Growing Olive Oil Export and Intra-Industry Trade in Mediterranean Countries: Application of Gravity Model

Abstract

While olive oil production is spreading to the non-traditional producer countries, including the US, Australia, and New Zealand, Mediterranean countries are still major producers and exporters. However, little is known about their olive oil exports simultaneously growing in tandem with their large volume of imports. This paper examines the factors that affect olive oil exports and imports in Mediterranean countries. Using balanced panel data of olive oil trade in Mediterranean countries from 1998 to 2016, we estimated the commodity-specific gravity model. Results suggest that an increase in the overall bilateral size of trading partners positively affects the flow of olive oil trade. The difference in factor endowments has a negative impact on exports, whereas its effect is positive on their imports. The members of the European Union (EU) are competitive in olive oil export, and the volume of its import is large among the EU countries whose per capita income and demand properties are similar. These results support Linder’s hypothesis rather than the predictions from the traditional Heckscher–Ohlin trade theory. The simultaneous export and import of olive oil in Mediterranean countries implies the relevance of a growing intra-industry trade rather than a country’s specialization following its comparative advantage.

1. Introduction

Over the last quarter-century, the size of the global olive oil market has been growing. According to statistics from the International Olive Council, the annual global export of olive oil increased from 337 to 778 thousand tones between 1990/91 and 2016/17. In the last five years, the European Union (EU) has not only dominated the export market, shipping out 66.4% of all olive oil globally, but has also had 15% of the world import share. While the EU’s export of olive oil continues to grow, a large import volume has also been observed. Traditionally, olive oil is produced in the Mediterranean basin and traded among Mediterranean countries, which have a comparative advantage in the production and export of olive oil [1]. However, little is known about the trend reflecting sizable and simultaneous growth in the import of olive oil. Figure 1 presents the increase in export volume in the Mediterranean countries, which also coincides with substantial import volume. As shown in Figure 2, these Mediterranean countries produce a surplus of olive oil but face declining consumption. Despite this excessive supply of olive oil, these countries continue to import a large volume of olive oil from other countries. This phenomenon is typically observed in traditional olive oil producers such as Italy, Spain, and France.

The questions are “What explains the simultaneous growth in sizable olive oil export and import?” and “What factors affect the export and import of olive oil?” Although there are Mediterranean countries that specialize in the production and export of olive oil, some countries import olive oil despite having a surplus from over-production. Meanwhile, others experience excess demand for olive oil but continue to export. The orthodox theories of international trade, including the Heckscher–Ohlin (H–O) theory, assume that countries tend to specialize in producing and exporting goods for which they have a comparative advantage. The difference in factor price caused by the difference in relative factor endowments is one of the major triggers of international trade. Following this argument, the rapid increase in the export of olive oil from Mediterranean countries seems to be consistent with the idea of specialization based on comparative advantage. However, olive oil export by Mediterranean countries is growing coincidentally with the sizable import. Particularly, imports of EU countries from southern Mediterranean countries and the subsequent re-export to the world are increasing rapidly. This simultaneous export and import of goods by a country is inconsistent with the trade patterns based on the comparative advantage. Currently, Krugman [2], and Helpman and Krugman [3] developed a new trade theory that emphasizes the significance of economies of scale combined with product differentiation and transportation costs. According to Helpman [4], large volumes of trade flow are typically observed between countries with similar factor proportions, where factor endowment differences do not drive a considerable proportion of trade.

Linder [5] suggests that the greater the similarity in factor endowments between two countries, the larger the expected trade flows between the two. The trade flow of olive oil observed in the Mediterranean region may have more relevance in the context of the new trade theory than the traditional discussions. Such trade flow of olive oil can be considered a typical pattern of intra-industry trade. Along these lines, the objective of this paper is to examine the factors that affect the export and import of olive oil in Mediterranean countries. To achieve this, the commodity-specific gravity model is estimated in order to investigate the determinant factors that explain the growth in exports and imports [6,7]. The data used in this study are the balanced panel data of export and import flows from 1998 to 2016. Following the empirical studies of the gravity model by Egger [8], Baltagi et al. [9], and Paas et al. [10], we examine the variables that assess the validity of the new trade theory, assuming that bilateral trade is a function of the sum of two countries’ real gross domestic product (GDP); the similarity of two trading countries; and the difference in relative factor endowments between the two. Further, we assume a positive effect of dummy variables including common border, common language, former colonial relationships, and EU membership as components for analysis.

This paper provides empirical evidence on olive oil trade that support Linder’s hypothesis rather than the traditional H–O trade theory. It suggests that the differences in relative factor endowments cannot adequately explain the trend of olive oil export observed among Mediterranean countries. This is because the trade pattern of olive oil including specialization of production and export does not follow the law of comparative advantage in a straightforward way. Trade volume is becoming larger among EU countries whose per capita income and demand properties are more similar. The simultaneous export and import of olive oil in Mediterranean countries implies the relevance of a growing phenomenon of intra-industry trade. The Grubel–Lloyd index applied to measure the degree of intra-industry trade of olive oil suggests that intra-industry trade constitutes more than half of trade flow. As policy implications, this paper emphasizes the significance of product differentiation among olive oils in response to the growing phenomenon of intra-industry trade. As it relates to the promotion of exports, we note that product differentiation is becoming more crucial for the quality improvement that results in competitive advantage.

2. Literature Review

The gravity model is one of the most frequently used models for empirical studies on international trade. Most traditional international trade theories do not consider the distance effect as a resistant factor. However, Isard [11] introduced a physics-based spatial interaction theory into international trade. According to Isard [11], Tinbergen [12], and Pöyhönen [13], the gravity force in the international economics, i.e., trade flows between two countries, increases with country size and decreases with trade costs. As the distance between countries that are trade partners increases, trade costs are expected to rise proportionately.

Formal theoretical foundations of the gravity model were provided by Anderson [14] and Bergstrand [15,16], who argue that international trade flows between two countries are proportional to the scale of their economies and are inversely affected by their distance as a resistance factor. Apart from the effect of the sizes of two countries, which can be represented by the GDP or GDP per capita [17,18], the difference in their GDP per capita is considered a proxy variable for the difference in relative factor endowments. Following the neoclassical trade theory of the H–O model, the difference in factor price caused by the difference in factor endowments determines the pattern of trade between countries. Under the assumption of the H–O model, larger volumes of trade are expected as the difference in factor endowments expands.

Regarding the effect of the difference in factor endowments, Linder [5] developed an alternative trade model in the early 1960s. Linder assumed that the trade volume between trading countries increases if per capita income levels between the two are similar. The income similarity model, as the hypothesis is known, states that the demand propensity becomes similar at the same time as the resemblance of per capita income increases [19]. The hypothesis also suggests that the trade volume between two countries increases as the relative factor endowments between them becomes more similar [9]. Helpman [4] argues that the quantity of trade between two countries is expected to be larger when they have similar factor proportions and the difference in factor endowments does not drive a substantial portion of the business. Unlike the classical H–O theory, Linder’s hypothesis assumes that bilateral trade is positively related to the similarity in relative country size and negatively related to increases in the difference in relative factor endowments. Since the mid-1990s, the specification of the gravity model derivations based on the new trade theory has become increasingly popular [3,20,21].

Based on this argument, we employ the commodity-specific gravity model to a single agricultural commodity. While many estimations of that model use aggregate data of trade flows, several studies applied it to a specific sector or a particular commodity. On the sectoral level, Martínez-Zarzoso and Nowak-Lehmann [22] estimated the modified gravity model by using disaggregated trade data to explain Mercosur’s export to the EU. For the agricultural sector, Koo et al. [6] estimated the gravity model to explain the trade flows of agricultural commodities. Crescimanno et al. [23] applied the gravity model to Italy’s agro-food exports to non-EU Mediterranean partner countries. This study found that the income of trading partners, colonial and historical relationships, and distance between trading partners affect trade flow. The gravity model has been applied to specific agro-food products as well to find factors that explain the trade flow of agricultural products, such as wine, beef, fruits, and vegetables [7,24,25]. Sheldon et al. [26] examined the effect of exchange rate uncertainty on the US bilateral trade of fresh fruit and fresh vegetables. The results showed that it has negatively affected fresh fruit trade and fresh vegetable export. For citrus exports of Turkey, Özer and Koksal [27] suggested that trade agreements and increases in real exchange rates have a positive effect on export volume, while transportation costs represented by the distance between Turkey and its trade partners affect exports negatively. The gravity model estimation by Márquez-Ramos and Martinez-Gomez [28] found that trade preferences positively affect exports of fruit and vegetables from Morocco to EU countries.

For the trade of olive oil, Vlontzos and Duquenne [29] estimated the commodity-specific gravity model for export in Greece using data from 1991 to 2005. This study found that the country’s virgin olive oil exports were positively impacted by the change in the GDP per capita of trade partners, the existence of a Greek community in partner countries, and tourist inflows to Greece. Regarding imports of olive oil, Kavallari et al. [30] estimated the gravity equation in Germany using the random effects model using panel data from 1995 to 2006 with 14 exporting partners. This study discovered that tourism, being a Mediterranean partner country of the EU, and the presence of direct marketing channels have a positive impact on Germany’s olive oil import. In southern Mediterranean countries, Angulo et al. [31] estimated the gravity model as applied to olive oil exports from Tunisia. Using panel data from 2001 to 2009, they found that the income level, human development index, and existence of a common language between Tunisia and its trading partners had a positive impact on Tunisia’s olive oil exports.

Most studies applied the gravity model on olive oil trade in line with the orthodox H–O discussion. However, studies examining Linder’s hypothesis are missing. The effect of variables related to the new trade theory (overall country size, the similarity index, and the difference in relative factor endowments) on the olive oil trade has not been explicitly examined. Most studies applied to olive oil trade have focused on exports and a single country’s case. However, our study uses panel data of multiple countries and examines determinant factors affecting exports and imports. Rallatou and Tzouvelekas [32] estimated the gravity equation using panel data of olive oil trade within 28 member states of the EU over the period from 2000 to 2015. They confirmed the positive effect of per capita GDP by eliminating the country-specific unobserved effects; however, the potential endogeneity problem of the explanatory variable was not well-controlled. This study attempts to fill these gaps. What is more, we employed Arellano–Bond dynamic panel-data estimation with the system Generalized Method of Moment (GMM) to solve the problems of serial correlation, heteroscedasticity, and endogeneity of some of the regressors.

3. Materials and Methods

3.1. Model Specification

Following Egger [8], Baltagi et al. [9], and Paas et al. [10], we employ the gravity model derived from the new trade theory. Bilateral trade is a function of the overall country size, relative country size, and the difference in relative factor endowments:

$$X_{ijt}= {\beta }_0+ {\beta }_{1}DI{S}_{ij}+ {\beta }_{2}GD{T}_{ijt}+ {\beta }_{3}SI{M}_{ijt}+ {\beta }_{4}RF{E}_{ijt}+ Z_{ijt}^{\prime }\delta + u_{ijt},$$

(1)

where X_ijt denotes the log of the volume of bilateral trade between country i and country j at year t; β_k (k = 0, 1, 2, 3, 4) denotes unknown parameters to be estimated; DIS_ij is the log of geographical distance between country i and j; δ is a vector of unknown parameters to be estimated; Z_ijt is the h × 1 vector of other explanatory variables. The error term, u_ijt, comprises the fixed or random unobserved effect that controls an individual country’s characteristics, time-specific fixed or random effects, and the remaining error term. DIS_ij is a proxy for trade costs and its coefficient should be negative because it is a resistance factor on trade. GDT_ijt is the log of the sum of the real GDPs of the trading partners i and j at year t, measured by the real value of US dollars. It is a measure of bilateral overall country size and is expected to positively impact trade flow:

$$GD{T}_{ijt}=\log \left(GD{P}_{it}+ GD{P}_{jt}\right).$$

(2)

SIM_ijt represents the similarity index calculated by two trading countries’ GDPs to measure the relative country size:

$$SI{M}_{ijt}=\log \left|1-{\left(\frac{GD{P}_{it}}{GD{P}_{it}+GD{P}_{jt}}\right)}^2-{\left(\frac{GD{P}_{jt}}{GD{P}_{it}+GD{P}_{jt}}\right)}^2\right|.$$

(3)

This similarity index is expected to have a positive effect on trade flows between country i and j because more similar countries prefer to engage in larger trade flow.

The absolute difference in relative factor endowments, RFE_ijt, is given as follows:

$$RF{E}_{ijt}=\left|\log \left(\frac{GD{P}_{it}}{PO{P}_{it}}\right)-\log \left(\frac{GD{P}_{jt}}{PO{P}_{jt}}\right)\right|,$$

(4)

where POP denotes a country’s population. According to Liner [5], the greater the difference in factor endowments between two countries, the smaller their expected trade flows. Linder’s hypothesis suggests that RFE_ijt is expected to have a negative effect on trade flow. Regarding the explanatory variables of Z_ijt, we include dummy variables of EU membership, contiguity, common language, and former colonial relationships. Many studies confirm a positive effect of free-trade agreements on trade flow [12,33,34,35,36,37,38]. For wine trade, Dsacal et al. [7] found a positive impact of having EU-membership on exports and imports. Kavallari et al. [28] confirmed with the olive oil import of Germany that being an EU member had a positive impact. Following these findings, we assume being an EU member has a positive effect on trade flow. Several studies confirmed the positive impact of contiguity, sharing a common official language, and former colonial relationships with a trade partner, on trade flow [39,40]. Angulo et al. [31] found with the olive oil export in Tunisia that having a common language has a positive impact. We follow the same assumption as these studies for the sign of coefficients of contiguity, common language, and former colonial relationships.

In this study, Equation (1) is applied to the export and import of olive oil in Mediterranean countries. The gravity equations of olive oil export and import are expressed as follows:

$$\begin{gathered} E{X}_{ijt}= {\gamma }_0+ {\gamma }_{1}DI{S}_{ij}+ {\gamma }_{2}GD{T}_{ijt}+ {\gamma }_{3}SI{M}_{ijt}+ {\gamma }_{4}RF{E}_{ijt}+ {\gamma }_{5}DE{I}_{it} \\ + {\gamma }_{6}DE{J}_{jt}+ {\gamma }_{7}DE{B}_{ijt}+ {\gamma }_{8}DC{B}_{ij}+ {\gamma }_{9}DC{L}_{ij}+ {\gamma }_{10}DC{R}_{ij}+ {\mu }_{ijt}, \end{gathered}$$

(5)

$$\begin{gathered} I{M}_{ijt}= {\pi }_0+ {\pi }_{1}DI{S}_{ij}+ {\pi }_{2}GD{T}_{ijt}+ {\pi }_{3}SI{M}_{ijt}+ {\pi }_{4}RF{E}_{ijt}+ {\pi }_{5}DE{I}_{it} \\ + {\pi }_{6}DE{J}_{jt}+ {\pi }_{7}DE{B}_{ijt}+ {\pi }_{8}DC{B}_{ij}+ {\pi }_{9}DC{L}_{ij}+ {\pi }_{10}DC{R}_{ij}+ v_{ijt}, \end{gathered}$$

(6)

where EX_ijt denotes the log of the volume of export, i.e., export of country i to country j at year t; IM_ijt is the log of the volume of import, i.e., import of country i from country j at year t, measured by the real value of US dollars; γ_l, π_m (l, m = 0, 1, 2, 3, …, 10) denote unknown parameters to be estimated; DIS_ij is the log of the geographical distance between capital cities of country i and j, measured by kilometers; DEI_it is a binary variable that takes the value of 1 if country i becomes an EU member, and 0 otherwise; DEJ_jt is a binary variable that takes the value of 1 if country j becomes an EU member, and 0 otherwise; DEB_ijt is a binary variable that takes the value of 1 if both i and j are EU members, and 0 otherwise; DCB_ij is a dummy variable that is equal to 1 if country i and j share a common border, and 0 otherwise; DCL_ij is a dummy variable that is equal to 1 if country i and j have a common official language, and 0 otherwise; DCR_ij is a dummy variable that is equal to 1 if country i and j had colonial relationships, and 0 otherwise. The dummy variables of EU membership are expected to positively impact trade flow. The variables of sharing a common border, having a common language, and having former colonial relationships are assumed to have a positive influence on trade flows. The error terms, μ_ijt, v_ijt, are decomposed into the fixed or random unobserved individual country’s characteristics, time-specific fixed or random effects, and remaining error terms.

3.2. Linear Dynamic Panel Gravity Model

We estimate Equations (5) and (6) by applying static panel data methods, including the random-effects and fixed-effects model. The random-effects estimator regards intercept as random variables, assuming no correlation between explanatory variables and unobservable country-specific characteristics. On the other hand, the fixed-effects model assumes that the country-specific effects appeared in intercepts. It provides a consistent estimator even if the country-specific effects correlate with explanatory variables. Yet, the time-invariant variables appearing in the equations, including DIS_ij, DCB_ij, DCL_ij, and DCR_ij, cannot be estimated in the fixed-effects model. What is more, due to the large panel data set, we have to consider the problem of heteroscedasticity and autocorrelation of error terms that lead to an unbiased estimator. Moreover, the bias caused by missing unobservable country-specific effects may be mitigated by the fixed-effects estimator, yet the endogeneity bias still remains, for instance, the relation between GDT_ijt and bilateral trade flows. Similar to GDT_ijt, SIM_ijt and RFE_ijt would not be strictly exogenous.

Considering these limitations, we extend our model from static to a dynamic panel data model, i.e., the system GMM technique augmented by Arellano and Bover [41] and developed by Blundell and Bond [42]. In GMM, the dynamic relationships are characterized by the introduction of the lagged dependent variable of the regressors. Based on Equation (1), a dynamic panel model can be rewritten as follows:

$$X_{ijt}= {\beta }_0+ \rho X_{ij,t-1}+ {\beta }_{1}DI{S}_{ij}+ {\beta }_{2}GD{T}_{ijt}+ {\beta }_{3}SI{M}_{ijt}+ {\beta }_{4}RF{E}_{ijt}+ Z_{ijt}^{\prime }\delta + {\phi }_{ij} + e_{ijt},$$

(7)

where X_ij,t−1 denotes the lagged dependent variable; ρ is a parameter to be estimated; φ_ij denotes country-pair specific effects; e_ijt is the error term. Following Arellano and Bond [43], the first difference transformation of Equation (7) removes both the constant terms and country-pair specific effects:

$$\Delta X_{ijt}= \rho \Delta X_{ij,t-1}+ {\beta }_2\Delta GD{T}_{ijt}+ {\beta }_3\Delta SI{M}_{ijt}+ {\beta }_4\Delta RF{E}_{ijt}+ \Delta Z_{ijt}^{\prime }\delta + \Delta e_{ijt}.$$

(8)

By taking the first difference, the potential bias due to the omitted variables arising from unobserved country-pair-specific effects is removed, but there is still a correlation between the differenced lagged dependent variables (∆X_ij,t−1) and the disturbance process (∆e_ijt). To deal with bias caused by the endogeneity problem, the first difference of this explanatory variable is instrumented by their lagged values (in levels). Under the assumption of no-autocorrelation in the error terms, the explanatory variable may strongly correlate with its lagged variables, but not with future values of the error term. Therefore, instrumental variable estimators can deal with the endogeneity of the lagged dependent variables.

However, the first difference GMM estimator is likely to suffer from poor performance and weakness instruments, as the lagged levels of explanatory variables are weak instruments for the first differences [41]. They perform poorly as instruments if the dependent variable is close to follow a random work [42]. Alternatively, Arellano and Bover [41] and Blundell and Bond [42] introduced a system GMM estimator to deal with the shortcomings of the first difference approach. The GMM system estimator combines the equation in difference with an equation in levels. Namely, it simultaneously estimates Equation (8) with Equation (7). In Equation (7), the explanatory variables of the equation in levels are instrumented by their most recent lags in the first differences. To examine the validity of using lagged variables as instruments, we test the Sargan/Hansen test of overidentification restrictions [41,43]. In addition, the GMM estimator is consistent if no second-order serial correlations exist in the residuals of the equation in differences [43]. We expect to reject the null hypothesis of first-order autocorrelation in Equation (8), but not the null of the second-order autocorrelation.

Given the above conditions, the following linear dynamic panel equation (in levels) is employed to empirically estimate the gravity model:

$$\begin{array}{l}E{X}_{ijt}= {\theta }_0+ {\theta }_{1}E{X}_{ij, t-1}+ {\theta }_{2}DI{S}_{ij,}+ {\theta }_{3}GD{T}_{ijt}+ {\theta }_{4}SI{M}_{ijt}+ {\theta }_{5}RF{E}_{ijt}+ {\theta }_{6}DE{I}_{it} + {\theta }_{7}DE{J}_{jt}+\\ {\theta }_{8}DE{B}_{ijt}+ {\theta }_{9}DC{B}_{ij} + {\theta }_{10}DC{L}_{ij}+ {\theta }_{11}DC{R}_{ij}+{ }_{ij}+ {\tau }_t+ {\epsilon }_{ijt},\end{array}$$

(9)

$$\begin{array}{l}I{M}_{ijt}= {\omega }_0+ {\omega }_{1}I{M}_{ij, t-1}+{\omega }_{2}DI{S}_{ij}+ {\omega }_{3}GD{T}_{ijt}+ {\omega }_{4}SI{M}_{ijt}+ {\omega }_{5}RF{E}_{ijt}+ {\omega }_{6}DE{I}_{it} + {\omega }_{7}DE{J}_{jt}+\\ {\omega }_{8}DE{B}_{ijt}+ {\omega }_{9}DC{B}_{ij}+ {\omega }_{10}DC{L}_{ij} + {\omega }_{11}DC{R}_{ij}+{ }_{ij}+ {ϵ}_t+{ }_{ijt},\end{array}$$

(10)

where EX_ij,t−1 denotes the lagged dependent variable of the export equation; IM_ij,t−1 is the lagged dependent variable of the import equation; θ_k, ω_n (k, n = 0, 1, 2, 3, …, 11) denote unknown parameters to be estimated. The terms λ_ij, k_ij are unobserved country-pair specific effects, which are independently and identically distributed, i.e., λ_ij ~ iid(0, σ_λ²), k_ij ~ iid(0, σ_λ²). The terms τ_t, ϵ_t are the time-fixed effects. The remaining error terms, ε_ijt, η_ijt, are also distributed independently and identically, and serially uncorrelated, i.e., ε_ijt ~ iid(0, σ_ε²), η_ijt ~ iid(0, σ_η²). In the estimation procedure, we regard GDT_ijt, SIM_ijt, and RFE_ijt as endogenous variables, which may correlate with the error term. Following Roodman [44], we use STATA’s estimator “xtabond2” to derive the coefficients using the system GMM estimator controlling potential bias due to the endogeneity of some of the regressors.

3.3. Data

The major exporters and importers of olive oil in the Mediterranean countries included in the analysis are reported in

Table 1. Major exporters and importers of olive oil in the Mediterranean countries.

Country Group	Countries
Exporters	Croatia, Cyprus, Egypt, France, Greece, Israel, Italy, Lebanon, Morocco, Spain, Tunisia, Turkey
Importers	Egypt, France, Greece, Italy, Morocco, Slovenia, Spain

. The estimation of the gravity model is implemented for the export of olive oil from 12 Mediterranean countries to their 125 major trading partners from 1998 to 2016 (Appendix A

Table A1. The trade partners of olive oil exports for the Mediterranean countries.

Albania	Cameroon	Gambia	Korea	Slovenia	Zimbabwe
Algeria	Canada	Germany	Kuwait	South Africa	Moldova
Andorra	Cambodia	Georgia	Kyrgyzstan	Spain	Myanmar
Angola	Cabo Verde	Ghana	Latvia Lebanon	Sri Lanka	Netherlands
Argentina	Chile	Greece	Liberia	Sweden	Nepal
Armenia	China	Guatemala	Lithuania	Switzerland	New Zealand
Australia	Columbia	Hong Kong	Luxemburg	Tanzania	Nicaragua
Austria	Cote d’Ivoire	Honduras	Malaysia	Thailand	Nigeria
Azerbaijan	Costa-Rica	Hungary	Malta	Tunisia	Norway
Bahrain	Croatia	India	Maldives	Turkey	Oman
Bangladesh	Cyprus	Indonesia	Mauritania	Turkmenistan	Panama
Belarus	Denmark	Iran	Mauritius	Trinidad-Tobago	Pakistan
Belgium	Dominica	Iraq	Mexico	United Arab	Paraguay
Belize	Ecuador	Ireland	Mongolia	Emirates	Peru
Benin	Egypt	Israel	Morocco	United Kingdom	Portugal
Bolivia	El-Salvador	Italy	Saudi Arabia	United States	Philippines
Bosnia-Herzegovina	Estonia	Jamaica	Senegal	Ukraine	Poland
Brazil	Ethiopia	Japan	Seychelles	Uruguay	Portugal
Brunei	Finland	Jordan	Singapore	Uzbekistan	Romania
Bulgaria	France	Kazakhstan	Slovakia	Vietnam	Russia
Burkina Faso	Gabon	Kenya		Yemen
				Zambia

). The dataset for the estimation contains balanced panel data of olive oil export with the number of observations set at 11,476. During the same period, we use panel data of the import from 7 Mediterranean countries to their 31 major trading partners (Appendix A

Table A2. The trade partners of olive oil imports for the Mediterranean countries.

Algeria	Croatia	Israel	Portugal	Turkey
Andorra	Cyprus	Japan	Saudi Arabia	United Kingdom
Argentina	Egypt	Jordan	Slovenia	United States
Austria	France	Lebanon	South Africa
Australia	Germany	Luxemburg	Spain
Belgium	Greece	Morocco	Switzerland
Bulgaria	Italy	Netherlands	Tunisia

). In total, we use balanced panel data of olive oil import with 1843 observations. Data regarding the export and import of olive oil are used for estimation and are extracted from the UN Comtrade Database, which is available from: https://comtrade.un.org/ (Accessed: 30 July 2018). We used data of olive oil trade by HS commodity code 1509. According to this database, olive oil is defined as “olive oil and its fractions, whether or not refined, but not chemically modified.” We measure the dependent variable as log(1 + X_ijt), where X_ijt indicates either the export value or the import value. The reason behind this transformation is that we cannot take the log of X_ijt in some cases, because the trade volume is reported as zero. Regarding independent variables, trade costs are proxied by the distance in kilometers between two countries’ capitals. The distance is measured by the straight line between capital cities, using the Distance Calculator. It is available from: https://www.distancecalculator.net/ (Accessed: 11 August 2018).

The real GDP and the real GDP per capita were used to calculate GDT_ijt, SIM_ijt, and RFE_ijt. The data source is the World Development Indicators provided by the World Bank.

Table 2. Summary statistics of the variables for the gravity model of olive oil export.

Variables	Number of Observations	Mean	Standard Deviation	Minimum	Maximum
EXijt	11,476	10.467	5.136	0	21.083
DISij	11,476	8.152	0.916	4.443	9.899
GDTijt	11,476	52.69	2.362	43.979	59.123
SIMijt	11,476	−2.245	1.428	−7.922	−0.693
RFEijt	11,476	1.508	1.101	0	5.244
DEIit	11,476	0.516	0.5	0	1
DEJjt	11,476	0.236	0.425	0	1
DEBijt	11,476	0.109	0.312	0	1
DCBij	11,476	0.05	0.217	0	1
DCLij	11,476	0.103	0.304	0	1
DCRij	11,476	0.056	0.23	0	1

and

Table 3. Summary statistics of the variables for the gravity model of olive oil import.

Variables	Number of Observations	Mean	Standard Deviation	Minimum	Maximum
IMijt	1843	10.457	5.565	0	21.119
DISij	1843	7.332	0.952	4.762	9.666
GDTijt	1843	54.128	2.175	48.23	59.123
SIMijt	1843	−1.717	0.954	−5.407	−0.693
RFEijt	1843	0.932	0.877	0.001	2.997
DEIit	1843	0.802	0.398	0	1
DEJjt	1843	0.569	0.495	0	1
DEBijt	1843	0.451	0.498	0	1
DCBij	1843	0.196	0.397	0	1
DCLij	1843	0.072	0.259	0	1
DCRij	1843	0.062	0.241	0	1

present the summary statistics of the variables for the gravity model for export and import, respectively.

4. Empirical Results

Equations (5) and (6) were applied to the static panel data regression approach, including four types of models: A pooled ordinary least squares model (POL), a fixed-effect model (FEM), a random-effects model (REM), and a Tobit model.

Table 4. Results of panel-data estimation of export model.

Variable	Pooled OLS		Fixed-Effects Model		Random-Effects Model		Tobit Model
Constant	−25.531	***	−58.132	***	−30.341	***	−30.075	***
	[3.945]		[21.987]		[4.456]		[4.493]
DISij	−0.120		-		−0.331		−0.138
	[0.186]				[0.201]		[0.212]
GDTijt	0.656	***	1.312	***	0.797	***	0.727	***
	[0.073]		[0.413]		[0.084]		[0.082]
SIMijt	0.051		−0.498		−0.123		0.029
	[0.124]		[0.511]		[0.140]		[0.139]
RFEijt	−0.465	***	−1.973	***	−0.862	***	−0.506	***
	[0.129]		[0.479]		[0.164]		[0.146]
DEIit	2.014	***	0.120		1.204	***	2.291	***
	[0.381]		[0.887]		[0.400]		[0.425]
DEJjt	0.302		−2.057	***	−1.146	*	0.384
	[0.488]		[0.927]		[0.670]		[0.554]
DEBijt	0.340		2.093	**	1.210	*	0.227
	[0.599]		[0.968]		[0.711]		[0.656]
DCBij	1.286	***	-		0.942		1.376	*
	[0.642]				[0.662]		[0.701]
DCLij	0.735		-		0.625		0.726
	[0.614]				[0.646]		[0.719]
DCRij	0.615		-		1.177		0.768
	[0.775]				[0.830]		[0.877]
Time dummies	Yes		Yes		Yes		Yes
R2	0.261
Within R2			0.155		0.151
Between R2			0.258		0.329
Overall R2			0.197		0.249
Pseudo-R2							0.048
Log pseudo-likelihood							−31,892.5
Hausman test			78.3	***
			(0.000)
Breush–Pagan test					21,692.2	***
					(0.000)
Number of groups			604		604
Number of observations	11,476		11,476		11,476		11,476

Notes: Figures in square brackets are standard errors clustered by country pair. Figures in parentheses are p-values. *, **, *** indicate significant at the 10% level, 5% level, and 1% level, respectively.

presents the estimation results of the export of olive oil by using the four models. The estimation results of the import of olive oil are presented in

Table 5. Results of panel-data estimation of import model.

Variable	Pooled OLS		Fixed-Effects Model		Random-Effects Model		Tobit Model
Constant	−14.583		−158.955	*	−30.139	*	−19.870
	[13.108]		[92.643]		[16.606]		[15.346]
DISij	−0.872		-		−0.959		−0.984
	[0.607]				[0.619]		[0.678]
GDTijt	0.543	**	3.237	*	0.905	***	0.639	**
	[0.251]		[1.688]		[0.311]		[0.289]
SIMijt	0.346		4.877		0.304		0.359
	[0.476]		[3.550]		[0.644]		[0.555]
RFEijt	0.532		6.812	*	0.996	*	0.509
	[0.718]		[3.142]		[0.735]		[0.841]
DEIit	−0.038		−6.343	***	−4.736	***	−0.088
	[1.786]		[1.101]		[1.161]		[2.116]
DEJjt	−0.710		−4.876	**	−4.132	***	−0.753
	[1.909]		[1.903]		[1.407]		[2.310]
DEBijt	2.715		8.130	***	6.692	***	3.055
	[2.292]		[1.253]		[1.338]		[2.734]
DCBij	−0.257		-		0.196		−0.268
	[1.315]				[1.223]		[1.441]
DCLij	−0.901		-		−2.613		−1.278
	[1.718]				[2.404]		[2.039]
DCRij	1.364		-		1.804		1.851
	[1.727]				[1.921]		[1.946]
Time dummies	Yes		Yes		Yes		Yes
R2	0.154
Within R2			0.133		0.120
Between R2			0.044		0.143
Overall R2			0.036		0.129
Pseudo-R2							0.028
Log pseudo-likelihood							−5309.8
Hausman test			35.4	*
			(0.062)
Breush–Pagan test					4152.48	***
					(0.000)
Number of groups			97		97
Number of observations	1843		1843		1843		1843

Notes: Figures in square brackets are standard errors clustered by country pair. Figures in parentheses are p-values. *, **, *** indicate significant at the 10% level, 5% level, and 1% level, respectively.

. As expected, the POL and Tobit estimator gave similar results with respect to signs and the significance of estimated coefficients. In the FEM, the time-invariant variables including DIS_ij, DCB_ij, DCL_ij, and DCR_ij were automatically dropped. The Hausman test is applied to test the null hypothesis of the REM against the FEM. In both estimations of export and import, the null hypothesis of the REM over the FEM is rejected. The Breuch–Pagan test rejects the null hypothesis that POL is more appropriate than the REM. The F-test strongly rejects the null hypothesis of coefficients of country-dummies being equal to zero. Thus, we confirm the proper estimators of the FEM over those of the POL. These results suggest that the estimation of FEM is appropriate for the specification in both export and import models.

Table 6. Results of dynamic panel-data estimation of export model.

	Dynamic Panel-Data Estimation
Variable	OLS		Fixed-Effects Model		Two-Step System GMM
Constant	−6.237	***	−7.467		−4.811	***
	[1.037]		[14.858]		[1.241]
EXij,t−1	0.496	***	0.327	***	0.661	***
	[0.014]		[0.018]		[0.037]
EXij,t−2	0.265	***	0.11	***	0.169	***
	[0.014]		[0.015]		[0.035]
DISij	−0.018		-		−0.017
	[0.046]				[0.038]
GDTijt	0.162	***	0.282		0.122	***
	[0.018]		[0.278]		[0.024]
SIMijt	0.001		0.145		−0.087	*
	[0.029]		[0.334]		[0.045]
RFEijt	−0.098	***	−1.061	***	−0.141	***
	[0.032]		[0.343]		[0.051]
DEIit	0.504	***	−0.449		0.314	***
	[0.099]		[0.495]		[0.105]
DEJjt	0.036		−0.853		0.006
	[0.126]		[0.553]		[0.104]
DEBijt	0.081		0.913		0.018
	[0.156]		[0.575]		[0.133]
DCBij	0.255		-		0.161
	[0.164]				[0.148]
DCLij	0.182		-		0.193
	[0.152]				[0.121]
DCRij	−0.048		-		−0.106
	[0.191]				[0.159]
Time dummies	Yes		Yes		Yes
R2	0.636
Within R2			0.253
Between R2			0.691
Overall R2			0.505
Hansen J statistics test					596.19
					−0.217
Arellano–Bond Test for AR(1)					−9.01	***
					(0.000)
Arellano–Bond Test for AR(2)					0.66
					(0.506)
Number of instruments					599
Number of groups			604		604
Number of observations	10,268		10,268		10,268

Notes: Figures in square brackets in OLS and fixed-effects model are robust standard errors clustered by country pair. Figures in square brackets in two-step system GMM are Windmeijer’s corrected standard errors. Figures in parentheses are p-values. *, *** indicate significant at the 10% level, and 1% level, respectively.

and

Table 7. Results of dynamic panel-data estimation of import model.

	Dynamic Panel-Data Estimation
Variable	OLS		Fixed-Effects Model		Two-Step System GMM
Constant	−5.703		−86.727		−83.070
	[4.953]		[76.501]		[55.571]
IMij,t−1	0.681	***	0.298	***	0.358	***
	[0.030]		[0.041]		[0.053]
DISij	−0.200		-		−1.169
	[0.181]				[0.767]
GDTijt	0.187	**	1.813		1.734	*
	[0.092]		[1.391]		[1.039]
SIMijt	0.164		3.659		−0.894
	[0.160]		[2.356]		[1.761]
RFEijt	0.302		3.939	*	3.250	**
	[0.245]		[2.096]		[1.396]
DEIit	−0.046		−2.832	***	−1.542
	[0.615]		[0.502]		[2.606]
DEJjt	−0.716		−2.701	*	−4.205
	[0.756]		[1.629]		[2.611]
DEBijt	1.519	*	4.647	***	7.215	**
	[0.883]		[0.693]		[3.397]
DCBij	−0.024		-		−0.306
	[0.417]				[1.224]
DCLij	0.322		-		5.010
	[0.676]				[3.850]
DCRij	0.636		-		−2.174
	[0.521]				[2.330]
Time dummies	Yes		Yes		Yes
R2	0.554
Within R2			0.192
Between R2			0.164
Overall R2			0.135
Hansen J statistics test					84.8
					(0.184)
Arellano−Bond Test for AR(1)					−5.53	***
					(0.000)
Arellano−Bond Test for AR(2)					1.16
					(0.244)
Number of instruments					97
Number of groups	97		97		97
Number of observations	1649		1649		1649

Notes: Figures in square brackets in OLS and fixed-effects model are robust standard errors clustered by country pair. Figures in square brackets in two-step system GMM are Windmeijer’s corrected standard errors. Figures in parentheses are p-values. *, **, *** indicate significant at the 10% level, 5% level, and 1% level, respectively.

present the results of the linear dynamic panel data model estimated by the two-step GMM estimation method. According to Windmeijer [45], the two-strep GMM gives lower bias and standard errors in estimating the coefficient. As a rule of thumb, we keep the number of instruments less than or equal to the number of groups. The Hansen J statistics test of overidentification restrictions is applied to check the validity of instrumental variables. In all models, the Hansen test fails to reject the null hypothesis that overidentification restrictions are valid, i.e., we cannot reject the null hypothesis that the instruments as a group are exogenous. The Arellano–Bond test with a null hypothesis of no autocorrelation is applied to the residuals of the equation in differences. The Arellano–Bond test rejects the null hypothesis of no first autocorrelation in the differenced residuals AR(1), but it fails to reject the null hypothesis in the differenced residuals AR(2). The lagged dependent variables are positive and statistically significant in all GMM estimators.

In most cases of both export and import models, the estimated coefficient of the overall country size (GDT_ijt) is positive and significant. This result suggests that an increase in the bilateral size of trading partner countries positively affects the trade flow in olive oil. The 0.122 estimated coefficient of GDT_ijt in the GMM estimator of the export model suggests that a 1% increase in overall bilateral country size will result in a 0.122% growth in olive oil exports, keeping other variables constant. Similarly, the 1.734 parameter size of GDT_ijt in the import model implies that a 1% increase in bilateral size will lead to a 1.734% increase in imports, ceteris paribus. These results are consistent with the finding of a positive impact of overall bilateral country size under the assumptions of the new trade theory [8,9,10].

More interestingly, for most of the export models, we found a negative and significant effect of the difference in relative factor endowments (RFE_ijt) at the 1% level of significance. This result suggests that olive oil exports decrease with the increase in the difference in relative factor endowments between two countries. In the GMM estimator, the estimated parameter of RFE_ijt is −0.141. This suggests that a 1% increase in the difference of relative factor endowments decreases olive oil exports by 0.141%, ceteris paribus. This finding does not support the classical H–O trade theory, which states that the difference in relative factor endowments is what drives trade. On the other hand, the difference in relative factor endowments has a positive and significant impact on olive oil imports. The relatively large size of the coefficient in the GMM estimator suggests that a 1% increase in the difference in factor endowments causes a 3.25% increase in imports.

For the export of olive oil, the coefficient of the dummy variable of EU membership for the home country (country i) (DEI_ijt) is positive and statistically significant. This result suggests exports from EU countries are greater in volume compared to those of non-EU countries. The GMM estimator found that the estimated parameter is 0.314, suggesting that exports from EU member countries are 31.4% higher compared to those of others. This finding is consistent with the fact that EU countries are competitive as major olive oil exporters. On the other hand, we found a positive and statistically significant effect of EU membership on imports if countries i and j (DEB_ijt) both join the EU. The positive sign of this parameter indicates that olive oil imports among EU members are intensive. This suggests that olive oil imports of EU countries from other EU countries are much higher compared to other combinations, including their imports from outside the EU and imports by non-EU members.

Regarding the time-invariant variables, the estimation results suggest that the coefficient of the geographical distance between trading partners (DIS_ij) does not significantly affect exports and imports. In addition, we cannot confirm the consistent and robust result of the effect of a common official language between trade partners (DCL_ij).

5. Discussion

The empirical results mentioned above are summarized in the following three points. First of all, we confirmed the significant effect of the variables that support the new trade theory, i.e., bilateral overall country size and the difference in relative factor endowments [9]. An increase in the GDPs of two countries positively affects the export and import of olive oil. More interestingly, we found the negative effect of difference in relative factor endowments on olive oil export. This result suggests that Linder’s hypothesis is relevant as the difference in relative factor endowments is not a major trigger of export. Rather, olive oil exports grow between two countries as their factor endowments become more similar. It is consistent with the results by Baltagi et al. [9] and Paas et al. [10], supporting Linder’s hypothesis that trade flows become smaller between two countries as their factor endowments grow more dissimilar. In the case of olive oil trade in Germany, Kavallari et al. [30] found a negative effect of the increase in GDP per capita on import, while the effect on the GDP per capita of its trade partner remained positive. These results imply that Germany’s olive oil importation decreases if the difference in relative factor endowments, i.e., the difference in GDP per capita of Germany and partner countries, becomes smaller. Although this study does not explicitly test Linder’s hypothesis, Germany’s case is also consistent with Linder’s hypothesis.

According to Paas et al. [10], the variable of difference in relative factor endowments would have a negative (or positive) effect on trade in situations where intra-industry (or inter-industry) trade dominates. Therefore, it implies that countries with more similar factor endowments would most likely engage in more intra-industry trade. As presented in Figure 1, the simultaneous export and import of olive oil in Mediterranean countries means that the intra-industry trade has more relevance rather than the specialization in production and export in particular countries.

Secondly, we could not confirm the relevance of Linder’s hypothesis for olive oil import. The positive coefficient of the difference in relative factor endowments suggests that Mediterranean countries import olive oil more from countries with a difference in relative factor endowments. This trade pattern of olive oil importation is consistent with the predictions of the traditional H–O trade model rather than Linder’s hypothesis. Indeed, olive oil production and, by consequence, the volume of exports fluctuate annually due to the biennial bearing of olives. Some olive oil-importing countries increase their import when their production is low in order to stabilize their domestic supply and consumption. What is more, some countries in the EU import olive oil in bulk as a raw material so that it can be processed as a domestic product for re-export. As most olive oil importers are also exporters (Table 1), we can infer that they import olive oil from countries that have different factor endowments in order to stabilize their exports to other countries that have similar factor endowments.

Thirdly, the enlargement of the EU had a positive impact on olive oil exports and imports among EU member states. Kavallari et al. [30] found similar results on Germany’s import of olive oil. This study suggests that being an EU member positively affects the import of olive oil by Germany. While the olive oil market is growing in non-traditional producer countries such as Australia, New Zealand, the US, and Chile, traditional exporter countries in EU such as Spain, Italy, and Greece are still competitive [46]. The latter group also imports olive oil from the EU members to stabilize their domestic supply and exports. These findings of trade patterns are consistent with Linder’s prediction. It is implied that trade volume is expected to be larger among EU member states whose per capita income and demand properties are more similar. As shown in Figure 1, an increase in the volume of olive oil export is observed with sizable imports among the Mediterranean countries. Among them, particularly in EU member countries, the volume of the intra-industry trade of olive oil is growing. From 1990/91 to 2016/17, the Grubel–Lloyd index was applied to data gleaned from major olive oil producers in the EU in order to measure the degree of intra-industry trade. The figure, calculated at 0.538 on average, suggests that intra-industry trade constitutes more than half of the trade flow in the region. The calculation of the index of intra-industry trade is based on Grubel and Lloyd [47,48]. The data source is World Olive Oil Figures, International Olive Oil Council, which is available from: http://www.internationaloliveoil.org/ (Accessed: 19 July 2018). During the same period, 69.6%, 46.0%, and 45.7% of trade flow was intra-industry trade occurring in Italy, France, and Spain, respectively.

6. Conclusions

The orthodox theories of international trade uphold the hypothesis that countries specialize in the production and export of goods in which they have a comparative advantage. The difference in factor price caused by the difference in factor endowments determines the pattern of trade between countries. However, actual trade flows often suggest the simultaneous import and export of goods by a particular country. This is true of olive oil trade in Mediterranean countries, where their exports grow simultaneously alongside the large volume of imports. This study investigated the factors that affect olive oil export and import in Mediterranean countries using balanced panel data from 1998 to 2016. We estimated the commodity-specific gravity model by employing POL, FEM, and REM. The results of the Hausman test, the Breuch–Pagan test, and the F-test suggest that the estimation of FEM is appropriate for the specification in both export and import models. In addition to the static analysis, we applied Arellano–Bond dynamic panel-data estimation with the two-step system GMM to solve the problem of serial correlation, heteroscedasticity, and endogeneity of some of regressors.

The empirical findings mostly support the predictions on new trade theory where an increase in the overall country size positively impacts the trade flow of olive oil. It is noteworthy that the difference in factor endowments has a negative impact on export flow. An increase in the similarity of factor endowments and, thus, in demand propensities, has a positive impact on export flow. The enlargement of the EU had a positive effect on the volume of olive oil export and import. Among EU member states whose per capita income and demand properties are similar, the volume of imports is larger than those of countries dealing in olive oil commerce outside of the EU. These results are consistent with Linder’s hypothesis as opposed to the traditional H–O trade theory. Unlike the predictions of the H–O trade theory, the difference in relative factor endowments does not seem to be a significant determinant of olive oil exportation. The specialization in olive oil production and exportation does not merely follow with comparative advantage.

The simultaneous increase in olive oil exportation alongside sizable import volume implies a growing phenomenon of intra-industry trade in the value chain of Mediterranean countries. Based on this empirical finding, this paper recommends two directions for the promotion of olive oil export. One direction is to encourage production of competitive olive oil at lower prices following the law of comparative advantage. This direction is significant for quantity expansion of olive oil production at lower costs. On the contrary, the other direction is focused on producing differentiated olive oil that responds to the growing phenomenon of intra-industry trade. Product differentiation in the interest of promoting exportation is becoming increasingly important in the context of product quality improvements that foster a competitive advantage in olive oil trade.