Impact of Hong Kong-ASEAN Free Trade Agreement: an assessment from the trade creation and trade diversion perspectives

Abstract

This study aims to investigate the effects of the ASEAN-Hong Kong Free Trade Agreement (AHKFTA) on exports, concentrating on the trade creation (TC) and trade diversion (TD) effects. Most previous studies evaluated the TC and TD effects of the agreements over a long period, ignoring the use of a short period. This empirical study compares the TC and TD effects of the AHKFTA between 2003-2021 and 2018-2021 to see if there is a difference in the degree of influence. This paper uses panel data covering 34 countries: Hong Kong, a territory Hong Kong’s 23 largest partners, and 10 ASEAN countries. To ensure the certainty to accurately capturing the impact of the factors in the gravity model, the study estimates the regression models for panel data by OLS, Random Effect (RE), Fixed Effect (FE), Hausman- Taylor and LASSO methods. Furthermore, the PPML method is also implemented to give a more solid basis for the conclusions. The research found that the current trade policy between Hong Kong and ASEAN is in good performance as it promotes both AHKFTA’s intra-regional trade growth and extra-bloc countries.

Keywords:

• AHKFTA
• TC
• TD
• gravity model
• PPML
• LASSO

REVIEWING EDITOR:

• Geng Yuqing, Shanghai Dianji University, China

SUBJECTS:

• Asian Economics
• Asian Development
• Economics

1. Introduction

In the past two decades, ASEAN has accelerated the pace of trade discussions and added key bilateral partners into ASEAN Free Trade Agreements (i.e. China and ASEAN signed in 2004, Korea and ASEAN in 2007, Japan and ASEAN in 2008, Australia and New Zealand in 2010, India in 2010, and most recently ASEAN and Hong Kong (RTAs database, 2023). The FTA between ASEAN and Hong Kong is the group’s sixth agreement with an external trading partner. Entering into force on 11 June 2019, this agreement marked the commitment of ASEAN and Hong Kong to deepen economic ties. From there, it is expected to promote trade and investment in the AHKFTA region and expand partnerships through economic cooperation activities. It can be said that Hong Kong is one of the region’s strategic trade and logistics centers (W. Zou et al., 2023). However, implementing a series of trade agreements in a short time in ASEAN has led to the question of whether creating a larger free trade area will bring more significant international trade to this region.

According to the traditional customs union resource reallocation after forming an FTA can lead to TC and TD effects. TC (the substitution of low-cost imports for high-cost indigenous production) enhances welfare. In contrast, TD (changing low-cost global imports with high-priced imports from free trade program members) is likely to harm welfare. The relative size of TC and TD effect then determines the net welfare effect. The TD effect of FTAs under customs union theory of Viner has been debated for a long time. Trade preferences make it possible for higher-cost producers in FTA member countries to eliminate efficient third-party producers, leading to TD and practical losses. On the other hand, the trade-generating benefits of an FTA will depend on the initial economic structure (Burfisher et al., 2001). Therefore, the extent of these TC and TD effects is still an empirically open question. Moreover, the stronger the institutional and economic reforms, the more benefits a country will receive.

Recent studies analyzed the impact of free trade agreements (FTAs) on trade flows focused on TC and TD effects (Martínez-Zarzoso et al., 2009; Repositori et al., 2014; Singh, 2021; Kashif & Akram, 2021). FTAs encourage overall trade volume between intra-bloc members as well as between intra-bloc and extra-bloc countries by cutting down on tariff barriers and non-barriers (Repositori et al., 2014; Singh, 2021). Previous research on the impact of FTAs produced mixed results, indicating that there is no general rule: FTAs can result in either TC or TD effects (Gharleghi & Shafighi, 2020). While the majority of empirical studies have focused on assessing the TC and TD effects of specific FTAs, such as the ASEAN-China FTA (Yang & Martinez-Zarzoso, 2014); the ASEAN-India FTA (Singh, 2021); and the ASEAN-Korea FTA (Mareta, 2018), previous studies have overlooked assessing the TC and TD effects of the ASEAN-Hong Kong FTA.

In addition to mixed results on an assessment from the TC and TD of FTAs, previous studies also led to conflicting results on the sign and the magnitude of the estimated coefficients of other determinations on international trade. The cause of the above problem may be that the researchers did not correctly account for the uncertainty of the models, which is one of the open-ended problems of economic theory (Brock & Durlauf, 2001). Especially in today’s context, data appears in a wide variety of forms, especially high-dimensional datasets or variables that exhibit co-integration and non-stationarity. This creates a difficult challenge that requires the use of a proper model, adequate explanatory factors, and lags that affect the dependent variable. According to Fan et al. (2007), if the sample size n is less than the parameter - dimensions, the OLS approach may result in the issue of pseudo-correlation (similar distributions and strong correlation) among the explanatory variables. Additionally, it is claimed that other problems with OLS-based models include the resilience of statistical estimators and the efficiency of statistical reasoning techniques. To overcome the above problems, RIDGE, LASSO, and ELASTIC NET methods based on shrinking coefficients have been raised. Tibshirani (1996) proposed the LASSO method, and then it has been widely used in model selection for general linear regression models (Fan & Peng, 2004; Nardi & Rinaldo, 2008; Zhao & Edu, 2006; H. Zou, 2006). The LASSO approach, which minimizes the sum of squares of the residuals with the restriction that the sum of the absolute of the coefficients is constrained by a constant, is based on the assumption that some explanatory factors in the regression are inconsequential. As a result, LASSO’s highlight is its ability to successfully and simultaneously perform model selection and estimation (Hastie et al., 2015).

The contributions of this study to the literature are three-fold. First, our study is distinct from previous literature in terms of the scope by considering ASEAN- Hong Kong Free Trade Agreement which had received less or no attention earlier. Second, as opposed to the previous which have evaluated the impact of FTAs on trade flows without dividing into subsamples to detect any changes in trade specialization for members in FTAs, this study specifically investigates an assessment from the TC and TD effects of AHKFTA by dividing into subsample (2018-2020) to compare the degree of TC and TD effects of full sample (2010-2020). Third, in addition to employing the conventional panel data methods (OLS, FE, RE, Hausman-Taylor), we explore methodologically by simultaneously applying PPML and LASSO regression approaches to investigate the effects on research variables.

The rest of the article is organized as follows. The first section of the article introduces the research motivation and clarifies the research gap in exploring how FTA benefits member countries. The second section presents the literature review, and the third discusses the gravity model. The main results and discussion are explained in the fifth section, followed by the fourth section on empirical models and data. The last section concludes the findings of the research.

2. Literature review

In the examination of the impact of Free Trade Agreements (FTAs) on trade, analysts commonly incorporate Viner’s concepts of Trade Creation (TC) and Trade Diversion (TD). TC occurs when reduced internal trade restrictions lead to the emergence of new trade between member countries. This shift in trade dynamics involves a change in the origin of a product, moving from an intra-bloc producer with higher resource costs to another intra-bloc producer with reduced resource costs. On the other hand, TD leads to resource misallocation, negatively affecting welfare. While TC brings positive benefits, changes in trade patterns among member countries and between member and non-member countries stem from the interplay of TC and TD.

The empirical literature on the effects of FTAs on trade flows is divided into two major strands. The first strand is case studies, which typically focus on evaluating a single FTA. Using aggregated and disaggregated data, Yang and Martinez-Zarzoso (2014) shed light on that the China-ASEAN FTA (ACFTA) had a positive trade effect. Notably, industrial products and chemical products have greater TC effects than agricultural products, machinery, and transportation equipment. In a different study, Koh et al. (2017) examine how the ASEAN Free Trade Agreement (AFTA) has affected the bilateral manufacturing trade between the 10 ASEAN member countries and its 39 trading partners. In terms of exports, their findings indicated that AFTA has had purely TC effects. Furthermore, AFTA has had more excellent TC effects in imports than import diversion effects. Khurana and Nauriyal (2017) found that the India-ASEAN FTA (AIFTA) has lowered exports between members, leading to pure TD. They did this by using aggregate data to estimate the effects of the AIFTA on export flows. Evaluating the impact of the AIFTA on export flows using aggregate data. These findings are in line with Singh (2021), who examined the impact of the AIFTA on trade and discovered that AIFTA creates trade in total bilateral trade in both exports and imports. Moreover, he found that AIFTA has caused the import creation effect greater than the export creation effect.

The second strand of empirical literature examines the effects of multi-FTAs TC and TD. Darma and Hastiadi (2018) investigated the effect of TC and TD on the export of Indonesian food and beverage industry products with trading partners. Their findings found that the export of goods from the Indonesian food and beverage industry is positively impacted by the TC and TD effect thanks to ACFTA, ASEAN-Korea FTA (AKFTA), and AIFTA. This indicates that the coming into force of ACFTA, AKFTA, and AIFTA has a TC effect by promoting intra-regional trade among member countries of these agreements while discouraging TD from non-member countries. Using panel data from 1996 to 2017, Timsina and Culas (2022) evaluate the TC and TD effects of major Australian free trade agreements (FTAs). This study discovered that intra-block export diversion is less significant than the overall TC effect in wheat trade.

Furthermore, the level of TC and TD effects of each FTA are different. Taguchi and Lee (2016) demonstrated that ACFTA’s TC effect was significantly bigger than that of AKFTA and AJFTA, and all three FTA’s TD effects were typically negative. Furthermore, Singh (2021) found that AIFTA generates exports and imports for all bilateral trade, with import generation having a bigger impact than export creation. In the AANZFTA, Gharleghi and Shafighi (2020) found TD from non-members and a lack of TD among its members.

Although the measurement methods and FTAs are pretty different and conclusions also differ slightly, most of the research results demonstrate that FTAs lead to TC and TD effects (detailed Table 1).

Table 1. Summary of latest literature review.

Study	Period	Study object	FTA/Regions	Method	Main findings
Khurana and Nauriyal (2017)	2000–2015	33 countries (11 countries of ASEAN, India, India’s top trading partners	India-ASEAN FTA	OLS, PPML, only time effects, time and country specific effects	FTA has led to rise trade diversion effects.
Timsina and Culas (2022)	1996–2017	Wheat export of Australia	Australia’s FTA	OLS, PPML	Trade creation effects is higher than Intra block export diversion.
Gharleghi and ShaFigurehi (2020)	2000–2017 2000–2007 2008–2017	Asia-Pacific region	AANZFTA AFTA APTA	2SLS, OLS	Trade has widened between AFTA and APTA members, but not between AANZFTA members.
Lateef et al. (2018)	2001–2014	110 partners countries, China, Pakistan	CPFTA	PPML	On Pakistan’s agricultural exports, CPFTA had a significant TC  effect
Koh et al. (2017)	1995–2014	10 member of ASEAN and 39 trading partners	AFTA	REM	AFTA encourages trade between ASEAN members to have  TC influences on exports and imports, as well as TD effects on imports.
Mareta (2018)	1990–2015	Indonesia and 20 trading countries partner	AKFTA	LSDV, OLS, FE	AKFTA have TC and TD effect on Indonesian export.
Taguchi (2015)	1993–2013	14 countries/regions	AKFTA AJFTA ACFTA	FEM	While the TD effects were typically negative in all three agreements, the TC effects in the ACFTA were much larger compared to the ones in the ASEAN-Korea FTA (AKFTA) and ASEAN-Japan FTA (AJFTA).
Zolin and Uprasen (2018)	1999–2015	38 countries including EU and Selected Asian countries	EU	FE-PPML, PPML, RE, FE	The TC effects in the ACFTA were significantly bigger compared to those in the AKFTA and ASEAN-Japan FTA (AJFTA), while the TD effects were typically adverse in all three agreements.
Muñoz et al. (2023)	Quanitative method (1996–2018) Qualitative method (October 2019 and March 2020)	Chile and Korea	Chile–Korea FTA	Mixed method (quanitative and qualitative method)	Due to limitations in the public and private sectors, Chile has been unable to change the level of concentration of its exports.
Geda and Yimer (2023)	1993–2017	North and South-Eastern African countries and the West and Central African countries	AfCFTA	OLS, PPML	The AfCFTA might increase intra-African exports by approximately US$ 13.7 billion (by around 19%) from an average yearly value of US$ 72.7 billion (2015–2017).
Alhassan and Payaslioğlu (2023)	2000–2018	ASEAN countries	FTAs in ASEAN	PPMLHDFE	The degree of the TC and TD effects of the FTAs on trade flows in ASEAN Depene upon the strength of he political and economic institutions in ASEAN.
(Kashif)	n.a.	ASEAN and India countries	ASEAN- India FTA	SMART	ASEAN-India FTA has resulted TD from non-member countries in favor of India.

3. The gravity model

The gravity model explains trade flows between countries of $i$ and $j$, which is favorably connected to the size of the economies and adversely related to the distance between them, which serves as a proxy for the cost of transportation. The basic gravity model with time dimension is given as: $$X_{\mathit{\text{ijt}}}=\alpha {\mathit{\text{GDP}}}_{\mathit{\text{it}}}^{{\beta }_1}{\mathit{\text{GDP}}}_{\mathit{\text{jt}}}^{{\beta }_2}{\mathit{\text{Dist}}}_{\mathit{\text{ij}}}^{{\beta }_3}{\in }_{\mathit{\text{ijt}}}$$

Applying natural logarithms, it can be expressed as: $${\mathit{\text{lnX}}}_{\mathit{\text{ijt}}}=\mathit{\text{ln}}\alpha +{\beta }_1\mathit{\text{ln}}{\mathit{\text{GDP}}}_{\mathit{\text{it}}}+{\beta }_2\mathit{\text{ln}}{\mathit{\text{GDP}}}_{\mathit{\text{jt}}}+{\beta }_3\mathit{\text{ln}}{\mathit{\text{Dist}}}_{\mathit{\text{ij}}}+\mathit{\text{ln}}{\in }_{\mathit{\text{ijt}}}$$

Where ‘$\mathit{\text{ln}}$’ denotes natural logs, $X_{\mathit{\text{ijt}}}$ denotes the bilateral trade between country i to country j in period t at the current US dollar; ${\mathit{\text{GDP}}}_{\mathit{\text{it}}}$ and ${\mathit{\text{GDP}}}_{\mathit{\text{jt}}}$ are the nominal values for GDP of the countries i and j in US dollars in period t, which represents countries’ consumption and demand level. It is anticipated that the GDP coefficients and trade flows will be positively correlated; ${\mathit{\text{Dist}}}_{\mathit{\text{ij}}}$ measures the average weighted distance between the capital cities of the countries $i$ and $j$. It is used as a proxy for transportation costs and the required delivery time; ${\in }_{\mathit{\text{ijt}}}$ is the log normally distributed error term.

The gravity model has developed as an increasingly common approach for analyzing and forecasting economic variables, notably bilateral trade flows (Kabir et al., 2017). One of the most advantages of gravity model is analyzing the specific effects of trade policies by introducing dummy variables, to indicate the existence of a regional trade agreement between countries i and j. This model, which may be extended to estimate trade creation and trade diversion, contributes significantly to the regionalism argument. Hence, the gravity model is popular in analyzing TC and TD effects of FTAs (Carrère, 2006; Endoh, 1999; Jagdambe & Kannan, 2020; Martínez-Zarzoso et al., 2009; Sun & Reed, 2010; Yang & Martinez-Zarzoso, 2014) using both cross-sectional data and panel data (Brülhart & Kelly, 1999; Lee & Oh, 2020; Nilsson, 2000; Rose & Van Wincoop, 2001).

4. Method

In addition to the basic estimation in the panel data model (OLS, FE, RE, Hausman-Taylor), this study employs the two methodologies mentioned in section Introduction: LASSO and PPML model.

4.1. LASSO regression

Multivariate linear regression with the least square method is useful only if the four conditions are satisfied. However, the data now appears in many structures and it is very hard to check those requirements. Then the LASSO method introduced in Tibshirani (1996) will correct the model appropriately by using the optimization problem $$s\left(\widehat{\beta }\right)={\displaystyle \sum }_{i=1}^n{\left(y_i-{\displaystyle \sum }_{j=1}^k{x}_{\mathit{\text{ij}}}{\beta }_j\right)}^2$$ to estimate the coefficients ${\beta }_j,j=1,k$ with constraints ${‖\widehat{\beta }‖}_1={{\displaystyle \sum }}_{j=1}^k\left|\widehat{{\beta }_j}\right|\le t$.

To solve the above optimization problem, we consider the Lagrange problem as follows: $$L\left(\widehat{\beta },\lambda \right)={\displaystyle \sum }_{i=1}^n{\left(y_i-{\displaystyle \sum }_{j=1}^k{x}_{\mathit{\text{ij}}}{\beta }_j\right)}^2+\lambda {\displaystyle \sum }_{j=1}^k\left|\widehat{{\beta }_j}\right|$$ where $\lambda$ is the Lagrange factor. According to Hastie et al. (2015), these $\lambda$ are used to adjust the model.

The fact that the ${\left|\left|\beta \right|\right|}_1\mathrm{ }$ term is not differentiable as ${\beta }_j\ne 0$ leads to a non- smooth optimization problem and therefore, a convex function in the form of $\mathit{\text{subderivative}}$ or $\mathit{\text{subgradient}}$ mentioned in Murphy (2012) is shown as $$\widehat{{\beta }_j}\left(c_j\right)=\left\{\begin{array}{c}\frac{c_j+}{a_j}\hspace{0.17em}\mathit{\text{if}}\hspace{0.17em}{c}_j<-\lambda \hspace{0.17em}\\ \frac{c_j-\lambda }{a_j}\hspace{0.17em}\mathit{\text{if}}\hspace{0.17em}{c}_j>\lambda \\ 0,\mathit{if}\hspace{0.17em}{c}_j\in \left[-\lambda ,\lambda \right]\end{array},\right.$$ where $a_j=2{\sum }_{i=1}^n{x}_{\mathit{\text{ij}}}^2$, $c_j=2{\sum }_{i=1}^n{x}_{\mathit{\text{ij}}}\left(y_j-{\beta }_{-j}^T{x}_{i,-j}\right)$. $\beta$ without component $j$ will form ${\beta }_{-j}$ and so do $X_{i,-j}$. We notice that $c_j\mathrm{ }$ is the correlation between the $j^{\mathit{\text{th}}}$ feature $X:,\mathrm{j}\mathrm{ }$ and the residual due to the other features, $r_{-j}=(Y-X:{,}_{-j}\text{*}{\beta }_{-j})$. This implies the magnitude of $c_j\mathrm{ }$ shows how relevant feature $j$ in predicting y. Shortly, the estimators can be defined as $$\widehat{{\beta }_j}=\mathit{\text{soft}}\left(\frac{c_j}{a_j},\frac{\lambda }{a_j}\right),$$

where $\mathit{\text{soft}}(a,\delta )=\mathit{\text{sign}}(a)(|a|-\delta )+\mathit{\text{and}}\mathrm{ }x+=\mathrm{\text{max}}(x,0)$ is the positive part of $x$.

By minimizing the mean square of error (MSE) using cross-validation method with given $X,Y$, (Hastie et al., 2015) gave the solution for $\lambda$; especially some regression parameters are proved to be removed as $\lambda$ is getting large.

4.2. PPML model

As it is preferred as a useful tool in structural gravity equations with multilateral resistance terms (MRTs) (Anderson & Yotov, 2016; Yang & Martinez-Zarzoso, 2014), the Poison Pseudo-Maximum Likelihood (PPML) model by Santos Silva and Tenreyro (2006) is utilized. Additionally, PPML enables the processing of zero trade flows and the generation of estimates that are more stable when both homogeneity and heterogeneity are present (Fally, 2015; Khurana & Nauriyal, 2017).

As a typical application, consider the following PPML regression: $$X_{\mathit{\text{ijt}}}=\mathrm{\text{exp}}[\mathit{\text{ln}}{T}_{\mathit{\text{it}}}+\mathit{\text{ln}}{G}_{\mathit{\text{jt}}}+\mathit{\text{ln}}{D}_{\mathit{\text{ij}}}+b\times {\mathit{\text{FTA}}}_{\mathit{\text{ijt}}}]+{ϵ}_{\mathit{\text{ijt}}}.$$

$X_{\mathit{\text{ijt}}}$ are trade flows; $i$, $j$, and $t$ are indices for export country, import country, and time. The goal is to consistently estimate the average effect of ${\mathit{\text{FTA}}}_{\mathit{\text{ijt}}}$, dummy variables for the presence of a free trade agreement on trade flows, using a ‘structural gravity’ specification. The origin-time and destination-time fixed effects—$T_{\mathit{\text{it}}}$ and $G_{\mathit{\text{jt}}}$—ensure the theoretical restrictions implied by structural gravity are satisfied. The pair fixed effect—$D_{\mathit{\text{ij}}}$—then absorbs all time-invariant pair characteristics that may be correlated with the likelihood of forming an FTA.

5. Empirical model and data

5.1. Empirical model

Consistent with the past studies on assessment from the TC and TD effects (see, e.g. (Singh, 2021; Yang & Martinez-Zarzoso, 2014), Equation (1)$$X_{\mathit{\text{ijt}}}=\alpha {\mathit{\text{GDP}}}_{\mathit{\text{it}}}^{{\beta }_1}{\mathit{\text{GDP}}}_{\mathit{\text{jt}}}^{{\beta }_2}{\mathit{\text{Dist}}}_{\mathit{\text{ij}}}^{{\beta }_3}{\in }_{\mathit{\text{ijt}}}$$ was used to measure the TC and Trade TD effects of AHKFTA, the basic augmented gravity model is given by:

$\begin{array}{l}\mathit{\text{lnE}}{X}_{\mathit{\text{ijt}}}={\beta }_0+{\beta }_1\hspace{0.17em}\mathit{\text{ln GDPit}}+2\hspace{0.17em}\mathit{\text{lnGD}}{P}_{\mathit{\text{jt}}}+{\beta }_3\hspace{0.17em}\mathit{\text{lnPOPit}}+{\beta }_4\hspace{0.17em}\mathit{\text{lnPO}}{P}_{\mathit{\text{jt}}}+{\beta }_5\hspace{0.17em}\mathit{\text{lnDIS}}{T}_{\mathit{\text{ij}}}+{\beta }_6\hspace{0.17em}\mathit{\text{lnLAN}}{G}_{\mathit{\text{it}}}+{\beta }_7\hspace{0.17em}\mathit{\text{LANDLOCKij}}\\ +{\beta }_8\hspace{0.17em}\mathit{\text{BORDERij}}+{\beta }_9\hspace{0.17em}\mathit{\text{FT}}{A}_{1\mathit{\text{ijt}}}+{\beta }_{10}\hspace{0.17em}\mathit{\text{FT}}{A}_{2\mathit{\text{ijt}}}+{\beta }_{11}\hspace{0.17em}\mathit{\text{FT}}{A}_{3\mathit{\text{ijt}}}+u_{\mathit{\text{ijt}}}\end{array}$  (1)

where: ${\mathit{\text{EX}}}_{\mathit{\text{ijt}}}$ indicates bilateral exports from exporter $i$ to importer $j$ in period $t$ at current US$. $\mathit{\text{GDPit}}$ and $\mathit{\text{GDPjt}}$ are the level of nominal gross domestic product in countries $i$ and $j$ in period $t$. ${\mathit{\text{POP}}}_{\mathit{\text{jt}}}$ and ${\mathit{\text{POP}}}_{\mathit{\text{jt}}}$ are the populations of countries $i$ and $j$ in period $t$. ${\mathit{\text{DIST}}}_{\mathit{\text{ij}}}$ is distance between the capital cities (or economic centres) of countries $i$ and $j$. Three dummy variables namely: speaking the same official language (${\mathit{\text{LANG}}}_{\mathit{\text{it}}}$), sharing common border (and being a landlocked country $(\mathit{\text{LANDLOCKij}})$ are included as regressors. $\mathit{\text{uijt}}\mathrm{ }$ is assumed to be a log-normally distributed error term.

${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$, $\mathit{\text{FT}}{A}_{2_{\mathit{\text{ijt}}}}$ and $\mathit{\text{FT}}{A}_{3_{\mathit{\text{ijt}}}}$ are dummy variables that measure the specific trade effects in the AHKFTA.

$\mathit{\text{FT}}{A}_{1_{\mathit{\text{ijt}}}}$ takes a value of 1 if both countries $i$ and $j$ in year $t$ belong to the AHKFTA and zero otherwise. A positive and statistically significant coefficient of $\mathit{\text{FT}}{A}_{1_{\mathit{\text{ijt}}}}$ demonstrates TD effects and shows that the FTA has more actively promoted intra-regional trade, which is higher than average levels of trade.

$\mathit{\text{FT}}{A}_{2_{\mathit{\text{ijt}}}}$ takes a value of one if exporter $i$ belongs to the AHKFTA in year $t$ and destination country $j$ does not and zero otherwise. A positive and statistically significant coefficient of ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ is defined as a TD effect in term of exports and shows that regional integration leads to a switch of export activities from AHKFTA member countries to non-AHKFTA member countries. Conversely, a negative and statistically significant coefficient of $\mathit{\text{FT}}{A}_{2_{\mathit{\text{ijt}}}}$ implies a decrease in exports from member countries to non-member countries and is defined as an export diversion effect.

$\mathit{\text{FT}}{A}_{3_{\mathit{\text{ijt}}}}$ takes a value of one if exporter $i$ is a non-AHKFTA member in year $t$ and destination country $j$ belongs to the AHKFTA and zero otherwise. A positive and statistically significant coefficient of $\mathit{\text{FT}}{A}_{3_{\mathit{\text{ijt}}}}$ is defined as a TD effect in terms of imports and indicates expanded imports from non-member countries to member countries. Conversely, a significantly negative $\beta 11\mathrm{ }$ indicates a TD effect in terms of imports.

5.2. Data

We use a panel dataset of 34 countries including Hong Kong, ASEAN-10 countries and Hongkong’s top 23 trading partners in 2021 (see Table 2) from 2010 to 2021 with 34*33*12 (13.464) observations. This study uses data from multiple sources, including: UN Comtrade Database; World Bank World Development Indicators; the gravity data set available at the Centre d’Etudes Prospectives etd’Informations Internationales (CEPII) (detail see Table 3).

Table 2. List of countries used in this study.

Member countries of AHKFTA	Top 23 trade partners of Hong Kong in 2020
Brunei	USA
Cambodia	Taipei
Indonesia	India
Laos	Netherlands
Myanmar	Korean
Malaysia	UK
Philipines	UAE
Singapore	Germany
Thailand	Macao
Vietnam	Switzerland
	Mexico
Hong Kong	Russian Federation
	France
	Australia
	Italy
	Brazil
	Poland
	Belgium
	Spain
	Hungary
	Israel
	Bangladesh
	Japan

Source: Authors elaboration using UNCOMTRADE database and information from the World Trade Organization. Website: https://www.wto.org.

Table 3. Data definitions and sources.

Variables	Descriptions	Expected sign	Source
${\mathit{\text{EX}}}_{\mathit{\text{ijt}}}$	Bilateral exports from exporter $i$ to importer $j$ in period $t$		Trademap
${\mathit{\text{GDP}}}_{\mathit{\text{it}}}$; ${\mathit{\text{GDP}}}_{\mathit{\text{jt}}}$	Nominal gross domestic product in countries $i$, $\mathit{j }$	+	WDI
${\mathit{\text{POP}}}_{\mathit{\text{jt}}};{\mathit{\text{POP}}}_{\mathit{\text{jt}}}$	the populations of countries $i$ and $j$ in period $t$	+	WDI
${\mathit{\text{DIST}}}_{\mathit{\text{ij}}}$	distance between the capital cities	–	CEPII
${\mathit{\text{LANG}}}_{\mathit{\text{it}}}$	speaking the same official language	–	CEPII
$\mathit{\text{LANDLOCKij}}$	being a landlocked country	+	CEPII
${\mathit{\text{BORDER}}}_{\mathit{\text{ij}}}$	sharing common border	+	CEPII
${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$	takes a value of 1 if both countries $i$ and $j$ in year $t$ belong to the AHKFTA and zero otherwise	+	WTO
${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$	takes a value of one if exporter $i$ belongs to the AHKFTA in year $t$ and destination country $j$ does not and zero otherwise	+	WTO
${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$	takes a value of one if exporter $i$ is a non-AHKFTA member in year $t$ and destination country $j$ belongs to the AHKFTA and zero otherwise	–	WTO

Note. UN Comtrade Database; World Bank World Development Indicators (WDI); the gravity data set available at the Centre d’Etudes Prospectives etd’Informations Internationales (CEPII); Website: https://www.wto.org (WTO).

6. Main results and discussion

Table 4 summarizes the main findings. In Column (1), we use the pooled OLS technique to estimate without regard to time or country effect. Because time-invariant unobserved heterogeneity and MRTs were disregarded, the coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$, and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ are probably skewed (Yang & Martinez-Zarzoso, 2014). As a result, we estimated using fixed and random effects models with country and time effects, as shown in columns (2) and (3), respectively. Gravity variable coefficients such as GDP, population, distance, language, and landlocked are all found to be consistent with previous studies (Singh, 2021; Yang & Martinez-Zarzoso, 2014). The GDP coefficients in columns (2) and (3) for both reporter and partner countries were positive and significant. The magnitude of GDP coefficients was greater in reporter countries than in partner countries. It implies that higher levels of trade and production are associated with trading partners who are wealthy and have higher incomes (Singh, 2021). The coefficients of the population’s reporter are also found to be positive and statistically significant in columns (1) and (3) and negative and statistically significant in columns (2). In columns (1) and (3), the population coefficient for partner countries is positive and significant. According to the positive coefficient of population partner countries, a greater population in a partner country helps imported commodities to compete more effectively with domestic goods and reimburses exporters for the expense of sales operations overseas (Brada et al., 1985). In columns (1) and (3), the distance coefficient is negative and significant, indicating that transportation costs negatively affect trade flows. The coefficient of the common official language is discovered to be substantial and positive. It implies that a common language not only facilitates trade flows through translation or direct communication but it also reflects many cultural aspects (Melitz, 2008). As expected, at the 1% level, the coefficient of landlocked was positive and statistically significant in columns (1) and (3).

Table 4. Panel data gravity estimatdata gravity estimation.

VARIABLES	(1)	(2)	(3)	(4)
	OLS	FE	RE	HTM
$\mathit{\text{lnGD}}{P}_{\mathit{\text{it}}}$	1.174***	0.485***	0.804***	0.582***
	(0.0155)	(0.123)	(0.0664)	(0.0469)
$\mathit{\text{lnGD}}{P}_{\mathit{\text{jt}}}$	1.059***	0.410***	0.680***	0.440***
	(0.0152)	(0.120)	(0.110)	(0.0379)
$\mathit{\text{lnPO}}{P}_{\mathit{\text{it}}}$	0.179***	−2.648***	0.394***	−0.820***
	(0.0150)	(0.729)	(0.0737)	(0.236)
$\mathit{\text{lnPO}}{P}_j$	0.0322**	0.769	0.287***	1.504***
	(0.0148)	(0.687)	(0.0946)	(0.227)
$\mathit{\text{lnDIST}}$	−1.118***		−1.052***	−0.921***
	(0.0207)		(0.0672)	(0.177)
$\mathit{\text{LANG}}$	0.678***		0.987***	1.119**
	(0.0584)		(0.181)	(0.489)
$\mathit{\text{LANDLOCK}}$	0.190***		0.459**	0.872*
	(0.0494)		(0.194)	(0.446)
$\mathit{\text{BORDER}}$	−0.0284		−0.0541	0.262
	(0.0903)		(0.236)	(0.786)
${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$	1.097***	0.221**	0.105	0.165***
	(0.129)	(0.0870)	(0.0822)	(0.0518)
${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$	0.395***	0.228***	0.133**	0.192***
	(0.0853)	(0.0659)	(0.0620)	(0.0348)
${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$	−0.0102	−0.0755	−0.0863*	−0.0732**
	(0.0857)	(0.0528)	(0.0521)	(0.0349)
Constant	−41.51***	21.80	−30.13***	−19.16***
	(0.427)	(18.78)	−2.235	−3.686
Observations	13,464	13,464	13,464	13,464
R-squared	0.675	0.062
Number of Country		1,122	1,122	1,122
Country FE	NO	YES	YES
Year FE	NO	YES	YES

Robust standard errors in parentheses.

***p < 0.01, **p < 0.05, *p < 0.1.

In colums (1), the coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}\mathit{\text{and FT}}{A}_{2\mathit{\text{ijt}}}$ are positive and statistically significant while the coefficients of ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ is nagative and statistically significant at the 1% level. The positive coefficient of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}} \mathit{\text{showed}}$ that the AHKFTA has caused an intra-regional TC effect and increases the welfare of member countries. The positive coefficient of ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ indicated welfare gain effect for countries outside the trading bloc as well. The coefficient of ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ is found to be negative and significant at the 1% level, which is indicating that TD effect in terms of imports. This outcome is in line with the previous study (see., (Alhassan & Payaslioğlu, 2023; Yang & Martinez-Zarzoso, 2014). However, the coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$, ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ could be skewed due to neglecting time-invariant and multilateral resistance terms.

FTAs differ significantly between columns (2) and (3). In column (2), the coefficient of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$ is found to be positive and significant, indicating that the AHKFTA has resulted in a TC effect. The ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ coefficient is positive and significant in columns (2) and (3), indicating an export creation effect. The ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ coefficient is also negative and significant in column (3). It indicates a TD effect in terms of imports. Column (4) gives the results of estimating with the $\mathit{\text{Hausman}}-\mathit{\text{Taylor}}$ estimator to avoid estimate bias (Hausman & Taylor, 1981). As expected, the coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$, ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$, and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ are significant in column (4). The coefficient of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$ is positive, indicating that the AHFTA has caused a TC effect. The coefficient of ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ is positive, which suggests an export creation effect. The coefficient of ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ is also found to be negative. It indicates a TD effect in terms of imports. The differences in the coefficients of these three FTA dummies in column (4) with columns (2) and (3) show that the intensity of bilateral trade has decreased with the inclusion of the country effect and time effect. The different results obtained by these models show that the estimated effects of FTAs on trade flows are highly dependent on how scholars control for unobserved country heterogeneity, implying that estimations for unbiased results are highly reliant on correct specifications for the model.

6.1. Robustness analysis

We perform two robustness checks in this section. We use the PPML technique to control for both time-varying MRTs and to avoid the gravity equation’s endogeneity bias by introducing country-and-time effects while maintaining country-pair fixed effects. Second, we use LASSO regression to reduce the variance resulting from the standard OLS approach by introducing some bias in the coefficients and reducing the problem of overfitting and high variance during the minimization process (Maruejols et al., 2022). Nonetheless, we compare the results of data collected for the years 2010-2021 (total sample) and 2018-2021(subsample) to detect any changes in trade specialization for members in FTAs (detailed Table 5).

Table 5. Panel data gravity estimation by using PPML and LASSO technique.

VARIABLES	Full sample		Subsample
	(1)	(2)	(3)	(4)
	PPML	LASSO	PPML	LASSO
$\mathit{\text{lnGD}}{P}_{\mathit{\text{it}}}$		1.166***		1.207***
		(0.0220)		(0.0283)
$\mathit{\text{lnGD}}{P}_{\mathit{\text{jt}}}$		1.050***		1.099***
		(0.0220)		(0.0322)
$\mathit{\text{lnPO}}{P}_{\mathit{\text{it}}}$		0.199***		0.209***
		(0.0211)		(0.0315)
$\mathit{\text{lnPO}}{P}_{\mathit{\text{jt}}}$		na		na
$\mathit{\text{lnDIST}}$		−1.170***		−1.445***
		(0.281)		(0.264)
$\mathit{\text{LANG}}$		na		na
$\mathit{\text{LANDLOCK}}$		na		na
$\mathit{\text{BORDER}}$		na		na
${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$	0.00751	1.067***	0.0574	1.047***
	(0.0567)	(0.292)	(0.0684)	(0.285)
${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$	0.0269	0.356***	0.0596	0.696***
	(0.0577)	(0.137)	(0.0688)	(0.141)
${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$	−0.0202*	0.0996	−0.0167**	0.331***
	(0.0113)	(0.118)	(0.00807)	(0.124)
Observations	13428	13,464	5,580	5,610
R-squared	0.963		0.978

Standard errors in parentheses.

***p < 0.01, ** p < 0.05, * p < 0.1. na: not applicable because not selected by the LASSO technique.

Column (1) displays the results of Equation (1) estimating using the PPML model with the whole sample. These results consider zero trade, exporter-and-year and partner-and-year effects, and country-pair fixed effects, all at the same time. These findings differ from the other estimates in Table 5’s columns (1), (2), (3), and (4). Because ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ differ in three dimensions $i$, $j$, and $t$), all determinants that differ in those ijt dimensions with $\mathit{\text{it}}$ and $\mathit{\text{jt}}$ (such as GDP and population in countries i and j), as well as time-invariant effects between two countries (such as distance, common language, and border), are controlled (Yang & Martinez-Zarzoso, 2014). As a result, these results provide unbiased estimates for ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}},},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$. Nonetheless, only ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ coefficients are negative and statistically significant, while ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}} \mathit{\text{and FT}}{A}_{2\mathit{\text{ijt}}}$ coefficients are positive and insignificant. The coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt }}}$ are similar to the results in column (3). The magnitude of ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ in column (1), however, is greater than the values in columns (2) and (3). It demonstrates that the overall TD effect of the AHFTA agreement is greater than the TD effect of the AHFTA agreement when the subsample is considered.

Under the LASSO approach, the coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ in columns (2) and (4) differ significantly from the PPML estimates. Column (2) shows that the coefficients of ${\mathit{\text{FTA}}}_1$ and ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ are positive and significant at the 1% level, while the coefficient of ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ is positive but insignificant. The FTA coefficients differ significantly between columns (2), (3) and (4). At the 1% level, the coefficients of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$ and ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ become positive and significant, while the coefficient of ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ becomes positive and significant. It demonstrates that the TC and TD effect of AHFTA is more pronounced in the subsample than in the total sample. The positive coefficient of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}}$ indicates that the AHFTA has resulted in intra-regional TC and increased member countries’ welfare. The average treatment effect is $184.9\%=[\mathrm{\text{exp}}(1.047)-1]100$ greater than normal export levels. The ${\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ dummy, which represents AHKFTA member countries’ exports to non-member countries, has a significantly positive coefficient, indicating a positive export diversion effect for countries outside the trade bloc. The ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$ coefficient is also positive and significant in terms of import diversion effects. It means that imports from non-member countries have increased to IHKFTA member countries. Import creation effects are typically smaller than export creation effects. It shows a growing trend in non-member country exports to AHKFTA member countries. In our model, a pure TC effect in terms of exports and imports is recognized as coefficients of of ${\mathit{\text{FTA}}}_{1\mathit{\text{ijt}}},{\mathit{\text{FTA}}}_{2\mathit{\text{ijt}}}$ and ${\mathit{\text{FTA}}}_{3\mathit{\text{ijt}}}$.

7. Conclusion and future research

This paper has focused on the impact of FTAs between ASEAN and Hong Kong on trade flows focusing on TC and TD effects. We used a panel dataset of 34 countries including Hong Kong, ASEAN-10 countries and Hong kong’s top 23 trading partners in 2021 from 2010 to 2021. Along with applying the traditional fundamental estimation in the panel data model, this work applies the LASSO and PPML techniques to address the problems of zero trade, heteroskedasticity in the error term, overfitting, and high variance during the minimization process. This study demonstrates that tariff reduction and elimination under AHKFTA boosts overall trade volume among intra-bloc member countries as well as intra-bloc and extra-bloc countries.

Thus, this paper makes important contributions to the literature in terms of assessing TC and TD effects of FTAs. Firstly, our study stands out from the existing literature due to its unique focus on the ASEAN-Hong Kong Free Trade Agreement, which has had no attention in previous studies. Secondly, unlike prior research that assessed the impact of FTAs on trade flows without subgroup analysis to identify changes in trade specialization among FTA members, our study specifically delves into an evaluation of the TC and TD effects of AHKFTA. Thirdly, our study takes a methodological stride by simultaneously applying PPML and LASSO regression approaches to investigate the evaluating TC and TD effects.

Our findings propose that the current trade policy between Hong Kong and ASEAN should be maintained, as it promotes both AHKFTA’s intra-regional trade growth and extra-bloc countries. To attain a more profound economic integration within the region, the AHKFTA should broaden its focus beyond tariff barriers. Efforts should also be directed towards enhancing production efficiency, bolstering product competitiveness, and optimizing trade complementarities (Akram, 2020a; Akram, 2020b; Akram, 2022). In future research, it is necessary to take into consideration disaggregated data for specific commodities to evaluate TC and TD effects of AHKFTA in different commodities goods.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research is funded by Vietnam National University HoChiMinh City (VNU-HCM) under grant number [B2021-34-03].

Notes on contributors

Uyen Pham

Uyen Pham has good experience in teaching graduate students, conducting quantitative research at University of Economics and Law. The research areas of the author include international trade, international tourism. She published different articles in reputable international journals.

Uyen Vo

Uyen Vo has good experience in teaching graduate students, conducting quantitative research at University of Economics and Law. The research areas of the author include international trade, tourism. She published different articles in reputable international journals.

Quy Trinh

Quy Trinh has good experience in conducting quantitative research at University of economics and law. The research areas of the author include international trade, tourism. He published different articles in reputable international journals.

Hoa Le

Hoa Le has good experience in teaching graduate students, conducting quantitative research at University of Economics and Law. The research areas of the author include international trade, tourism. She published different articles in reputable international journals.