Gravity Model: research questions 2 & 3

The majority of scholars applied gravity model to analyze the trade impacts of TBTs. For example, Metha and Nambier (2005), Baller (2006), Disdier et al. (2005 and 2007), Yoann et al. (2014), Bao and Qiu (2010, 2012), Siyakiya (2017), and Wood et al. (2017). To address the second and third research questions, the gravity model was used. However, the gravity model includes Hecksher-Ohlin variables: market size (G), income similarity (S) (Warin et al., 2009). Both the Heckscher-Ohlin variables take the following forms:

!"#,% = '() !*+"%+ !*+#% and, S"#,% = '() 1 − (_-./-./01 012-./31) 5_-( -./31 -./312-./01) 5₎

The databases used to answer question 2- differentiated by categories, how do TBTs affect the international trade? - and research question 3- how do un-harmonized ecolabelling program impact international trade? - , are cross-section time-series databases. Each cross-section has its own individual features, which may (or may not) influence the predictor variables (Eisenhart, 1947). A Hausman test is performed to see whether time-invariant characteristics are unique to the individuals (Stock and Watson, 2003; Bartels, 2008). And regarding the data, both fixed and random effects models are tested. As a result, the best estimation technique is a set of multilevel linear regressions. As Hox and Kreft (1994) explained: "multilevel models assume a hierarchically structured population, with random sampling of groups both groups and individuals within groups". These models are linear models with (1) fixed effects to take into consideration parameters corresponding to an entire population and (2) random effects, parameters corresponding to individual units drawn at random from a population. Since multilevel models are selected, some underlying assumptions must be checked.9

The estimation technique is thus a set of multilevel models, with some temporal pseudo- replication due to the time-series cross-section (TSCS) type of the data. The Generalized Least Squares (GLS) technique (Parks, 1967) is the method that is often used with TSCS data. However, GLS technique for TSCS may produce inaccurate standard errors and violates the Gauss-Markov assumption (Beck and Katz, 1995). Indeed, in our data, each country may have its own error variance (heteroscedasticity). To deal with heteroscedasticity, dummy variables are created to represent each country. Thus, each country has its own intercept. Hsaio (1986) shows that fixed effects are suitable if one wants to make inferences to the units observed.

For validity, a set of models is tested. First, to deal with heterogeneity, the random coefficients model (RCM) is used (Beck and Katz, 2006; Swamy, 1970). Regarding our data, the RCM as the "Random Intercepts" (Model 2) is selected to add some more validity to the analysis.10 Second,

9 _{There are five fundamental assumptions for multilevel models: (1) within-group errors are independent with mean}

zero and variance σ^2, (2) within-group errors are independent of random effects, (3) random effects are normally distributed with mean zero and covariance matric ψ, (4) random effects are independent in different cross-sections, and (5) the covariance matrix does not depend on the cross-section.

10 _{The random part of the model is specified as the name of the country, which means only the intercepts vary across}

the current random model is augmented with time fixed effects. The third model is calibrated with time as a predictor of trade inflows and random intercepts across countries (see column "Time RI" - Model 3). Fourth, a next model is calibrated with the effect of time being different across countries (varying slopes across countries) (see the "Time RS" column - Model 4). The fifth model introduces a term that models the covariance structures and errors (see the "Auto Regressive" column - Model 5). The empirical analysis is based on a variant of the gravity model, commonly used to analyze bilateral trade flows. Since the dataset includes missing observations, the actual dataset is unbalanced.

Data: research question 2

To address the second research question - differentiated by categories, how do TBTs affect international trade? -, we created two databases for the countries of the case studies: China and United States of America. The databases contain the number of TBT notifications in agricultural and industrial sectors in three more important categories of TBTs: protection of human and health or safety’s, protection of the environment, and quality requirements.

The database created based on counting regulation approach, through counting TBTs notification that are issued in WTO TBT Agreement. Each regulation in TBT agreement, include the primary objective (category), and sector of the product that the notification applied. The TBT notifications are classified upon the product sectors they cover. The databases cover 96 classifications on agricultural and industrial products at the HS2-digit level. The products under HS code of 01 to 24 belong to agricultural sectors and the product under HS codes of 24 to 95 belongs to industrial sectors. Therefore, the database includes number of TBTs notifications, in 6 categories (primary objectives regarding TBT Agreement), and in two sectors of agricultural and industrial. That leads to creation of unique database on TBT notifications that give the opportunity for further studies in this matter.

The dependent variable is exports from China and the US, to the 27 country members of the EU covering the period of 2001-2015. The dependent variable is the logarithm of the exports share of exporting country (country i: China and the US) of GDP of the importing country (country j: the EU country members). The independent variables include, the length of membership of the EU country members, market size, market similarity, and distance. The exports and TBT notifications are grouped in two sectors: agricultural and industrial. One of the objectives of this

study is also to compare the impact of the TBTs in agricultural sectors with the impact of TBTs in industrial sectors. Therefore, the exports from China and the US is grouped in two sectors: agricultural and industrial as well as the TBTs notifications. The classification is made upon the HS-2Level classification.

Data: research question 3

To address the third research question- how do un-harmonized ecolabelling program impact international trade?- we designed a cross-section time-series database. The dependent variable is the exports to Canada on period of 2003-2013. The data is collected from the Statistics Canada database in Canadian dollars. The data cover export values of all categories based on the 6-digit commodity level using the harmonized system (HS).

The Independent variables include the gravity variables (distance, common border and common language), the Heckscher-Ohlin variables (market size and market similarity), variables related to harmonization program on ecolabelling (ISO 14001 and GEN) and variables regarding the trade agreements concerning ecolabelling regulations (WTO, FTA, and MRA). The model chose the dependent variable is exports to Canada in period of 2003-2013. Similarly, to answer question 3 a series of the estimation techniques is applied.

The data for geographic distance, common border and common language are obtained from CEPII.11 The data regarding the FTA and MRA are collected from Global Affairs Canada and Industry Canada.12 Information about the membership for current GEN members are available on the GEN website.13 Also, the data related to WTO are collected from the WTO database available online.14

The main difference between the two gravity models of article 2 and article 3, is in adding TBTs notifications as independent variable. To measure TBTs there are four main approaches: counting TBT notifications, overage ratio, frequency index, and price-wedge approach. Regarding the

11 _{Research and expertise on the world economy. For further information, please refer to: http://www.cepii.fr} 12 _{For further information, please refer to: http://www.international.gc.ca}

13 _{For further information, please refer to: https://globalecolabelling.net} 14 _{For further information, please refer to: http://stat.wto.org}

accessible data, we applied the counting TBT notifications approaches. As we compared the EU imports from China and the US in both sectors of agricultural and industrial, therefore we elaborate two databases on TBTs notification. The database on agricultural sectors includes TBT notifications based on TBT agreements in agricultural sector (01HS – 24HS), and the database on industrial sectors includes TBT notifications in industrial sectors (25HS-96HS).