Export, Logistics Performance, and Regional Economic Integration: Sectoral and Sub-Sectoral Evidence from Vietnam

The trade gravity model was primarily proposed by Pöyhönen (1963) and Tinbergen and Tinbergen (1962), which was based on a rule in physics which indicates the gravitational force between the two objects is proportional to their weights and inversely proportional to their spatial distance. This rule when being applied to the explanation of trade considers the cross-border flows of goods as gravitational force between the two objects, which are simultaneously considered as the two economies whose gross domestic product (GDP) and geographical distance are corresponding to physical weights and spatial distance of the two objects respectively. Thus, the original form of the gravity model is as follows:

In the above equation, EXP_ijt represents the export from country i to country j in year t, while Y_it and Y_jt are GDP in year t of country i and j respectively, and DIST_ij is the geographical distance between country i and j. From the above equation, it is argued that bilateral trade is positively influenced by the size of the economy of the trading partners while being restricted by the geographical distance between the two nations. The negative impact of distance could be explained by the logistics-related costs arising due to the transportation, forwarding and storage activities from place of departure to place of destination. For regression models, the original form was transformed to a natural logarithmic function as follows:

Where ln represents the natural logarithm of attached variables and ε_ijt is the residuals of the estimation, while β₀, β₁, β₂, and β₃ are the coefficients of the included regressors. Since the introduction of gravity model, a large number of scholars have been employing this model to analyze the antecedents of bilateral trade flows between economies. The incorporation of several variables reflecting national characteristics results in augmented trade gravity model which allows quantitative analysis of policy-related issues (Kabir et al. 2017; Yean and Yi 2014). Regarding economic performance, Yean and Yi (2014) have argued that income per capita should be considered for the estimation rather than gross income as it more effectively reflects national purchasing power and prosperity, which is also included in the models of Natale et al. (2015).

A set of variables have been incorporated into the augmented trade gravity model, which includes linguistic similarity (Bensassi et al. 2015; Gani 2017; Kahouli 2016; Kahouli and Maktouf 2015; Kahouli and Omri 2017; Kuik et al. 2019; Martí and Puertas 2017; Martí et al. 2014; Puertas et al. 2014; Yean and Yi 2014), continental location (Kahouli 2016; Kahouli and Maktouf 2015; Kuik et al. 2019; Martí and Puertas 2017; Martí et al. 2014; Puertas et al. 2014), exchange rate fluctuations (Bui and Chen 2017; Kahouli 2016; Kahouli and Maktouf 2015; Kahouli and Omri 2017; Narayan and Nguyen 2016; Nasrullah et al. 2020), geographical advantages and proximity (Bensassi et al. 2015; Gani 2017; Kahouli 2016; Kahouli and Maktouf 2015; Kahouli and Omri 2017; Leng et al. 2020; Martí and Puertas 2017; Martí et al. 2014; Puertas et al. 2014), market access (Bensassi et al. 2015; Besedeš and Cole 2017; Gani 2017; Kahouli and Maktouf 2015; Narayan and Nguyen 2016), trade barriers (Bensassi et al. 2015; Besedeš and Cole 2017; Gani 2017), population (Bui and Chen 2017; Kahouli 2016; Kahouli and Maktouf 2015; Kahouli and Omri 2017; Liu et al. 2016; Martí et al. 2014), environmental deterioration (Duarte et al. 2018; Kahouli and Omri 2017), regional economic integration (Carrère 2006; Gani 2017; Kahouli 2016; Kahouli and Maktouf 2015; Kahouli and Omri 2017; Kuik et al. 2019; Narayan and Nguyen 2016; Natale et al. 2015), foreign investment (Kahouli and Omri 2017; Liu et al. 2016; Nasrullah et al. 2020; Wang et al. 2010; Yean and Yi 2014), logistical capacity (Bensassi et al. 2015; Bottasso et al. 2018; Gani 2017; Martí and Puertas 2017; Martí et al. 2014; Puertas et al. 2014), special events like global financial crisis (Kahouli and Maktouf 2015), and factor endowments including labor and technology (Liu et al. 2016; Wang et al. 2010). In addition, as included in the model by Kahouli (2016), Kahouli and Maktouf (2015), Kahouli and Omri (2017), and Tham et al. (2018), similarity of economic size and income per capita difference between exporting and importing country are arguably determinants of bilateral trade flows. Especially, the latter was originally recommended by Linder (1961) which is known as Linder hypothesis or Linder effect, which argues that countries of similar income-levels share common consumer preferences which results in increased bilateral trade flows.

Prior to the emergence of Logistics Performance Index (LPI), a global logistics performance index has been introduced and involved in the augmented trade gravity model, which measure the time, cost, complexity and risks of logistics-related activities (Hausman 2004; Hausman et al. 2013; Lee and Whang 2005). Those logistics-related aspects are measured by surveyed data but not based on 5-point scale like LPI and its sub-indicators. Thus, as could be seen in Table 1, a limited number of recent scholarly studies have been included LPI and its components in the augmented trade gravity model. Sectoral data appears to be absent from relevant literature while this may be more insightful for policy-makers to align resources with the national plan on export-oriented sectors.

The case of individual emerging economy is not concerned especially that of Vietnam, which are only included in the Far East group by Martí et al. (2014) and Martí and Puertas (2017). Most of the studies employ Heckman selection model with two-step procedure as research methodology (Martí and Puertas 2017; Martí et al. 2014; Puertas et al. 2014) to solve the zero-trade issue recorded in trading relationship of some partners at specific time. Findings are inconclusive, which vary by geographical regions, product groups and income levels (Çelebi 2019; Martí and Puertas 2017; Zaninović et al. 2021). Nevertheless, those studies reveal a positive impact of logistics performance on bilateral trade or export which needs to be further tested in future research. Noticeably, most recent research has been increasingly concerning the sectoral evidence of the trade gravity model and employing the Poisson Pseudo-Maximum Likelihood (PPML) with a variety of fixed effects (FE) included to tackle the zero trade and endogeneity issues (Taguchi and Thet 2021; Zaninović et al. 2021).

Panel data is considered to be the most appropriate for testing the trade gravity model, which allows the observation in both spatial and temporal aspects. This means that panel data allows examining the heterogeneity of a sample of selected economies. To analyze static panel data, several recent studies have employed FE and random effect (RE) models, which subsequently uses Hausman test to determine which model would be retained (Leng et al. 2020; Nasrullah et al. 2020; Tham et al. 2018; Yean and Yi 2014). Previously, trade gravity model was applied to a single exporting country, which has been recently modified by including the reverse trade flow to involve the multilateral resistance effect in the proposed models. This results in the inclusion of importer-fixed and exporter-FEs in the estimation. However, to consider the export of a single economy, only the importer-fixed, year-fixed and importer-year FEs could be included in the estimation.

In terms of dynamic panel data estimation, a large amount of research has employed the difference and system Generalized Method of Moments (GMM) models, which are arguably the solution to solving the endogeneity in the trade gravity model (Kahouli 2016; Kahouli and Omri 2017; Tham et al. 2018). Several studies have considered the income and difference, economic similarity, exchange rate, and dummy variables of free trade agreements to be the endogenous variables in the trade gravity model. In addition to GMM models, some research uses Hausman-Taylor (HT) model to treat the endogeneity (Kahouli and Omri 2017). To deal with the zero trade issue, some research uses the Pooled Ordinary Least Squares (POLS), PPML and Heckman Sample Selection estimation (Kuik et al. 2019; Natale et al. 2015; Taguchi and Thet 2021; Zaninović et al. 2021).

Previous research has been increasingly using the gravity model to examine determinants of export growth (Jagdambe and Kannan 2020; Leng et al. 2020; Narayan and Nguyen 2016; Nasrullah et al. 2020; Natale et al. 2015). Among those, the link between export growth and logistics performance has received a limited amount of scholarly concern despite the fact that logistics-related advantages are the influencers of export competitiveness (Taguchi and Thet 2021; Zaninović et al. 2021). In addition, empirical evidence of sectoral and sub-sectoral levels is not adequate to extend the applicability of trade gravity model (Kahouli 2016). Thus, this paper examines the validity of the logistics-augmented trade gravity model in seafood export industry of Vietnam with sectoral and sub-sectoral data. The impacts of a number of RTAs signed by Vietnam and trading partners are included in the proposed models. The zero trade and endogeneity concerns in trade gravity model are addressed in this paper, which provides more robust results. Policy implications for the interlink between seafood export, logistics performance, and regional economic integration are discussed to enhance the practicality of the findings.

This paper proposes three separate models with the incorporation of aggregated (or overall) LPI and six sub-indicators as regressors and export growth of seafood products (aggregated value of HS 03 and HS 16), HS 03 and HS 16 (aggregated value of 1603, 1604, and 1605) are included as dependent variables. Three proposed models are presented as follows.

where ln indicates the natural logarithm of attached variables, i is the exporting country (i.e., Vietnam), j is Vietnam’s seafood importing countries (j = 1, … 96), t is the year of observation (t = 2007, 2010, 2012, 2014, 2016, and 2018 in accordance with the years of LPI collection and calculation), β₀ is the y-intercept which reflects cross-country variations in the estimation, β_1→11 represents the parameters which indicate the association between dependent and independent variables, ε_ijt is the error term, and μ_t indicates the time-FE included in the estimation. EXP_ijt, HS03_ijt, and HS16_ijt are the export revenue of Vietnam to country j in year t (in thousand US$) of aggregated seafood, HS03, and HS16 products, respectively. GDPpc_it and GDPpc_jt are the per capita GDP in year t (in thousand US$/person) of Vietnam and country j, respectively. DIST_ij is the geographical distance (in km) between Vietnam and country j. AD_GDPpc_ijt is the absolute value of annual difference of income per capita of Vietnam and country j in year t (in thousand US$/person). SIML_ijt is the annual similarity of economic size of Vietnam and country j in year t. EXG_jit is the annual bilateral exchange rate (local currency unit of Vietnam per local currency unit of country j) in year t. OLPI_it and OLPI_jt are the overall logistics performance index of Vietnam and country j in year t, respectively. RTA_ijt is the dummy variable which equals 1 if both Vietnam and country j are members to a specific RTA (please see the list in Table 2) in year t otherwise equals 0. LAND_j is the dummy variable which equals 1 if country j is landlocked otherwise equals 0. BORD_ij is the dummy variable which equals 1 if Vietnam and country j share a common land border otherwise equals 0.

The three models are used to investigate the central linkage between seafood export, logistics performance, and regional economic integration. Model (1) uses the aggregated (sectoral) seafood export as the dependent variable, while Model (2) and Model (3) use the disaggregated (sub-sectoral) export of HS03 and HS16 products, respectively. The detailed interpretation of incorporated variables is presented in Table 2.

Pearson’s correlation test is used to test the linear relationship between the independent and the dependent variable. The paper conducts this test to predict the correlation between the variables in the sample. Pearson’s correlation coefficient takes a specific value from +1 to −1. Regarding the multi-collinearity in the POLS, VIF index is considered. If VIF index is less than 10, the multi-collinearity in the model is inconsiderable. In terms of heteroskedasticity, Breusch-Pagan test (Breusch and Pagan 1979) and White test (White 1980) will be conducted for POLS estimation of proposed models. If the p-value of the tests is less than 0.05 (significance level is 5%), the hypothesis H₀ is rejected which confirms the heteroskedasticity of the models. Prior to regression, Levin-Lin-Chu (Levin et al. 2002), Harris-Tzavalis (Harris and Tzavalis 1999), and Breitung (Breitung and Das 2005) tests are applied for variables with balanced data, while Fisher-type tests (Choi 2001) are used for those with unbalanced data. Those tests are employed for examining the stationarity of included variables. The p-value of the tests is expected to be less than 0.05 (significance level is 5%) to reject the hypothesis H₀ which confirms the appropriateness of the data set before regression.

To test the proposed linkages between seafood export, logistics performance, and regional economic integration, this paper employs three groups of analytical techniques. The first group includes the baseline models, which are applied to static panel data estimation, namely FE and RE models. Time-FE is also included in the estimation. The baseline models with fixed and RE are employed by Kahouli (2016), Kahouli and Maktouf (2015), Leng et al. (2020), and Tham et al. (2018) to include national characteristics in the estimation. Meanwhile, models in the second group are employed to examine the dynamic panel data estimation, namely difference and system GMM models with one-step and two-step option. GMM estimation is employed to solve the endogeneity as employed by Kahouli (2016), Kahouli and Maktouf (2015), and Tham et al. (2018), which allows the examination of time-invariant variables in the trade gravity model. The third group is used to deal with the zero trade issue, which includes POLS, PPML, and Heckman Sample Selection models as employed by Jagdambe and Kannan (2020), Khurana and Nauriyal (2017), and Shahriar et al. (2019). All the groups require distinctive tests to confirm their validity, which would also be employed and presented in the results. Only the results of those valid models are discussed and explained.

The stable long-term relationship between dependent and independent variables are considered using the panel cointegration tests of Kao (MDFt, DFt, ADFt, UmDFt, and UDFt) (Kao 1999), Pedroni (MPPt, PPt, and ADFt) (Pedroni 1999), and Westerlund (Variance ratio) (Westerlund 2005) (including and excluding time trend). Accordingly, included variables have a stable relationship in the long run if p-value is less than 0.05. Those tests are performed after stationarity tests have been conducted which indicate relevant variables is not stationary. The multivariate regression for the dataset can then be replaced by a panel cointegration test. In addition, the panel cointegration test can be used to test the robustness of the regression results (which is performed after the regression results confirm the proposed hypotheses).

Data is sourced from international organizations, namely WTO, World Bank (WB), and United Nations Conference on Trade and Development (UNCTAD) in 2007, 2010, 2012, 2014, 2016, and 2018 based on the availability of biennial LPI and its dimensions collected by WB (Table 2). Income per capita difference and economic similarity are calculated by the author based on previous research. Meanwhile, geographical distance and landlockedness are established based on information of online websites. A panel of selected 96 Vietnam’s trading partners is considered. In total, the dataset includes a number of 576 observations (6 × 96) for the strongly balanced panel data analysis.

Table 3 indicates descriptive statistics of the data. The minimum values of seafood export variables EXP_ijt, HS03_ijt, and HS16_ijt are 0, which indicates the possible existence of non-randomly distributed zero trade issue in the panel data. The mean value of logistics performance of Vietnam, specifically the variable OLPI_it is approximately 3.04 out of 5.00, which implies that much needs to be considered to enhance this indicator. Meanwhile, there is insignificant difference between Vietnam and its trading partners in logistics performance as the mean value of their logistics performance, specifically the variable OLPI_jt is about 3.10.

Pearson’s correlation tests are subsequently conducted for the natural logarithm of included variables at the significance level of 5%. Results confirm the significant correlations between seafood export variables and independent determinants. In addition, income per capita, income difference, and exchange fluctuations are significantly correlated with many other independent variables, which reveals the threat of multi-collinearity. Table 4 indicates the result of panel unit root tests for all of the included variables in the three proposed models. At the significance level of 5%, all of the null hypotheses are rejected which confirms the stationarity of considered variables.

The first group of analytical techniques is applied for testing the proposed models whose results are presented in Table 5. Model (1) includes Column (1), (2), and (3), Model (2) includes Column (4), (5), and (6), and Model (3) includes Column (7), (8), and (9). The Hausman tests indicate that the FE estimations are more appropriate than the RE estimations. The F tests validate the use of FE estimation against the POLS. Therefore, the results of Column (1), (2), (4), (5), (7), and (8) are appropriate for the discussion. In the Column (1), (2), (4), (5), (7), and (8), findings confirm the positive links between income per capita of exporting and importing countries, which are represented by variables lnGDPpc_it and lnGDPpc_jt, and the seafood export at sectoral and sub-sectoral levels, which are represented by variables lnEXP_ijt, lnHS03_ijt and lnHS16_ijt. This may reveal that foreign markets of high-income level are potential for Vietnam’s exported seafood products because their demand is proportionate to their consumption budget which is considerably determined by the income level.

In terms of national logistics performance, findings in Column (1) and (4) indicate that the positive impact of importing country’s logistics performance is confirmed only in the cases of aggregated seafood and HS03 export with the coefficients of 1.242 and 1.517, respectively. Meanwhile, exporting country’s logistics performance does not exert significant impact on seafood export at all levels. This may be explicable in the case of Vietnamese exporting enterprises because transport- and logistics-related obligations are frequently undertaken by the importing partners in a contract with higher logistics management capabilities, which results in the sole significant impact of importing country’s logistics performance on seafood export of Vietnam.

Furthermore, the increase of bilateral exchange rate, which is represented by the variable lnEXG_jit, seems to boost seafood export at aggregated and disaggregated level in Column (1), (2), (4), (7), and (8). This may imply that the relative depreciation of Vietnamese domestic currency would result in the seafood export price decrease in foreign markets, which ultimately enhances seafood export revenue. Similarly, difference in income per capita, which is represented by the variable lnAD_GDPpc_ijt, appears to enhance aggregated seafood and HS03 export in Column (1), (2), (4), and (5). This result indicates that international labor division exists in the case of Vietnamese seafood industry. Specifically, income per capita is in most of the cases positively associated with labor productivity. Low and middle income-level countries are frequently capital-scarce and labor-abundant, which consequently enables the specialization in labor-intensive industries. That logic may apply to the situation of Vietnam’s seafood industry.

Regarding regional economic integration which is represented by RTA-related dummy variables, the VCFTA appears to increase export growth of aggregated seafood, HS03 and HS16 products in Column (1), (2), (5), (7), and (8), while the AIFTA increases aggregated seafood and HS03 export in Column (1), (2), (4), and (5). Noticeably, the VNEAEU appears to decrease export of HS16 products in Column (7) and (8). Interestingly, the AANZFTA negatively influences the growth of aggregated seafood and HS03 export in Column (1), (2), (4), and (5), and significantly boosts HS16 export in Column (8). Overall, findings indicate that joining a RTA may not necessarily facilitate immediate gains for members. In addition, region-to-country RTAs such as AANZFTA, which include more than two members, could increase the intra-bloc competition in exporting activities because the market access is equally available for all of the members. Furthermore, the impact of a RTA on Vietnam’s seafood export may be two-fold, which depends on specific sectoral and sub-sectoral products.

The second group of analytical techniques includes difference and system GMM estimation with one-step and two-step procedures, which solves the endogeneity in proposed models. As suggested by Kahouli (2016), income, difference in income, economic similarity, and exchange rate are identified as endogenous variables whose lagged values are significant instruments. The results of difference and system GMM (D-GMM and S-GMM) with one-step (I) and two-step (II) procedures and relevant tests are presented in Table 6. Model (1) includes Column (10), (11), (12), and (13), Model (2) includes Column (14), (15), (16), and (17), and Model (3) includes Column (18), (19), (20), and (21).

As could be seen in Table 6, the p-value of Sargan tests for D-GMM (I) and S-GMM (I) of three models is less than 0.05, which reject the null hypothesis of restrictions on over-identifications. Thus, GMM models with one-step procedure are not appropriate for the estimation of three models. Meanwhile, the p-values of Hansen tests and Arellano-Bond tests (AR2) of D- GMM (II) and S-GMM (II) estimations are greater than 0.05, which confirms the appropriateness of GMM models with two-step procedure for the three proposed models. Therefore, the results of Column (11), (13), (15), (17), (19), and (21) are appropriate for the discussion.

Noticeably, findings only confirm the positive relationship between HS16 export and exporting and importing country’s income per capita, which are represented by the variables lnGDPpc_it and lnGDPpc_jt respectively, with D-GMM (II) estimation in Column (19). In the other estimations of Column (11), (13), (15), and (17), only the negative links between exporting country’s income per capita and seafood export are valid, which may reveal that as economic development is improved, seafood export tends to be reduced. Finding indicate that when the control of endogeneity is employed the enhanced national income level may reduce the seafood export of Vietnam. The increase of income level may result in the corresponding increase of capital endowment, which consequently decreases labor-intensive industries which include seafood industry. This may imply the inclusion of export-oriented maritime policy in Vietnam’s sustainable development framework to maintain export growth. From the supply-side perspective, income per capita of importing country does not influence seafood export except the case of S-GMM (II) estimation for aggregated seafood export and D-GMM (II) estimation for HS16 export in Column (13) and (19). This may imply that in the endogeneity-controlled estimation the positive impact of income level of importing country on Vietnam’s seafood export varies by data levels and product groups.

Interestingly, findings of Column (13), (17), and (21) reveal that geographical distance, which is represented by the variable lnDIST_ij, and importing country’s landlockedness, which is represented by the variable LAND_j, are not the inhibitors of seafood export at all levels. When controlling the endogeneity, those results may imply the expanding and potential global market for Vietnam’s seafood export despite geographical disadvantages relevant to trading partners. The transport- and logistics-related costs are offset by the benefits of increased Vietnamese seafood consumption of foreign markets. The economic similarity between exporting and importing countries, which is represented by the variable lnSIML_ijt, is confirmed to increase aggregated seafood and HS03 export in the D-GMM (II) and S-GMM (II) estimation in Column (11), (13), (15), and (17). This result may unveil that to some extent the similar economic size would shape similar patterns of consumption, especially in the case of food-related industries which include the seafood sector. Therefore, potential markets of Vietnam’s seafood products may be the economies of similar size and rapid growth. Meanwhile, the impact of difference in income per capita, which is represented by the variable lnAD_GDPpc_ijt, on aggregated seafood, HS03, and HS16 export in the D-GMM (II) estimation appears to be significantly positive in Column (11), (15), and (19). This result is in accordance with the baseline estimation. The results may support the North-South trade pattern of agricultural and aquatic products. In the case of bilateral exchange rate, which is represented by the variable lnEXG_jit, findings of the S-GMM (II) estimations in Column (13) and (21) indicate the negative impact on aggregated seafood and HS16 export. Nevertheless, the positive impact on aggregated seafood export is confirmed in Column (11) with D-GMM (II) estimation. Thus, the impact of bilateral exchange rate on Vietnam’s seafood export is inconclusive in the endogeneity-controlled estimations. Nevertheless, the inconclusiveness may to some extent imply that trade-related monetary measures appear to be volatile and unpredictable, and could only generate benefits in a short-term period.

In terms of regional economic integration which is represented by RTA-related dummy variables, the impact of VNEAEU on seafood export varies by data levels and product groups. Specifically, Column (11) and (15) with D-GMM (II) estimations confirm the positive impact while Column (19) and (21) with D-GMM (II) and S-GMM (II) estimations indicates the negative relationship. Similarly, the positive impact of the AIFTA on aggregated seafood and HS03 export is confirmed in Column (13) and (17) with S-GMM (II) estimation, whereas the negative relationship with HS16 export is significant in Column (19) with D-GMM (II) estimation. With regard to the AANZFTA, its negative relationship with aggregated seafood and HS03 export is significant in Column (13) and (17) with S-GMM (II) estimations, whereas Column (19) with D-GMM (II) estimation demonstrates the positive linkage with HS16 export. With regard to the VCFTA, its positive impact on HS16 export is confirmed with D-GMM (II) estimation in Column (19). Those results unveil the two-fold effect of RTA membership in the case of Vietnam’s seafood export. Accordingly, sub-sectoral and product-specific evidence should be considered when assessing the impact of RTAs on export growth because the effect of RTAs varies by data levels and product groups.

In terms of national logistics performance, it appears that exporting country’s logistics performance, which is represented by the variable lnOLPI_it, is positively associated with seafood export in all proposed models with D-GMM (II) and S-GMM (II) estimations in Column (11), (13), (15), (17), (19), and (21). In the case of importing country’s logistics performance which is represented by the variable lnOLPI_jt, the negative relationship is significant in Column (11) of aggregated seafood export with D-GMM (II) estimation, while the positive relationship is confirmed in Column (21) of HS16 export with S-GMM (II) estimation. Interestingly, in the endogeneity-controlled estimation, it turns out that the national logistics performance could enhance Vietnam’s seafood export. However, the impact of importing country’s logistics performance on seafood export is inconclusive and unpredictable.

The third group of analytical techniques includes POLS, PPML and Heckman Sample Selection estimations, which solve the non-randomly distributed zero trade issue in empirical trade studies. In the Heckman Sample Selection estimation, the paper includes income per capita, difference in income per capita, economic similarity, geographical distance, and membership to RTAs (Table 2) as independent variables for the selection model. Model (1) includes Column (22), (23), and (24), Model (2) includes Column (25), (26), and (27), and Model (3) includes Column (28), (29), and (30). The LR tests for independent equations in Table 7 demonstrate that the Heckman Sample Selection estimations are more appropriate the POLS (p-value < 0.05). In addition, the results of Breusch-Pagan test (Breusch and Pagan 1979) and White test (White 1980) for POLS estimation of the three models demonstrate that the multi-collinearity and heteroskedasticity may exist. Therefore, the results of Column (23), (24), (26), (27), (29), and (30) are appropriate for the discussion.

Table 7 indicates the positive relationship between seafood export at all levels and income per capita of exporting and importing country which are represented by the variables lnGDPpc_it and lnGDPpc_jt respectively, except Column (27) with the income per capita of the exporting country. This is consistent with findings of the baseline models when not considering the zero trade issue in the estimation. In addition, the geographical distance and importing country’s landlockedness, which are represented by the variables lnDIST_ij and LAND_j respectively, are detrimental to aggregated seafood, HS03 and HS16 export in Column (23), (24), (26), (27), (29), and (30). Therefore, in the zero trade-controlled estimation, geographical disadvantages appear to hamper Vietnam’s seafood export, which is not concluded in the endogeneity-controlled estimation. This result may imply that the transport- and logistics-related costs, which are proportional to geographical distance and landlockedness, could decrease Vietnam’s seafood export.

Regarding difference in income per capita which is represented by the variable lnAD_GDPpc_ijt, findings in Column (29) with PPML estimation confirm the negative relationship with HS16 export. This result is inconsistent with the baseline and endogeneity-controlled models. However, the significant negative impact may unveil that Vietnam’s HS16 export is specifically attractive to foreign consumers of the similar income level. In terms of economic similarity which is represented by the variable lnSIML_ijt, all of the models with PPML and Heckman Sample Selection estimations in Column (23), (24), (26), (27), (29), and (30) demonstrate its negative impacts on seafood export at all levels. These results are almost contrary to those of the endogeneity-controlled models. Nevertheless, reasons for the significant negative relationship may be practically two-fold. Firstly, Vietnam’s seafood products are attractive to foreign consumers of larger economies. Secondly, Vietnam’s exported seafood products are the input of a regional or global value chain which is mainly located in larger countries. In terms of bilateral exchange rate which is represented by the variable lnEXG_jit, findings do not confirm the significant relationship with seafood export at all models. These results are contrary to those of the baseline and endogeneity-controlled models. Nevertheless, these findings may imply that trade-related monetary measures for boosting export are ineffective in the case of Vietnam’s seafood industry.

In terms of regional economic integration which is represented by RTA-related dummy variables, the ATIGA appears to hamper export growth of aggregated seafood and HS03 export in Column (23), (24), (26), and (27). The VNEAEU and ACFTA positively influence HS03 export in Column (26) and (27), while significantly decrease HS16 export in Column (29) and (30). In all of the models, the VJEPA and AKFTA seem to enhance seafood export at all levels. Finally, the AANZFTA is only conducive to HS16 export in Column (29) and (30). These findings are consistent with those of the baseline and endogeneity-controlled models. Overall, the impact of RTAs on seafood export varies by estimation methods, data levels, and product groups. In terms of national logistics performance which is represented by the variables lnOLPI_it and lnOLPI_jt, all of the models with the Heckman Sample Selection estimation confirm the positive relationship between seafood export and national logistics performance of importing and exporting country. These findings are inconsistent with those of the baseline and endogeneity-controlled models. Nevertheless, the significant positive relationship unveils the role of logistics performance in enhancing the export growth of Vietnam’s seafood products.

For robustness check, panel cointegration tests have been employed to examine the stable long-term relationship between seafood export (aggregated and disaggregated levels), national logistics performance (supply-side and demand-side approaches), and regional economic integration (membership to any RTA) (Table 8). Findings demonstrate the significant stable long-term relationship between relevant variables as almost all the tests (09 tests in total for each combination of relevant variables) appears to be significant (0.01 < p-value < 0.1), which reject the hypotheses of no co-integration in panels. Despite the fact that those variables may not be significantly associated in distinct models and estimation methods, Table 8 reveals the interconnectedness of seafood export, national logistics performance, and regional economic integration in the long run, which to some extent may lend support to the comprehensive national plan of international economic integration, logistics development and export-led marine economy.