1. Introduction
Eight multilateral rounds of negotiations under the General Agreement on Tariffs and Trade (GATT) and international agreements under the World Trade Organization (WTO) have contributed significantly to a reduction of tariffs among WTO members. However, legitimate reasons for the imposition of non-tariff measures (NTMs) within regulations have triggered their extensive use over the years. Aiming at trade liberalization, protectionist and discriminatory motives for trade policy measures are not permitted by the regulations, while some specific motives are endorsed in good faith by NTMs. Among these measures, technical barriers to trade (TBTs) and sanitary and phytosanitary (SPS) measures allow countries to impose restrictions on the import of low-quality products suspected of harming the health of domestic consumers and the global environment, or violating safety etc. Such trade policy instruments may induce higher standards in the import market, in addition to improving market efficiency via information requirements such as mandatory labelling. In this paper, we analyse the quality improvement of the imported products, which might be a general underlying motive for the imposition of different types of NTMs. Applying a monopolistic competition framework involving both the supply and the demand side of trade, we will assess the impact of different types of NTMs on the quality of traded products at the four-digit level of the Standard International Trade Classification (SITC) rev. 2. The analysis modifies and uses the existing information on NTM notifications to the WTO from the Integrated Trade Intelligence Portal (I-TIP) over the period 1995–2011.
According to the MAST (2008, p. 99)Footnote 1 classification, ‘Non-tariff measures (NTMs) are policy measures, other than ordinary customs tariffs, that can potentially have an economic effect on international trade in goods, changing quantities traded, or prices or both.’ Classifications of NTMs are mostly based on legal international regulations mandated by the WTO and other organizations. In addition, scholars have classified NTMs based on their nature and implications into two broad categories. The first category includes quantitative NTMs such as anti-dumping measures, quantitative restrictions, and safeguard measures. Despite having quantitative implications, this broad category of NTMs is sometimes based on some qualitative reasoning (e.g. national legal basis, national security, health and environment issues, market adjustments). The second category refers to NTMs that have qualitative implications. TBTs and SPS measures are the core NTM category that aims to achieve better regulations and higher standards. Irrespective of the complex motives behind such trade policy measures – i.e. following good faith and legitimate motives, unlike discriminatory motives – they are basically caused by technology, domestic standards and innovations, and qualitative, health, and environmental issues (Ghodsi, Reference Ghodsi2018). Therefore, these core qualitative or regulative NTMs (i.e. TBTs and SPS measures) are considered to also have qualitative upgrading effects on trade flows.
Such core NTMs are aimed at improving the quality of the imported product to align it with domestic standards. Standard-based regulations can potentially improve production procedures or the quality of products (Wilson and Otsuki, Reference Wilson and Otsuki2004; Trienekens and Zuurbier, Reference Trienekens and Zuurbier2008; Ing and Cadot, Reference Ing and Cadot2017). The various impacts of NTMs on trade values and quantities have already been studied (Ronen, Reference Ronen2017a). For instance, using a gravity model on traded HS six-digit products, Essaji (Reference Essaji2008) found that the technical regulations imposed by the US resulted in a huge cost to poor exporting countries with lower capacities. Using the data on TBT notifications to the WTO, Bao and Qiu (Reference Bao and Qiu2012) found that these regulations reduce the export-extensive margins while increasing the intensive margins. In a computable general equilibrium (CGE) framework, Francois et al. (Reference Francois, Manchin and Norberg2011) analysed trade liberalization gains from preferential trade agreements. They found that a reduction in NTMs would have a much larger impact than a tariff reduction. Disdier et al. (Reference Disdier, Fekadu, Murillo and Wong2008), Li and Beghin (Reference Li and Beghin2012), Yousefi and Liu (Reference Yousefi and Liu2013), and Ghodsi (Reference Ghodsi2020a) also found evidence of a negative impact of core NTMs on trade flows while Ronen (Reference Ronen2017b) finds a trade-stimulative impact of quality NTMs imposed on virgin olive oil. Several other studies in the literature have analysed the trade restrictiveness of NTMs at the HS six-digit level by estimating the ad-valorem equivalent of NTMs (Kee et al., Reference Kee, Nicita and Olarreaga2009; Beghin et al., Reference Beghin, Disdier and Marette2015; Cadot and Gourdon, Reference Cadot and Gourdon2016; Ghodsi et al., Reference Ghodsi, Gruebler and Stehrer2016; Bratt, Reference Bratt2017; Cadot et al., Reference Cadot, Gourdon and Van Tongeren2018; Niu et al., Reference Niu, Liu, Gunessee and Milner2018). While these studies provide evidence of the price equivalence of NTMs to make them comparable to tariffs, the literature regarding the quality impact of core regulative NTMs is still lacking.
Therefore, this paper extends the literature by focusing specifically on the role of two types of qualitative NTMs, i.e. TBTs and SPS measures, notified to the WTO during the period 1995–2011 regarding the quality of products traded bilaterally at the SITC four-digit level. The quality of traded products is measured using the theoretical framework developed by Feenstra and Romalis (Reference Feenstra and Romalis2014). It is important to note that the legitimate motive behind the imposition of TBTs and SPS measures is to improve the quality of products traded to a country when these products may harm, for example, human health, plant life, and environmental quality, or violate consumer's safety and protection. Therefore, the results of this analysis may provide a better understanding of whether these regulative NTMs fulfil their intended purpose.
The paper is structured as follows. In the next section, we summarize the theoretical model for the calculation of the quality index for products traded bilaterally, as developed by Feenstra and Romalis (Reference Feenstra and Romalis2014). This methodology provides a framework to disentangle quantity, price, and quality effects of bilateral trade flows from traded values. Section 3 discusses data issues and the econometrics specification. Section 4 provides a discussion of the estimation results, and Section 5 concludes.
2. The Theoretical Framework of the Quality Index
The starting point of the analysis is the model presented in Feenstra and Romalis (Reference Feenstra and Romalis2014) – subsequently referred to as F&R (2014) – which provides a framework to disentangle quantity, quality, and price effects of exports and imports. Here the intuition of the model is presented allowing for a proper interpretation of results concerning the econometric outcomes of the effects of NTMs.Footnote 2
The model starts from an expenditure function given by
$$E^k = U^k\left[{\mathop \smallint \limits_i {\left({\displaystyle{{\,p_i^k } \over {{( {z_i^k } ) }^{\alpha^k}}}} \right)}^{1-\sigma }di} \right]^{{1 \over {1-\sigma }}} = U^k\left[{\mathop \smallint \limits_i {\left({\displaystyle{{\,p_i^k } \over {z_i^k }}\displaystyle{1 \over {{( {z_i^k } ) }^{\alpha^k-1}}}} \right)}^{1-\sigma }di} \right]^{{1 \over {1-\sigma }}}$$
(with σ > 1) implying non-homothetic demand for quality for α k(U k) ≥ 1. The price 
$p_i^k $ of good i sold in market k is divided by quality 
$z_i^k \;$, which allows us to model the consumer decision in quality-adjusted prices and quantities. The quality-adjusted price is denoted by 
$P_i^k : = p_i^k /z_i^{\alpha ^k} $, where for brevity we set 
$( {z_i^k } ) ^{\alpha ^k}: = z_i^{\alpha ^k}. \;$ Note that the quality-adjusted price depends on both the level of quality 
$z_i^k $ and how consumers evaluate quality α k. Both lead to lower quality-adjusted prices. Correspondingly, quality-adjusted demand is denoted by 
$Q_i^k : = z_i^{\alpha ^k} q_i^k $. The validity of the expenditure function is shown in F&R (2014). From the above, we can also see that quality-adjusted demand increases with quality.
It is assumed that firms can produce multiple products (one for each market), and that firm h in country r simultaneously chooses quality 
$z_{ih}^{rk} $ and the freight-on-board (f.o.b.) price 
$p_{ih}^{{\rm fob}, rk} $ to sell in market k. Further, the production function for quality 
$z_{ih}^{rk} $ is assumed to be Cobb–Douglas given by productivity of labour 
$\varphi _{ih}^{rk} $ and amount of labour 
$l_{ih}^{rk} $ used: 
$z_{ih}^{rk} = ( {l_{ih}^{rk} \varphi_{ih}^{rk} } ) ^\theta $ with 0 < θ < 1 reflecting diminishing returns to quality. The wage rate for (the composite) input 
$l_{ih}^{rk} $ is given by w r. Factor demand therefore is 
$l_{ih}^{rk} = ( {z_{ih}^{rk} } ) ^{1/\theta }/\varphi _{ih}^{rk} $ and total per-unit variable costs are 
$w^rl_{ih}^{rk} = w^r( {z_{ih}^{rk} } ) ^{1/\theta }/\varphi _{ih}^{rk} $. Further, firms pay fixed costs of exporting given by 
$f_{ih}^{rk} ( {\varphi_{ih}^{rk} } ) $, i.e. depending on productivity. Productivity levels are assumed to be Pareto-distributed, with 
$G_i^r ( {\varphi_i} ) = 1-\left({{{\varphi_i} \over {\varphi_i^r }}} \right)^{-\gamma _i}$, where 
$\varphi _i < \varphi _i^r $ (φi is the lower bound of productivities in country r).Footnote 3
Concerning trade costs, the assumption is that there are both specific (per-unit) trade costs denoted by 
$T_i^{rk} $ and ad-valorem trade costs 
$\tau _i^{rk} $. Tariffs might be included and considered similarly denoted by 
$t_i^{rk} $. Likewise, other ad-valorem costs (e.g. AVEs of NTMs) might be part of the specific trade costs or enter as tariff-equivalents. These trade costs are applied to the value, including the specific trade costs, giving the c.i.f. price (including tariffs) asFootnote 4
$$p_{ih}^{{\rm cif}, rk} = ( {1 + t_i^{rk} } ) ( {1 + \tau_i^{rk} } ) ( {\,p_{ih}^{{\rm fob}, rk} + T_i^{rk} } ) $$
The marginal costs are the same as the total costs for producing one unit of a good with quality 
$z_{ih}^{rk} $, i.e. 
$c_{ih}^{rk} ( {z_{ih}^{rk} , \;w^r} ) = w^rl_{ih}^{rk} = w^r( {z_{ih}^{rk} } ) ^{1/\theta }/\varphi _{ih}^r $. These are thus increasing in the wage rate w r and the quality 
$z_i^{rk} $, and decreasing in productivity 
$\varphi _{ih}^r $. The firm maximization problem is thus given by
$$\mathop {\max }\limits_{\,p_{ih}^{{\rm fob}, rk} , z_{ih}^{rk} } \left[ {\,p_{ih}^{{\rm fob}, rk} -w^r{( {z_{ih}^{rk} } ) }^{1/\theta }/\varphi_i^r } \right] \displaystyle{{( {1 + \tau_i^{rk} } ) q_{ih}^{rk} } \over {( {1 + t_i^{rk} } ) }}$$This can be reformulated in quality-adjusted terms and to tariff-exclusive c.i.f. prices, which can be rewritten in quality-adjusted c.i.f. prices net of tariffs (see Appendix)
$$\mathop {\max }\limits_{P_i^{{\rm cif}, rk} , z_i^{rk} } \left[{P_{ih}^{{\rm cif}, rk} -( {1 + \tau_i^{rk} } ) \displaystyle{{\left({\displaystyle{{w^r{( {z_{ih}^{rk} } ) }^{{1 \over \theta }}} \over {\varphi_{ij}^r }} + T_i^{rk} } \right)} \over {{( {z_{ih}^{rk} } ) }^{\alpha_k}}}} \right]\displaystyle{{Q_{ih}^{rk} } \over {( {1 + t_i^{rk} } ) }}$$The assumption of a Cobb–Douglas production function and the resulting cost function results in a log-linear form of the optimal quality choice – see F&R (2014) for the derivation
$$z_{ih}^{rk} = \left({\displaystyle{{T_i^{rk} \displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \over {w^r/\varphi_{ih}^r }}} \right)^\theta $$Thus, quality is increasing with higher specific trade costs (referred to as the ‘Washington apples effect’) of which NTMs might be a part, higher productivity, and higher parameter values α kθ Footnote 5. If α k is increasing with income, richer countries import higher qualities. If θ (depending on the exporter r) is larger, the returns to quality diminish less quickly and thus quality is increasing. Conversely, quality is decreasing with higher wages, i.e. higher costs of production.
The marginal costs become proportional to the specific trade costs
$$c_{ih}^{rk} \left( {z_{ih}^{rk} , \;w^r} \right) = \left[{\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right]T_i^{rk} $$These are increasing in α k (as quality increases and therefore marginal costs increase) and the specific trade costs. The assumption of the CES expenditure function and the optimal choice of the f.o.b. price yield the familiar mark-up equation
$$\left( {\,p_{ij}^{{\rm fob}, r\;k} + T_i^{rk} } \right) = \left[ {c_{ih}^{rk} \left( {z_{ih}^{rk} , \;\;w^r} \right) + T_i^{rk} } \right] \left({\displaystyle{\sigma \over {\sigma -1}}} \right)$$Using the proportionality of marginal costs and specific trade costs gives f.o.b. and c.i.f. (inclusive tariffs) prices
$$p_{ih}^{{\rm fob}, rk} = T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right] = : \overline {\,p_i^{{\rm fob}, rk} } $$
$$p_{ih}^{{\rm cif}, rk} = \left( {1 + t_i^{rk} } \right) \left( {1 + \tau_i^{rk} } \right) T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)} \right] = : \overline {\,p_i^{{\rm cif}, rk} } $$Thus, the prices do not depend on firm productivity, as more efficient firms sell higher-quality products.Footnote 6 The implication of these assumptions is that all firms selling to market k charge the same price but only differ with respect to quality.
Finally, it can be shown that the quality index is related to the log f.o.b. price
$$z_{ih}^{rk} = \left({\displaystyle{{\kappa_1^k \overline {\,p_i^{{\rm fob}, rk} } } \over {w^r/\varphi_{ih}^r }}} \right)^\theta = \left({\displaystyle{{\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right]} \over {w^r/\varphi_{ih}^r }}} \right)^\theta $$
with 
$\kappa _1^k = {{\alpha ^k\theta ( {\sigma -1} ) } \over {1 + \alpha ^k\theta ( {\sigma -1} ) }}$. Thus, quality is increasing with the specific trade costs (which might include the costs of NTMs) and productivity.
Let 
$\hat{\varphi }_i^{rk} $ denote the cut-off productivity of the marginal exporter (i.e. the firm just covering the fixed costs of exporting). The c.i.f. (including tariffs) quality-adjusted price for the marginal exporter is defined as 
$\hat{P}_i^{{\rm cif}, rk} : = {{p_i^{{\rm cif}, rk} } \over {{( {z_i^{rk} ( {\hat{\varphi }_i^{rk} } ) } ) }^{\alpha ^k}}}$, which after inserting yields
$$\hat{P}_i^{{\rm cif}, rk} = \displaystyle{{\,p_i^{{\rm cif}, rk} } \over {{\left( {z_i^{rk} \left( {\hat{\varphi }_i^{rk} } \right) } \right) }^{\alpha ^k}}} = \displaystyle{{\,p_i^{{\rm cif}, rk} } \over {{\left({{\left({\displaystyle{{\kappa_1^k \overline {\,p_i^{{\rm fob}, rk} } } \over {w^r/\hat{\varphi }_i^{rk} }}} \right)}^\theta } \right)}^{\alpha ^k}}} = p_i^{{\rm cif}, rk} \left({\displaystyle{{w^r/\hat{\varphi }_i^r } \over {\kappa_1^k \overline {\,p_i^{{\rm fob}, rk} } }}} \right)^{\alpha ^k\theta }$$
which includes tariffs. 
$\hat{X}_i^{rk} = \hat{P}_i^{{\rm cif}, rk} \hat{Q}_i^{rk} $ is the (tariff-inclusive) export revenue. Firm profits are given by
$$\displaystyle{{\hat{X}_i^{rk} } \over {( {1 + t_i^{rk} } ) }}\displaystyle{1 \over \sigma } = f_i^{rk} ( {\hat{\varphi }_i^{rk} } ) $$which covers fixed costs. Assuming a special function for fixed costs as argued in F&R (2014)
$$f_i^{rk} ( {\hat{\varphi }_i^{rk} } ) = \left({\displaystyle{{w^r} \over {\hat{\varphi }_i^{rk} }}} \right)\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)^{\beta _0}{\rm exp}( {\beta^{\prime}F_i^{rk} } ) $$(with β 0 > 0) allows us to derive the quality-adjusted c.i.f. price (tariff-inclusive) (under the assumption of homogeneous firms)
$$\hat{P}_i^{{\rm cif}, rk} = \left[{\displaystyle{{\overline {\,p_i^{{\rm cif}, rk} } } \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\overline {\,p_i^{{\rm fob}, rk} } } \right)}^{\alpha^\kappa \theta }}}} \right]\left[{\displaystyle{{X_i^{rk} } \over {\sigma ( 1 + t_i^{rk} ) N_i^r }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta^{\prime}F^{rk}} ) } \right]^{\alpha ^\kappa \theta }$$which after some manipulations (see Appendix) can be written as
$$\hat{P}_i^{{\rm cif}, rk} = \left[{\displaystyle{{{( {1 + t_i^{rk} } ) }^{1-\alpha^\kappa \theta }( {1 + \tau_i^{rk} } ) \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)} \right]{( {T_i^{rk} } ) }^{1-\alpha^\kappa \theta }} \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right]} \right)}^{\alpha^\kappa \theta }}}} \right]\;\;\left[{\displaystyle{{X_i^{rk} } \over {\sigma N_i^r }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta^{\prime}F^{rk}} ) } \right]^{\alpha ^\kappa \theta }$$
The quality-adjusted price is decreasing with 
$\kappa _1^k $ (which is increasing in its arguments) and a larger f.o.b. price, and is increasing with a larger c.i.f. price. In the second term, it is decreasing with tariffs, σ, and the number of exporters. It is further decreasing with the size of the market and the fixed costs. The value of exports 
$X_i^{rk} $ and the quality-adjusted price 
$\hat{P}_i^{{\rm cif}, rk} $ are positively related – in contrast to the demand-side interpretation. A similar equation holds in the case of heterogeneous firms (see Apppendix).
From the CES demand it would follow that
$$\hat{X}_i^{rk} = \displaystyle{{X_i^{rk} } \over {N_i^r }} = \left({\displaystyle{{\hat{P}_i^{{\rm cif}, rk} } \over {P^k}}} \right)^{-( {\sigma -1} ) }Y^k$$i.e. a higher quality-adjusted price results in lower export values. Using supply-side information results in an export equation that is close to a gravity equation.
$$\displaystyle{{X_i^{rk} } \over {M_i^r {\left({\displaystyle{{\varphi_i^r } \over {w^r}}} \right)}^\gamma }} = \left({\displaystyle{{\overline {P_i^{{\rm cif}, rk} } } \over {P^k}}} \right)^{-( {\sigma -1} ) ( {1 + \gamma } ) }( {Y^k} ) ^{1 + \gamma }\left({\sigma \kappa_2^k \left( {1 + t_i^{rk} } \right) {\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{\beta_o}\exp ( {\beta^{\prime}F_i^k } ) } \right)^{-\gamma }$$F&R (2014) used equation (16) for two representative countries, r and j, exporting to market k, which could equally be used for two different markets, k and l, as export destinations for a representative country i. Therefore, equation (16) is modified to be estimated using GMM to calculate the unknown parameters of the model. With the estimated parameters, we can further calculate the quality index in equation (9). For the sake of simplicity, the quality preference parameter of the US is assumed to be equal to 1, and other countries’ preferences are then calculated relative to the US with iterated estimations.
Further, the model suggests that NTMs (TBTs or SPS measures) impact positively on the quality of traded products if these enter as specific trade costs.
3. Data and Specifications
3.1 Regression Framework
The analysis is conducted for a sample of countries over the period 1995–2011. The sample includes all 151 countries that were WTO members in 2011, mainly because the NTM database includes the notifications of the WTO members. The primary data source on product quality is the index quality of exports derived in F&R (2014), which are downloadable from their website.Footnote 7 These data provide information for bilateral flows of goods at the four-digit SITC rev. 2 (782 products) over the period 1984–2011. Based on the theoretical framework outlined in F&R (2014) discussed above, GMM estimations are performed for each of the 712 products to estimate the relevant elasticities with other parameters partly taken from the literature. The quality index is thus disentangled from quality-adjusted prices and quality-adjusted quantities in the values of trade, which is taken from F&R (2014) available on Feenstra's websiteFootnote 8.
$$\ln z_{ht}^{rk} = \alpha _0 + \alpha _1\ln ( {1 + t_{ht}^{rk} } ) + \alpha _2TBT_{ht}^{rk} + \alpha _3SPS_{ht}^{rk} + \alpha _4PTA_t^{rk} + \alpha _5EU_t^{rk} + \mu _{St}^r + \mu _{St}^k + \mu _h^{rk} + \varepsilon _{ht}^{rk} $$
where 
$\ln z_{ht}^{rk} $ is the log of the average quality index of (all exporting firms) exporting product h from country r to importing country k in year t; 
$\ln ( {1 + t_{ht}^{rk} } ) $ is the log of tariffs plus 1; 
$TBT_{ht}^{rk} $ and 
$SPS_{ht}^{rk} $ represent four different proxies for TBTs and SPS measures included in the analysis (see discussion below) imposed by the importing country on the export of the product in that year; 
$PTA_t^{rk} $ is a dummy variable indicating whether the two countries have a preferential trade agreement (PTA)Footnote 9 in that year; 
$EU_t^{rk} $ is a dummy variable indicating whether the two countries are both members of the EU in that year. In order to control for the technological change across firms in the same sector and in the production side of the exporting country, exporter–sector–time fixed effects (FE) 
$\mu _{St}^r $, where S indicates the three-digit sector, are included; moreover, importer–sector–time FE 
$\mu _{St}^k $ are included to control for demand-side characteristics. Thus, using these two sets of FE, time-varying country-level characteristics such as size, capital, and factor endowments of the economy are controlled for. Furthermore, 
$\mu _h^{rk} $ is the bilateral-product FE that controls for any time-invariant characteristics inherent in the bilateral trade flows at the four-digit product level in addition to other gravity variables such as distance, common border, and historical relations. The whole set of fixed effects controls for multilateral resistance terms elaborated in the gravity framework (Anderson and Van Wincoop, Reference Anderson and Van Wincoop2003); moreover, 
$\varepsilon _{ht}^{rk} $ is the error term.
The estimation is run using the Ordinary Least Squares (OLS) on the whole sample of bilateral products over the period 1995–2011. Estimators are robust against heteroscedasticity in the error term. Since qualitative NTMs may have a heterogenous impact on the quality of products in different sectors, the sample of estimations is also separated in ten one-digit SITC sectors, including all bilateral flows of four-digit products, and the estimations are run separately for each sector. As a robustness test, the analysis is run on first-lagged independent variables to control for the endogeneity bias due to the reverse causality, which is available on request.
Furthermore, in a separate model, we analyse how the effect of NTMs might vary across exporting countries.Footnote 10 The reason for this is that a country such as the US might already be exporting very high quality such that TBTs and/or SPSs are non- binding, whereas China, may initially produce low quality. In such a case, the same NTM applied to the US or China might have very different effects. The impact of the NTM may vary with the initial quality level of the product. While the initial quality would be absorbed by the exporter–sector–time effects, this country-specific effect would not. Therefore, to explore for some heterogeneity in the impact, the NTM variables are interacted with exporting countries μ r as follows:
$$\ln z_{ht}^{rk} = \alpha _0 + \alpha _1\ln ( {1 + t_{ht}^{rk} } ) + \mathop \sum \limits_r \alpha _2\mu ^rTBT_{ht}^{rk} + \mathop \sum \limits_r \alpha _3\mu ^rSPS_{ht}^{rk} + \alpha _4PTA_t^{rk} + \alpha _5EU_t^{rk} + \mu _{St}^r + \mu _{St}^k + \mu _h^{rk} + \varepsilon _{ht}^{rk} $$where α 2μ r and α 3μ r show the exporter-specific impact of TBT and SPS measures separately. To avoid exhausting the degree of freedom, including all interactions with both NTM variables, and collinearity, these exporter-specific effects must be estimated in two separate estimations. In one estimation, one of the regulative NTMs is interacted with the exporter dummies, while keeping the other regulative NTM as control variable. The results of these exporter-specific impact of NTMs on quality of traded goods are provided in the online appendix. Moreover, the exporting countries are grouped according to their income-levels defined by the World Bank, and their groupsFootnote 11 are interacted with the regulative NTMs to provide more insights. The results, which are provided in the appendix, help to explain the variation in the sector results if the distribution of NTMs across target country groups varies across the different subsamples (i.e. one sector may be characterized by many NTMs, whereas in others the main focus of NTMs is on low quality exporters).
3.2 Measurement of Qualitative NTMs
Four types of measurement of TBTs and SPS measures are used in this analysis in four separate models. In Model 1, the simplest measure, which is often used in several studies in the literature (e.g. Kee et al., Reference Kee, Nicita and Olarreaga2009; Beghinet al., Reference Beghin, Disdier and Marette2015; Bratt, Reference Bratt2017; Niu et al., Reference Niu, Liu, Gunessee and Milner2018), uses a dummy variable that takes the value of 1 when an importing country k imposes any qualitative NTMs against the import of product h from exporting country r in year t. Thus, this is a dummy variable on the flows of NTMs, and for TBTs and SPS measures it is shown respectively as 
$TBT_{rkht}^{DF} $ and 
$SPS_{rkht}^{DF} $.
Previous studies in the literature have relied on cross-sectional data analysis, whereas we apply a panel-data analysis here. Thus, it matters whether an NTM, which was imposed in previous years, is still in force and has not been withdrawn. Therefore, as the second measure of NTMs and in Model 2, we use a dummy variable on stocks of qualitative NTMs that takes the value of 1 when there exists any qualitative NTM that was in force until year t and has not yet been withdrawn, and was notified by the importing country k against the exports of product h from exporting country r. Thus, the dummy variables on the stock of TBTs and SPS measures are shown respectively as 
$TBT_{rkht}^{DS} $ and 
$SPS_{rkht}^{DS} $.
To make standards and regulations more stringent, authorities may impose several NTMs on a given product to achieve the highest quality (Ing and Cadot, Reference Ing and Cadot2017; Cadot et al., Reference Cadot, Gourdon and Van Tongeren2018). To expand the impact of the existence of any qualitative NTMs on the product quality to the impact of stringency of these NTMs on quality, in Model 3 we use the count measures of NTMs. Therefore, we expand the dummy variable on flows of NTMs to the count variable of flows of the total number of NTMs notified in a given year t by the importing country k reporting the product h. The variable for TBTs and SPS measures is thus included as 
$TBT_{rkht}^{CF} $ and 
$SPS_{rkht}^{CF} $, respectively.
It might be the case that it takes time to adjust the quality of a product to ensure that it complies with the new regulative NTMs imposed in each year. Therefore, the count variable of stocks of the total number of NTMs might indicate a stronger impact on the average quality of the products at the sector level than the count flows variable might have. Consequently, as the benchmark measure in Model 4 we use the count variable of stocks of the existing number of NTMs that is calculated as the accumulated number of NTMs that came into force until year t and have still not been withdrawn, which were notified by the importing country k against the exports of products h. This count variable of stocks for TBTs and SPS measures is included as 
$TBT_{rkht}^{CS} $ and 
$SPS_{rkht}^{CS} $, respectively. This variable has been used in earlier studies, such as Ghodsi et al. (Reference Ghodsi, Gruebler and Stehrer2016), Ghodsi et al. (Reference Ghodsi, Grübler, Reiter and Stehrer2017), Ghodsi and Stehrer (Reference Ghodsi and Stehrer2019), and Ghodsi (Reference Ghodsi2020a, Reference Ghodsi2020b).
Furthermore, for both stocks and flows measures, the average number of measures per six-digit product at HS within a four-digit SITC is calculated. Thus, in Model 5, this count variable of flows for TBTs and SPS measures is included as 
$TBT_{rkht}^{AVG, CF} $ and 
$SPS_{rkht}^{AVG, CF} $, respectively. In Model 6, this count variable of stocks for TBTs and SPS measures is included as 
$TBT_{rkht}^{AVG, CS} $ and 
$SPS_{rkht}^{AVG, CS} $, respectively.
Finally, EU Member States can impose unilateral NTMs that affect third-party countries. However, because of the mutual recognition clause of the single market agreement these regulations cannot affect intra-EU trade but can only affect extra-EU trade. Therefore, due to both the harmonization and the mutual recognition of trade policy measures, regulations and standards within the EU, bilateral tariffs, and non-tariff measures imposed against intra-EU trade are set to zero. Furthermore, the dependent variable does not include zeros as the price of a traded good cannot take the value of zero. Therefore, it is not meaningful to include zero trade flows in the regression on quality or price of traded goods.Footnote 12
3.3 Data
The quality framework presented above considers both the demand and the supply side of markets, which improves the former frameworks proposed by Hallak and Schott (Reference Hallak and Schott2011) and Khandelwal (Reference Khandelwal2010). The following analysis is based on the quality of imported products measured by the framework discussed above.
The data on explanatory variables are collected from various sources. The data on tariffs are taken from F&R (2014), which include preferential rates and Most-Favoured Nations (MFN) tariffs wherever applicable. Two types of NTMs are included: TBTs and SPS measures allow countries to impose restrictions on imports of low-quality products suspected to harm domestic consumers’ health, plant life, the environment, etc. It is expected that these core NTMs induce higher standards in the import market, in addition to improving market efficiency via information requirements such as mandatory labelling. TBTs and SPS measures are usually imposed unilaterally on the imports from all other countries in the worldFootnote 13. The data on NTMs are collected from the WTO I-TIP database. The data have many missing HS codes, which are improved, harmonized, and matched to the trade data using the approach in Ghodsi et al. (Reference Ghodsi, Grübler, Reiter and Stehrer2017). The data on PTAs are borrowed from Foster and Stehrer (Reference Foster and Stehrer2011).
Table A1 in the appendix provides summary statistics of the variables used in the estimations for the various Models 1–6. Table A2 in the appendix provides summary statistics of TBT and SPS measures used in Model 5 and Model 6 for the nine aggregate SITC groups. As one should expect, TBT measures are imposed on the imports of goods from all sectors. However, one may expect that SPS measures are imposed only on food and edible products. An SPS measure that should protect human health and safety against hazardous and toxic use of chemicals in the cooling system of automobiles may even target products in the transports industry, which is part of the machinery and transport equipment sector (SITC 7). This table indicates that SPS measures are imposed on many product groups. The average number of stocks of SPS measures across six-digit products within four-digit SITC sectors is largest in food and live animals mainly for food (SITC 0), then in animal and vegetable oils, fats, and waxes (SITC 4), chemicals and related products, n.e.s. (SITC 5), and then beverages and tobacco (SITC 1); pharmaceutical products and medicines ( SITC 5) tare usually targeted by SPS measures. Moreover, one can observe that SPS measures are also imposed on other sectors such as crude materials, inedibles, except fuels (SITC 2) that could be used as intermediate inputs in many edible products.
4. Results
In this section, we present the results of the econometric analysis. First, we start with the overall results, which include bilateral trade flows of all products in the regressions. In the second sub-section, we present the results of estimations for each one-digit SITC group encompassing all its bilateral four-digit products.
4.1 The Whole Sample
The estimation results for the whole sample are presented in Table 1. Overall, the regressions perform quite well, as indicated by a high R-squared. The Akaike information criterion (AIC) and Bayesian information criterion (BIC) both indicate that Model 6 has the best fit. As explained above, this model includes count variables on stocks of TBTs and SPS measures averaged from the six-digit level of HS to four-digit level of SITC. Therefore, this model is taken as the benchmark specification. In all models, coefficients of tariffs have statistically insignificant coefficients. PTA has strongly positive coefficients across all models, and EU has a strongly negative coefficients across many models. The negative coefficient shows that the countries entering the EU in 2004 or 2007 are characterized by lower quality levels in their exports and imports from other countries that were already EU members.
Table 1. Estimation results on quality of SITC products traded bilaterally during the period 1995–2011
Notes: Model 1 includes dummy variables of flows of NTMs; Model 2 includes dummy variables of stocks of NTMs; Model 3 includes count variables of flows of NTMs; Model 4 includes count variables of stocks of NTMs; Model 5 includes count variables of flows of NTMs averaged across all HS six-digit products within the four-digit SITC; Model 6 includes count variables of stocks of NTMs averaged across all HS six-digit products within the four-digit SITC as the benchmark specification.
Standard errors in parentheses; *p < 0.1; **p < 0.05; ***p < 0.01.
Source: Authors’ estimation of equation (17) on quality index compiled from F&R (2014); NTMs data from Ghodsi et al. (Reference Ghodsi, Grübler, Reiter and Stehrer2017).
In all models, the TBT measure strongly affects the quality of traded products. Since country-sector-time fixed effects (with sector at three-digit level) are included in the regressions, what remains of the impact of NTMs on quality of traded goods is through the variations of NTMs and quality of traded goods at the four-digit level within each three-digit level sector. This impact is statistically significant at the 1% level across all models except model 1. According to the estimation results of Model 2, the existence of a TBT on a product traded bilaterally can increase its quality by 1.3%, though according to the results obtained from Model 4 an additional TBT imposed on a product traded bilaterally can increase its quality by 0.08% only. In model 6, this impact is about 0.14%. In this respect, it is worth noting that the average number of stocks of TBTs imposed on bilaterally traded products in the sample is about 1.78, and the maximum number of stocks of TBTs imposed on a bilateral four-digit SITC product is 155 (see Table A1 in the Appendix). However, the average value of the dummy variable on stocks of TBTs shows that only about 30% of all products traded bilaterally report an existing TBT. This comparison indicates that regulative TBTs on one-third of observations are very stringent, which gives a statistically significant coefficient of TBTs in Model 4. This could also explain the much smaller effect on quality (i.e. 1.3% compared with 0.09%).
The estimation results of Model 1 and Model 3 including flows of TBTs also show that when TBTs are imposed in each year, a higher quality of traded product is expected in that same year too. The impact of TBTs in Model 1 is smaller than in Model 2, which suggests that the existence of stocks of TBTs that have remained in force over time has a stronger impact on the quality of traded products than the newly imposed flows of TBTs in each year. However, the impact of TBTs in Model 3 (Model 5) is stronger than in Model 4 (Model 6). This indicates that the stringency of a regulative TBT on a traded product has the strongest impact in the year in which it is imposed, rather than when the regulative TBT remains in force and gets accumulated over time.
The results also indicate a positive impact of SPS measures on the quality of traded products. While SPS coefficients are positive, the estimation results indicate that the level of significance of SPS coefficients is strongest in Model 1, that it is gradually weakened through Model 2, and that it becomes statistically insignificant in other models. This indicates that the existence of an SPS measure has the strongest positive impact on the quality of a traded product in the year in which it is imposed. The existence of SPS measures which were imposed until a specific year that is measured as a stock dummy variable also has a positive impact on the quality of traded products, but this impact is statistically significant at a 5% level only. Using the number of flows of SPS measures as an indicator of stringency of these standard-like measures in Model 3 and Model 5 has a weakly significant effect on the quality of a traded product. Therefore, these findings point to the fact that the existence of an SPS measure that protects human health and safety is the most important factor for the quality improvement of traded products, while the impact over time (i.e. proxied in stock measure) and stringency (i.e. proxied in count measure) on the quality of traded products fades.Footnote 14
4.2 Quality Impact of Regulative NTMs across Sectors
This sub-section provides an overview of the estimation results on the quality of traded products at the four-digit level of SITC in each sample of a one-digit SITC sector based on Model 6 specification. These results are presented in Table 2. Moreover, the point estimates and confidence intervals of TBT and SPS coefficients are presented graphically in Figures 1 and 2, respectively.
Table 2. Estimation results on quality of SITC products traded bilaterally during the period 1995–2011 across SITC sectors, Model 6
Note: Model 6 is used here that includes count variables of stocks of NTMs averaged across all HS six-digit products within the four-digit SITC as the benchmark specification.
Standard errors in parentheses; *p < 0.1; **p < 0.05; ***p < 0.01.
Sources: Authors’ estimation of equation (17) on quality index compiled from F&R (2014); NTMs data from Ghodsi et al. (Reference Ghodsi, Grübler, Reiter and Stehrer2017).
Figure 1. Point estimates and confidence intervals of TBT coefficients in Model 6 across sectors
Note: The diamond shows the point estimates, the bar around the diamond shows the confidence intervals, the darkest colour in the bar indicates 10% level of significance, and the brightest colour indicates 1% level of significance.
Source: Estimations presented in Table 2.
Figure 2. Point estimates and confidence intervals of SPS coefficients in Model 6 across sectors
Note: The diamond shows the point estimates, the bar around the diamond shows the confidence intervals, the darkest colour in the bar indicates 10% level of significance, and the brightest colour indicates 1% level of significance.
Source: Estimations presented in Table 2.
As is shown in Appendix Table A2, the number of TBTs and SPS measures imposed in each sector varies with different means and standard deviation. According to the estimation results presented in Table 2, in almost all other sectors, a positive impact of these regulative measures is observed. TBTs affect the quality of traded products positively and statistically significantly in the ‘food and live animals chiefly for food’, ‘mineral fuels, lubricants and related materials’, ‘animal and vegetable oils, fats and waxes’, ‘chemicals and related products, n.e.s.’, ‘manufactured goods classified chiefly by material’, and ‘machinery and transport equipment’.
For traded products in SITC groups ‘beverages and tobacco’, ‘animal and vegetable oils, fats and waxes’, ‘chemicals and related products, n.e.s.’, and ‘machinery and transport equipment’ industries SPS measures show to be positively affecting the quality. One can understand that SPS measures are imposed on machineries to secure the health issues regarding these products that improve their traded quality significantly.
4.2 Quality Impact of Regulative NTMs across Sectors and Countries
Table A3 in the appendix presents the results of estimations on quality of traded goods across SITC sectors with the impact of NTMs varying across exporting country groups. These country groups are based on the definition of the World Bank. Furthermore, estimation results on a more detailed exporter-specific impact of NTMs are presented in Tables A4 and A5 in the online appendix. Table A3 shows that the impact of the two types of NTMs on quality of traded goods from various exporting groups is very heterogeneous. For instance, TBT measures improve the quality of exported goods from high-income countries for food and live animals (SITC 0), chemicals and related products, n.e.s. (SITC 5), and machinery and transport equipment (SITC 7), while they reduce the quality of exports of beverages and tobacco (SITC 1) from these advanced economies. In contrast, SPS measures do not reduce the quality of any goods exported from high-income countries statistically significantly, while they even improve the quality of beverages and tobacco (SITC 1) from these advanced economies. The positive elasticity of quality of trade in beverages and tobacco (SITC 1) from these advanced economies with respect to SPS measures is even much larger than the negative elasticity of quality of trade in these goods with respect to TBT measures.
The results also suggest that TBT and SPS measures improve the quality of exports from upper-middle and lower-middle income countries for many sectors. TBTs improve the quality of exports of animal and vegetable oils, fats, and waxes (SITC 4) and chemicals and related products, n.e.s. (SITC 5) from low-income countries significantly. SPS measures improve the quality of exports of beverages and tobacco (SITC 1), crude materials, inedibles, except fuels (SITC 2), and machinery and transport equipment (SITC 7) from low-income countries significantly. However, one can also observe that the TBTs reduce the quality of food and live animals (SITC 1) and machinery and transport equipment (SITC 7) exported from low-income countries, and global SPS measures reduce the quality of manufactured goods classified chiefly by material (SITC 6) exported from low-income countries.
5. Summary and Concluding Remarks
Following the establishment of GATT and the WTO and the subsequent fall in tariffs, NTMs have increasingly been used as a trade policy tool. However, the various causes and motivations behind the imposition of NTMs make their implications hard to interpret. The complex and opaque nature of these trade policy instruments has been emphasized in the literature. Despite the trade-impeding consequences of NTMs on the quantity of traded products, the quality improvement of these products can point to the direction of legitimate motives behind them. The discriminatory behaviour and trade restrictiveness of these trade policy instruments have been studied extensively in the literature. However, a visible gap has remained regarding the analysis of the impact of these complex measures on the quality of traded products. This study aims to contribute to the literature by filling this gap.
Using the rich database of NTM notifications by WTO members, we have analysed the diverse impacts of two types of regulative and standard-like NTMs on the quality of traded products at the four-digit level of the SITC during the period 1995–2011. For this, we borrowed the quality index from a demand–supply theoretical framework proposed by Feenstra and Romalis (Reference Feenstra and Romalis2014) and used four different proxies of TBTs and SPS measures in the analysis. The results of our analysis for the whole sample of traded products indicate a positive impact of TBTs and SPS measures on the quality of a traded good. However, the results differ quantitatively for the various proxies. The positive impact of TBTs is strongest when we use a dummy variable on the stocks of TBTs that are imposed before a given year and are still in force in a year. Using a count variable on the stocks of TBTs imposed at the six-digit level of HS average within each three-digit SITC sector would render a model with the best fit. Using this model, we find that an additional TBT imposed on a product traded bilaterally can improve the quality of that product by 0.01%. The positive impact of an SPS measure is strongest when we include it as a dummy variable on the flows of SPS measures that are imposed each year. When we include SPS measures as a count measure on stocks of existing SPS measures up to date, then the positive impact on quality becomes statistically insignificant.
As NTMs may have a heterogeneous impact across products and sectors, we have also run regressions on the sample of traded products for each one-digit SITC sector using the most-fit model. The positive impact of TBTs on traded quality remains strong and statistically significant in many sectors, such as ‘food and live animals chiefly for food’, ‘mineral fuels, lubricants, and related materials’, ‘animal and vegetable oils, fats, and waxes’, ‘chemicals and related products, n.e.s.’, ‘manufactured goods classified chiefly by material’, and ‘machinery and transport equipment’. We also find a strong positive impact of SPS measures on the quality of traded products in many sectors, such as ‘beverages and tobacco’, ‘animal and vegetable oils, fats, and waxes’, ‘chemicals and related products, n.e.s.’, and ‘machinery and transport equipment’.
Overall, our results suggest that the imposition of TBTs or SPS measures is conducive to the quality of the imported products. This quality-enhancing effect has to be taken into account when discussing the effects of such NTMs on the quantity of traded products. This is also consistent with the model outlined in Feenstra and Romalis (Reference Feenstra and Romalis2014) when NTMs enter as specific trade costs.
Supplementary Materials
To view supplementary material for this article, please visit https://doi.org/10.1017/S1474745621000392.
Acknowledgements
Research for this paper was financed by the Anniversary Fund of the Oesterreichische Nationalbank (Project No. 18044). The support provided by Oesterreichische Nationalbank for this research is gratefully acknowledged.
Special thanks should go to Simona Jokubauskaite, Oliver Reiter, and David Zenz for statistical support in the preparation of the data and econometric exercise.
Technical Appendix
Equation (3) can be reformulated in quality-adjusted terms as follows:
$$\mathop {\max }\limits_{\,p_{ij}^{{\rm fob}, rk} , z_{ij}^{rk} } \left[{\displaystyle{{\,p_{ij}^{{\rm fob}, rk} } \over {{( {z_{ij}^{rk} } ) }^{\alpha_k}}}-\displaystyle{{w^r{( {z_{ij}^{rk} } ) }^{1/\theta }/\varphi_{ij}^r } \over {{( {z_{ij}^{rk} } ) }^{\alpha_k}}}} \right]\displaystyle{{( {1 + \tau_i^{rk} } ) Q_{ij}^{rk} } \over {( {1 + t_i^{rk} } ) }}$$
(as 
$Q_{ij}^{rk} = q_{ij}^{rk} ( {z_{ij}^{rk} } ) ^{\alpha _k}{\rm \;}$) and to tariff-exclusive c.i.f. prices
$$\mathop {\max }\limits_{\,p_{ij}^{{\rm fob}, rk} , z_{ij}^{rk} } \left[{( {1 + \tau_i^{rk} } ) \displaystyle{{\,p_{ij}^{{\rm fob}, rk} + T_i^{rk} } \over {{( {z_{ij}^{rk} } ) }^{\alpha_k}}}-( {1 + \tau_i^{rk} } ) \displaystyle{{\left({\displaystyle{{w^r{( {z_{ij}^{rk} } ) }^{{1 \over \theta }}} \over {\varphi_i^r }} + T_i^{rk} } \right)} \over {{( {z_{ij}^{rk} } ) }^{\alpha_k}}}} \right]\displaystyle{{Q_{ij}^{rk} } \over {( {1 + t_i^{rk} } ) }}$$which can be rewritten in quality-adjusted c.i.f. prices (net of tariffs)
$$\mathop {\max }\limits_{P_i^{{\rm cif}, rk} , z_i^{rk} } \left[{P_{ij}^{{\rm cif}, rk} -( {1 + \tau_i^{rk} } ) \displaystyle{{\left({\displaystyle{{w^r{( {z_{ij}^{rk} } ) }^{{1 \over \theta }}} \over {\varphi_{ij}^r }} + T_i^{rk} } \right)} \over {{( {z_{ij}^{rk} } ) }^{\alpha_k}}}} \right]\displaystyle{{Q_{ij}^{rk} } \over {( {1 + t_i^{rk} } ) }}$$
From equation (8a) and (8b), the c.i.f./f.o.b. margin can be derived: denote 
$\mu ^k: = \left({{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({{\sigma \over {\sigma -1}}} \right)$, then the c.i.f./f.o.b.-margin (c.i.f. price including tariffs) is given by
$$\displaystyle{{\overline {\,p_i^{{\rm cif}, r{\rm \;}k} } } \over {\overline {\,p_i^{{\rm fob}, r{\rm \;}k} } }} = \displaystyle{{( {1 + t_i^{rk} } ) ( {1 + \tau_i^{rk} } ) T_i^{rk} \mu } \over {T_i^{rk} [ {\mu -1} ] }} = {\rm \;}( {1 + t_i^{rk} } ) ( {1 + \tau_i^{rk} } ) \displaystyle{{\mu ^k} \over {\mu ^k-1}}$$Including the expression for c.i.f. and f.o.b. prices, one gets from equation (8):
$$\hat{P}_i^{{\rm cif}, rk} = {\rm \;}\left[{\displaystyle{{( {1 + t_i^{rk} } ) ( {1 + \tau_i^{rk} } ) T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)} \right]} \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right]} \right)}^{\alpha^\kappa \theta }}}} \right]{\rm \;\;}\left[{\displaystyle{{X_i^{rk} } \over {\sigma ( 1 + t_i^{rk} ) N_i^r }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta {\rm^{\prime}}F^{rk}} ) } \right]^{\alpha ^\kappa \theta }$$which can be rearranged to
$$\hat{P}_i^{{\rm cif}, rk} = {\rm \;}\left[{\displaystyle{{{( {1 + t_i^{rk} } ) }^{1-\alpha^\kappa \theta }( {1 + \tau_i^{rk} } ) \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)} \right]{( {T_i^{rk} } ) }^{1-\alpha^\kappa \theta }} \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right]} \right)}^{\alpha^\kappa \theta }}}} \right]{\rm \;\;}\left[{\displaystyle{{X_i^{rk} } \over {\sigma N_i^r }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta {\rm^{\prime}}F^{rk}} ) } \right]^{\alpha ^\kappa \theta }$$In the case of heterogenous firms, this equation becomes
$$\overline {P_i^{{\rm cif}, rk} } = {\rm \;}\left[{\displaystyle{{\overline {\,p_i^{{\rm cif}, rk} } } \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\overline {\,p_i^{{\rm fob}, r{\rm \;}k} } } \right)}^{\alpha^\kappa \theta }}}} \right]\left[{\displaystyle{{X_i^{rk} /\kappa_2^k ( {1 + t_i^{rk} } ) } \over {M_i^r {( {\varphi^r/w^r} ) }^\gamma }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta {\rm^{\prime}}F^{rk}} ) } \right]^{{{\alpha ^\kappa \theta } \over {1 + \gamma }}}( {\kappa_2^k } ) ^{{1 \over {1-\sigma }}}$$
with 
$\kappa _2^k = {\gamma \over {\gamma -\alpha ^k\theta ( {\sigma -1} ) }} > 1$.
Again inserting and re-arranging yields
$$\eqalign{\overline {P_i^{{\rm cif}, rk} } & = {\rm \;}\left[{\displaystyle{{( {1 + t_i^{rk} } ) ( {1 + \tau_i^{rk} } ) T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)} \right]} \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}T_i^{rk} \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right]} \right)} \right)}^{\alpha^\kappa \theta }}}} \right]\cr & \quad \times \left[{\displaystyle{{X_i^{rk} /\kappa_2^k ( {1 + t_i^{rk} } ) } \over {M_i^r {( {\varphi^r/w^r} ) }^\gamma }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta {\rm^{\prime}}F^{rk}} ) } \right]^{{{\alpha ^\kappa \theta } \over {1 + \gamma }}}( {\kappa_2^k } ) ^{{1 \over {1-\sigma }}}} $$which can be simplified to
$$\eqalign{P_i^{{\rm cif}, rk} & = \left[{\displaystyle{{{( {1 + t_i^{rk} } ) }^{1-{{\alpha^\kappa \theta } \over {1 + \gamma }}}( {1 + \tau_i^{rk} } ) \left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)} \right]{( {T_i^{rk} } ) }^{1-\alpha^\kappa \theta }} \over {{\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\left({\displaystyle{{\alpha^k\theta ( {\sigma -1} ) } \over {1 + \alpha^k\theta ( {\sigma -1} ) }}\left[{\left({\displaystyle{{\alpha^k\theta } \over {1-\alpha^k\theta }}} \right)\left({\displaystyle{\sigma \over {\sigma -1}}} \right)-1} \right]} \right)} \right)}^{\alpha^\kappa \theta }}}} \right]\cr & \quad \times \left[{\displaystyle{{X_i^{rk} /\kappa_2^k } \over {M_i^r {( {\varphi^r/w^r} ) }^\gamma }}{\left({\displaystyle{{Y^k} \over {\,p^k}}} \right)}^{-\beta_0}\exp ( {-\beta {\rm^{\prime}}F^{rk}} ) } \right]^{{{\alpha ^\kappa \theta } \over {1 + \gamma }}}( {\kappa_2^k } ) ^{{1 \over {1-\sigma }}}} $$
Table A1. Summary statistics of the variables used in the estimation of the whole sample
Table A2. Summary statistics of the NTM measures used in models 5 and 6 across sectors
Table A3. Estimation results on exporting country group-specific impact of NTMs on quality of SITC products traded bilaterally during the period 1995–2011 across SITC sectors, Model 6
