PPML, three types 0.85 0.43 0.32 0.51 0.23 0.00 
Export(1)(2)(3)(4)(5)(6)
(1) OLS, no pair dummies 0.00      
(2) OLS, no country-year dummies 1.20 0.00     
(3) OLS, three types 0.68 0.60 0.00    
(4) PPML, no pair dummies 0.89 0.79 0.66 0.00   
(5) PPML, no country-year dummies 0.97 0.29 0.43 0.65 0.00  
(6) PPML, three types 0.85 0.43 0.32 0.51 0.23 0.00 

Source: Author's calculations.

Note: OLS = ordinary least squares; PPML = Poisson pseudo maximum likelihood.

Table 7.
Matrix of root mean square differences between coefficients for Japan's FTAs (Imports)
Import(1)(2)(3)(4)(5)(6)
(1) OLS, no pair dummies 0.00      
(2) OLS, no country-year dummies 1.89 0.00     
(3) OLS, three types 1.62 0.46 0.00    
(4) PPML, no pair dummies 1.29 1.08 0.92 0.00   
(5) PPML, no country-year dummies 1.74 0.44 0.53 0.99 0.00  
(6) PPML, three types 1.69 0.38 0.32 0.89 0.34 0.00 
Import(1)(2)(3)(4)(5)(6)
(1) OLS, no pair dummies 0.00      
(2) OLS, no country-year dummies 1.89 0.00     
(3) OLS, three types 1.62 0.46 0.00    
(4) PPML, no pair dummies 1.29 1.08 0.92 0.00   
(5) PPML, no country-year dummies 1.74 0.44 0.53 0.99 0.00  
(6) PPML, three types 1.69 0.38 0.32 0.89 0.34 0.00 

Source: Author's calculations.

Note: OLS = ordinary least squares; PPML = Poisson pseudo maximum likelihood.

In this subsection, the effects of the individual FTAs are explored. We found that each of Japan's FTAs affects trade values in a different way. Therefore, we conclude that one of the reasons for the absence of the trade creation effects of Japan's FTAs is that aggregation across all trade partners obscures the positive effects. We should note that about half of Japan's FTAs have positive and statistically significant impacts.

5.4  Trade creation effects of each individual FTA

In this subsection, we consider the determinants of the trade creation effects. To this end, we first obtain the coefficients for each individual FTA. We then regress the estimated coefficients on some variables. This two-step estimation is suggested in Baier, Yotov, and Zylkin (2019).

The first stage is estimated by OLS as in Kohl (2014) because of the computational difficulty. As explained in the previous subsection, the differences between coefficients in the two estimates are small. Among the 725 directional flows within active FTAs, 256 (35 percent) of these flows have positive and statistically significant values at the 5 percent level. Mean and median values are 0.113 and 0.105, respectively.14 Compared with Baier, Yotov, and Zylkin (2019), these values are small and suggest that recent FTAs have weaker effects.

To examine the effects of individual FTAs, we regress the estimated coefficients on some other variables in the second stage. The estimating equation is as follows:
(5)
where δX¯i and δM¯j are the means of exporter-year and importer-year fixed effects estimated in the first stage, respectively. These variables measure the size of countries in terms of trade values. δijB is the estimated country-pair fixed effect and a proxy for the strength of trade linkages before the FTAs were established. Intuitively, these three variables are expected to have negative coefficients because there is no room for more trade if a country is already open to trade. The last variable, FTAyearsij, equals the number of years for which the country pair has an FTA. The coefficient is expected to be positive if recent FTAs are not effective. The cohort effects of FTAs are not identified from phase-in effects, so the positive coefficient may be a result of phase-in effects.15
In another specification, we replace the estimated fixed effects with some gravity variables as follows:
(6)
where the expected signs are the same as for equation (5), except ν3 because the country-pair fixed effects are smaller for longer distances. Following Baier, Yotov, and Zylkin (2019), we simply use heteroskedasticity-robust standard errors. Furthermore, we include country fixed effects again instead of these variables in some estimations of the second stage.

The estimation results of equation (5) are shown in Table 8. In the first column, the signs of the coefficients are as expected. The FTAs generally increase trade values if the trading countries are smaller and the relationship between the two countries is weak. In addition, the results imply that relatively old FTAs are more effective.

Table 8.
Estimation results for the effects of individual FTAs using estimated fixed effects
(1)(2)(3)(4)(5)
Importer-year fixed effect −0.0529*** −0.0550*** −0.0369   
 (−3.828) (−3.729) (−1.519)   
Importer-year fixed effect  −0.0759 −0.0869***   
× Export from Japan  (−1.423) (−3.583)   
Exporter-year fixed effect −0.0447* −0.0437*  −0.0631*  
 (−1.822) (−1.650)  (−1.868)  
Exporter-year fixed effect  −0.0860  −0.0716**  
× Import to Japan  (−1.331)  (−2.121)  
Country pair fixed effect −0.128*** −0.139*** −0.160*** −0.163*** −0.178*** 
 (−4.329) (−4.531) (−3.241) (−4.786) (−3.366) 
Country pair fixed effect  −0.0191 −0.0182 0.137 0.146** 
× Export from Japan  (−0.128) (−0.368) (1.534) (2.123) 
Country pair fixed effect  0.171** 0.0406 0.161*** 0.0637 
× Import to Japan  (2.070) (0.526) (4.753) (1.000) 
Number of years under FTA 0.0147*** 0.0158*** 0.0253*** 0.0231** 0.0289** 
 (2.795) (2.981) (3.088) (2.540) (2.351) 
Number of years under FTA  0.0249 −0.0328*** 0.0147 0.00231 
× Export from Japan  (1.319) (−4.009) (1.095) (0.153) 
Number of years under FTA  0.000187 −0.00323 −0.0536*** −0.0391*** 
× Import to Japan  (0.0142) (−0.286) (−5.884) (−3.888) 
Observations 725 725 723 723 721 
Adjusted R-squared 0.070 0.075 0.184 0.155 0.250 
Exporter fixed effects No No Yes No Yes 
Importer fixed effects No No No Yes Yes 
(1)(2)(3)(4)(5)
Importer-year fixed effect −0.0529*** −0.0550*** −0.0369   
 (−3.828) (−3.729) (−1.519)   
Importer-year fixed effect  −0.0759 −0.0869***   
× Export from Japan  (−1.423) (−3.583)   
Exporter-year fixed effect −0.0447* −0.0437*  −0.0631*  
 (−1.822) (−1.650)  (−1.868)  
Exporter-year fixed effect  −0.0860  −0.0716**  
× Import to Japan  (−1.331)  (−2.121)  
Country pair fixed effect −0.128*** −0.139*** −0.160*** −0.163*** −0.178*** 
 (−4.329) (−4.531) (−3.241) (−4.786) (−3.366) 
Country pair fixed effect  −0.0191 −0.0182 0.137 0.146** 
× Export from Japan  (−0.128) (−0.368) (1.534) (2.123) 
Country pair fixed effect  0.171** 0.0406 0.161*** 0.0637 
× Import to Japan  (2.070) (0.526) (4.753) (1.000) 
Number of years under FTA 0.0147*** 0.0158*** 0.0253*** 0.0231** 0.0289** 
 (2.795) (2.981) (3.088) (2.540) (2.351) 
Number of years under FTA  0.0249 −0.0328*** 0.0147 0.00231 
× Export from Japan  (1.319) (−4.009) (1.095) (0.153) 
Number of years under FTA  0.000187 −0.00323 −0.0536*** −0.0391*** 
× Import to Japan  (0.0142) (−0.286) (−5.884) (−3.888) 
Observations 725 725 723 723 721 
Adjusted R-squared 0.070 0.075 0.184 0.155 0.250 
Exporter fixed effects No No Yes No Yes 
Importer fixed effects No No No Yes Yes 

Source: Author's calculations.

Note: Robust t-statistics in parentheses. OLS = ordinary least squares; PPML = Poisson pseudo maximum likelihood. ***Statistically significant at the 1 percent level; **statistically significant at the 5 percent level; *statistically significant at the 10 percent level.

In columns (2)–(5), the interaction terms with Japan's export and import dummies are added to explore the characteristics of Japan's FTAs.16 Although the results are not stable, the effects of Japan's FTAs are larger when the partner countries are smaller. However, the trade linkages before FTAs are signed are not important. More interestingly, recently enforced FTAs increase Japan's import values. In the case of Japan, agricultural goods are excluded from the negotiations in initial agreements. Protection is likely to be weakened and tariff rates for some products such as beef are lowered in the recent agreements with Australia.17,18

In Table 9, we present the estimation results of equation (6). While the main results are unchanged from the previous analyses, we find some notable differences. First, exporter GDP does not explain the effectiveness of an FTA.19 This implies that some factors related to export values but not to GDP are key determinants of trade creation effects. One potential explanation is the endowment of natural resources. The export values of natural resources are not closely related to tariff rates, so the countries specializing in those resource sectors cannot increase exports via FTAs. Similarly, the coefficients for distance are statistically insignificant in some specifications, and therefore the role of distance is less clear than the country-pair fixed effects. However, distance is important for Japan's import under FTAs.

Table 9.
Estimation results for the effects of individual FTAs using GDP and distance
(1)(2)(3)(4)(5)
Importer ln(GDP) −0.0345** −0.0350* −0.0238   
 (−2.047) (−1.893) (−0.878)   
Importer ln(GDP)  −0.101** −0.118***   
× Export from Japan  (−2.263) (−4.345)   
Exporter ln(GDP) −0.0251 −0.0261  −0.0307  
 (−0.957) (−0.905)  (−0.942)  
Exporter ln(GDP)  −0.124**  −0.148***  
× Import to Japan  (−2.427)  (−4.550)  
ln(Distance) 0.0751* 0.0728* 0.105 0.0904 0.174* 
 (1.756) (1.671) (1.474) (1.309) (1.689) 
ln(Distance)  0.347** 0.271*** 0.0148 0.161 
× Export from Japan  (2.439) (3.799) (0.743) (1.573) 
ln(Distance)  0.409*** 0.0326** 0.144** −0.151 
× Import to Japan  (2.695) (2.393) (2.089) (−1.370) 
Number of years under FTA 0.0143*** 0.0150*** 0.0210** 0.0210** 0.0268** 
 (2.621) (2.694) (2.420) (2.544) (2.107) 
Number of years under FTA  −0.0361 −0.0456*** 0.000480 0.0109 
× Export from Japan  (−1.444) (−5.261) (0.0209) (0.383) 
Number of years under FTA  −0.0357 −0.0416*** −0.0617*** −0.0383*** 
× Import to Japan  (−1.604) (−3.720) (−7.478) (−3.884) 
Observations 725 725 723 723 721 
Adjusted R-squared 0.010 0.008 0.119 0.067 0.181 
Exporter fixed effects No No Yes No Yes 
Importer fixed effects No No No Yes Yes 
(1)(2)(3)(4)(5)
Importer ln(GDP) −0.0345** −0.0350* −0.0238   
 (−2.047) (−1.893) (−0.878)   
Importer ln(GDP)  −0.101** −0.118***   
× Export from Japan  (−2.263) (−4.345)   
Exporter ln(GDP) −0.0251 −0.0261  −0.0307  
 (−0.957) (−0.905)  (−0.942)  
Exporter ln(GDP)  −0.124**  −0.148***  
× Import to Japan  (−2.427)  (−4.550)  
ln(Distance) 0.0751* 0.0728* 0.105 0.0904 0.174* 
 (1.756) (1.671) (1.474) (1.309) (1.689) 
ln(Distance)  0.347** 0.271*** 0.0148 0.161 
× Export from Japan  (2.439) (3.799) (0.743) (1.573) 
ln(Distance)  0.409*** 0.0326** 0.144** −0.151 
× Import to Japan  (2.695) (2.393) (2.089) (−1.370) 
Number of years under FTA 0.0143*** 0.0150*** 0.0210** 0.0210** 0.0268** 
 (2.621) (2.694) (2.420) (2.544) (2.107) 
Number of years under FTA  −0.0361 −0.0456*** 0.000480 0.0109 
× Export from Japan  (−1.444) (−5.261) (0.0209) (0.383) 
Number of years under FTA  −0.0357 −0.0416*** −0.0617*** −0.0383*** 
× Import to Japan  (−1.604) (−3.720) (−7.478) (−3.884) 
Observations 725 725 723 723 721 
Adjusted R-squared 0.010 0.008 0.119 0.067 0.181 
Exporter fixed effects No No Yes No Yes 
Importer fixed effects No No No Yes Yes 

Source: Author's calculations.

Note: Robust t-statistics in parentheses. OLS = ordinary least squares; PPML = Poisson pseudo maximum likelihood. ***Statistically significant at the 1 percent level; **statistically significant at the 5 percent level; *statistically significant at the 10 percent level.

In this subsection, we discussed which types of countries have effective FTAs. In a nutshell, a small ex ante trade value means substantial scope to increase trade via FTAs. Japan's FTA partners are so far mainly located in the Asia–Pacific region and actively transacting with Japan before signing the agreements. Japan can be integrated with the global market through FTAs along with many developing countries.

FTAs are one of the major commercial policies of the 21st century, yet Japan's FTAs have not been evaluated adequately. This paper therefore investigates the effects of Japan's FTAs using a recently developed gravity framework and explores the determinants of the effects of FTAs.

Our estimation results do not indicate the presence of trade creation effects for Japan's FTAs on average. Nonetheless, the effects of Japan's FTAs vary substantially across partners and around half of the FTAs increase Japan's trade values. Positive impacts on Japan's exports are more likely to be observed for small partners. Japan's imports from FTA partners tend to increase when the partner countries are small and distant. More importantly, Japan's recent FTAs have larger effects.

Our results suggest that there is a little scope to increase trade values with some countries. This implies that political resources should be directed toward negotiations with developing countries if the government is aiming to integrate the economy into the global market through FTAs. Large-scaled multilateral trade agreements may not be effective in terms of trade creation, however, even though large amounts of effort are spent on the negotiation. Of course, the impacts on investment and other forms of international cooperation are different and those multilateral agreements may play important roles in regulating world trade systems and supporting new types of globalization. These are issues open for future research.

*

I am grateful to Sungbae An, Mateus Silva Chang, Vu Quoc Huy, Banri Ito, Hayato Kato, Fukunari Kimura, Kozo Kiyota, and participants at the 7th Spring Meeting of the Japan Society of International Economics at Matsuyama University and the Asian Economic Panel at Keio University for useful comments and suggestions.

1 

See http://rtais.wto.org/UI/PublicMaintainRTAHome.aspx (last accessed 30 August 2018).

2 

See Table 1 for the list of partner countries. More detailed information is available in Ando and Urata (2015). While most of Japan's trade deals are economic partnership agreements, covering a wide range of issues, we use the term FTA throughout the paper for brevity.

3 

In the first stage, we estimate the crude effects of FTAs. Various kinds of related factors such as interactions among other FTAs and the formation of production networks are included.

4 

The roles of these fixed effects are discussed in Section 4.

5 

Among the 166 agreements studied in Kohl (2014), only 44 agreements have a trade-promoting effect.

6 

Some papers study the effects of FTAs in East Asia other than for Japan. Yean and Yi (2014), for example, explore the ASEAN–China FTA. Chia (2013, 2015) discusses the prospect of economic integration in the region.

7 

In our data set, the correlation of trade values reported by importers and by exporters is 0.88, but the mean of the logged trade gap (the difference between log of import values and log of export values) is 0.26 and the median is 0.14.

8 

The data set is available at Centre d'Etudes Prospectives et d'Informations Internationales (CEPII): www.cepii.fr/CEPII/en/bdd_modele/presentation.asp?id=8 (last accessed 2 September 2018).

9 

The data set is available at www.ewf.uni-bayreuth.de/en/research/RTA-data/index.html (last accessed 2 September 2018).

10 

Mongolia is included in the sample as an FTA partner of Japan; however, the FTA was not enforced in the sample period.

11 

The estimated coefficients for the FTA dummies are considered to be average effects over time. Although we do not explicitly consider the phase-in effects in the first stage, it is partially addressed in the second stage.

12 

PPML with high-dimensional fixed effects is computationally demanding. In this paper, we use the Stata command ppml_panel_sg written by Larch et al. (2017). See their paper for the detailed procedure.

13 

As robustness checks, we also estimate the model (1) without zero trade values, (2) without interpolated values, (3) limiting the sample to impose one-year intervals, and (4) including pair-specific linear trends. The results do not change substantially.

14 

301 (42 percent) are insignificant and 168 (23 percent) are negatively significant.

15 

Yamanouchi (2017) does not find phase-in effects for Japan's FTAs.

16 

We also check the results of regressions estimated using Japan's variables only. Of course, the sample size is so small that the results are only indicative, but the main results are unchanged from Tables 8 and 9.

17 

We cannot see the effects of lowered tariff rates using simple statistics because Japan's import values of agricultural goods from Australia decreased after the FTA entered into force. As explained in Section 3, we cannot make conclusions about the effects of FTAs from the descriptive analysis. In fact, Australia's total exports also decreased in 2015.

18 

It is difficult to explain the large effects of Japan's recent FTAs by the differences of contents across Japan's FTAs. Almost all FTAs have chapters on investment and trade in services. Provisions on intellectual property, movement of natural persons, and government procurement are not limited to recent FTAs.

19 

Baier, Yotov, and Zylkin (2019) report positive coefficients for both exporter and importer GDP.

Anderson
,
James E.
, and
Eric van
Wincoop
.
2003
.
Gravity with Gravitas: A Solution to the Border Puzzle
.
American Economic Review
93
(
1
):
170
192
.
Ando
,
Mitsuyo
, and
Shujiro
Urata
.
2011
. Impacts of the Japan–Mexico EPA on Bilateral Trade.
RIETI Discussion Paper No. 11-E-020
.
Tokyo
:
Research Institute of Economy, Trade and Industry
.
Ando
,
Mitsuyo
, and
Shujiro
Urata
.
2015
. Impacts of Japan's FTAs on Trade: The Cases of FTAs with Malaysia, Thailand, and Indonesia.
RIETI Discussion Paper No. 15-E-104
.
Tokyo
:
Research Institute of Economy, Trade and Industry
.
Arvis
,
Jean-François
, and
Ben
Shepherd
.
2013
. The Poisson Quasi-Maximum Likelihood Estimator: A Solution to the “Adding Up” Problem in Gravity Models.
Applied Economics Letters
20
(
6
):
515
519
.
Baier
,
Scott L.
, and
Jeffrey H.
Bergstrand
.
2007
. Do Free Trade Agreements Actually Increase Members’ International Trade?
Journal of International Economics
71
(
1
):
72
95
.
Baier
,
Scott L.
,
Jeffrey H.
Bergstrand
, and
Matthew W.
Clance
.
2018
.
Heterogeneous Effects of Economic Integration Agreements
.
Journal of Development Economics
135
(
1
):
587
608
.
Baier
,
Scott L.
,
Yoto V.
Yotov
, and
Thomas
Zylkin
.
2019
.
On the Widely Differing Effects of Free Trade Agreements: Lessons from Twenty Years of Trade Integration
.
Journal of International Economics
116
(
1
):
206
226
.
Chia
,
Siow Yue
.
2013
.
The Emerging Regional Economic Integration Architecture in East Asia
.
Asian Economic Papers
12
(
1
):
1
37
.
Chia
,
Siow Yue
.
2015
.
Emerging Mega-FTAs: Rationale, Challenges, and Implications
.
Asian Economic Papers
14
(
1
):
1
27
.
Cipollina
,
Maria
, and
Luca
Salvatici
.
2010
.
Reciprocal Trade Agreements in Gravity Models: A Meta-Analysis
.
Review of International Economics
18
(
1
):
63
80
.
Egger
,
Peter
, and
Mario
Larch
.
2008
.
Interdependent Preferential Trade Agreement Memberships: An Empirical Analysis
.
Journal of International Economics
76
(
2
):
384
399
.
Eicher
,
Theo S.
,
Christian
Henn
, and
Chris
Papageorgiou
.
2012
.
Trade Creation and Diversion Revisited: Accounting for Model Uncertainty and Natural Trading Partner Effects
.
Journal of Applied Econometrics
27
(
2
):
296
321
.
Fally
,
Thibault
.
2015
.
Structural Gravity and Fixed Effects
.
Journal of International Economics
97
(
1
):
76
85
.
Head
,
Keith
, and
Thierry
Mayer
.
2014
. Gravity Equations: Workhorse, Toolkit, and Cookbook. In:
Handbook of International Economics
, vol. 4,
edited by
Gita
Gopinath
,
Elhanan
Helpman
, and
Kenneth
Rogoff
, pp.
131
195
.
Amsterdam
:
Elsevier B.V
.
Head
,
Keith
,
Thierry
Mayer
, and
John
Ries
.
2010
. The Erosion of Colonial Trade Linkages After Independence.
Journal of International Economics
81
(
1
):
1
14
.
Kohl
,
Tristan
.
2014
.
Do We Really Know That Trade Agreements Increase Trade
?
Review of World Economics
150
(
3
):
443
469
.
Larch
,
Mario
,
Joschka
Wanner
,
Yoto V.
Yotov
, and
Thomas
Zylkin
.
2017
. The Currency Union Effect: A PPML Re-Assessment with High-Dimensional Fixed Effects.
CESifo Working Paper Series No. 6464
.
Munich
:
CESifo Group Munich
.
Magee
,
Christopher S. P.
2008
. New Measures of Trade Creation and Trade Diversion.
Journal of International Economics
75
(
2
):
349
362
.
Okabe
,
Misa
.
2015
. Impact of Free Trade Agreements on Trade in East Asia.
ERIA Discussion Paper Series No. 2015-1
.
Jakarta
:
Economic Research Institute for ASEAN and East Asia
.
Santos
Silva
,
J. M.
C.
, and
Silvana
Tenreyro
.
2006
.
The Log of Gravity
.
Review of Economics and Statistics
88
(
4
):
641
658
.
Yamanouchi
,
Kenta
.
2017
. Nihon no Jiyuu Boueki Kyoutei (FTA) no Boueki Soushutsu Kouka [The Trade Creation Effects of Japan's Free Trade Agreements].
Tokyo
:
Mitsubishi Economic Research Institute (in Japanese)
.
Yean
,
Tham Siew
, and
Andrew Kam Jia
Yi
.
2014
. Re-examining the Impact of ACFTA on ASEAN's Exports of Manufactured Goods to China.
Asian Economic Papers
13
(
3
):
63
82
.
Yotov
,
Yoto V.
,
Roberta
Piermartini
,
José-Antonio
Monteiro
, and
Mario
Larch
.
2016
.
An Advanced Guide to Trade Policy Analysis: The Structural Gravity Model
.
Geneva
:
World Trade Organization
.
Zylkin
,
Thomas
.
2016
. Beyond Tariffs: Quantifying Heterogeneity in the Effects of Free Trade Agreements.
GPN Working Paper Series No. GPN2016_006
.
Singapore
:
Global Production Networks Centre at National University of Singapore
.