Exports and governance: the role of private voluntary agrifood standards

Exports and governance: the role of private voluntary agrifood standards Dela-Dem Doe Fiankor, Inmaculada Martínez-Zarzoso, Bernhard Brümmer a University of Goettingen, Department of Agricultural Economics and Rural Development, Germany b University of Goettingen, Department of Economics and Center for Statistics, Germany c University Jaume I, Institute of International Economics, Castellón, Spain d University of Goettingen, Centre for Biodiversity and Sustainable Land Use (CBL), Germany

governance and institutions, they apply a principal component analysis to the WGIs. They find that good institutions improve food exports and duration in each of the EU market segments. Estimating a gravity model, Bojnec et al. (2014) show that the quality of institutions in both exporting and importing countries enhance bilateral agrifood trade for the BRIC countries. de Mendonça et al. (2014) show that issues such as property rights, quality of rural employment and adoption of national and international norms in agricultural activity are essential to enlarge trade flows between countries.
Premised on this discussion we hypothesise that increasing bilateral governance distance has a negative effect on agrifood trade ceteris paribus, i.e., the farther away countries are from each other in terms of their institutional quality, the less trade we will observe.

Voluntary food standards as private governance institutions
In many instances, retailers in developed countries ("the North") import their agricultural and food products from developing countries ("the South"). But, institutions and the ability to enforce strict food safety regulations in the North are better than in the South (Levchenko, 2007). Consider the case of the EU/EFTA; according to the EU Food Law (Regulation EC No 178/2002), where any food which is unsafe is part of a batch, it shall be presumed that all the food in that batch is also unsafe. It is the responsibility of retailers in the EU to ensure that banned substances are not applied or present on their imports from third countries. Retailers stand the risk of damaging their reputation and losing out financially if the quality of their imports is compromised.
Ensuring due diligence increases the transaction costs for retailers, especially where they cannot trust domestic institutions in the producing countries to ensure high standards. Exportoriented producers and firms operating in institutionally weak countries face difficulties in this regard, as buyers tend to infer the quality of their products partly from the generally poor reputation of their home countries' institutions (Montiel et al., 2012). As Hudson and Jones (2003) point out, because perceptions of quality have become associated with the level of development in the country of origin, developing countries find it especially difficult to signal quality to buyers. They are disproportionately hampered by information asymmetries and negative reputation effects (Goedhuys and Sleuwaegen, 2013) which necessitates signalling quality to their international partners through other means. For example, Dimitrova et al. (2017) find that when the differences in country-pair quality of institutions increase, uncertainty about exchanges heightens, and importers tend to rely more on an exporter's reputation for its people as a reassurance that exporting firms will be honest in their dealings. In other words, the more bilaterally distant the formal institutional environments between countries, the more beneficial the use of informal arrangements (Abdi and Aulakh, 2012).
Our point of departure is the argument that voluntary certification by exporters to a standard that is accepted in the importing country improves exporting country reputations by reducing the bilateral governance distance between the two countries. This effect is moderated through the transaction cost reducing effect of the standard for retailers in the importing country. When the quality of institutions differs widely between two countries, we argue that standards can act as surrogate governance institutions. They level the playing field by placing geographically dispersed firms on a common ground in terms of managerial practices, business language and conflict-settling procedures (Hudson and Jones, 2003;Goedhuys and Sleuwaegen, 2016). This will reduce the bilateral institutional distance across countries engaged in bilateral trade. In agricultural trade, importers can in many cases only judge the final product. In the presence of increasing bilateral governance distance, information asymmetries are pronounced and signalling quality becomes even more important. With bounded rationality, importers will look for proxies to assess product quality. Exporters that can provide quality assurance, e.g., via certification, gain a competitive edge (Cao and Prakash, 2011).
As traceability requirements (e.g., article 18.2 of the EU Food Law) 5 get stricter, retailers are increasingly interested in the guarantee that not only the final products but also the production processes meet the required standards. The surge in the number of retailer-led standards, e.g., GlobalGAP, International Featured Standards, British Retail Consortium standards is, therefore, not surprising. The case of GlobalGAP standards is particularly interesting because it is fast becoming quasi-mandatory to assess high-value markets despite being legally voluntary. As a business-to-business standard, GlobalGAP certification resembles an attempt by retailers to enforce a system where individual farmers' skills are benchmarked against each other. This provides a mechanism for retailers to identify producers, regardless of country of origin, producing according to industry accepted standards, i.e., those who can signal quality through the possession of a certificate of conformity. This enhances the scopes of importers to gauge the quality performance of their suppliers and ensures the inclusion of distant suppliers (Ouma, 2010).
In the process, this reduces the transaction costs for retailers dealing with producers scattered across various countries, who may have different food safety standards and different abilities to enforce them. By outsourcing both the knowledge acquisition and the technical expertise required for design and ex-post monitoring of the standard, GlobalGAP allows both for a reduction in the costs of monitoring food safety standards at the farm level and ensuring that they comply with EU public regulations (Maze, 2017). However, it also induces extra costs for the producing party, which some have interpreted as the increasing power of retailers to pass 5 The regulation states that "food and feed business operators shall be able to identify any person from whom they have been supplied with a food, a feed, a food-producing animal, or any substance intended to be, or expected to be, incorporated into a food or feed". on food safety risks through their supply chains (Lockie et al., 2013). For producers, GlobalGAP has "major" and "minor" musts that should be met along each stage of the production chain before certification is granted. 6 But, the harmonisation of production processes across farms overrides to some extent the institutional quality differences between high-value importing countries and suppliers, especially from countries with weak domestic food safety regulations. Hence, producers who bear the costs and comply, nevertheless, may achieve a competitive advantage.
In summary, supply chain governance via GlobalGAP standards is an attempt to normalise spatially dispersed farming practices across countries (Ouma, 2010). Certification provides a shared frame of reference for both parties and increases importers' trust in products irrespective of the country of origin. By increasing the visibility of actions of actors on the supply-side (i.e., producers and suppliers) to actors on the demand-side (i.e., retailers and importers) of the value chain, standards enable the maintenance of trust in distant relationships (Lockie et al., 2015).
Based on these arguments, we hypothesise that by reducing the transaction costs for retailers, private voluntary food certifications decrease the bilateral governance distance between countries.

Empirical application
To test our hypotheses, we estimate a structural gravity model of international trade. The gravity model describes one of the most stable relationships in economics: "interactions between large economic clusters are stronger than between smaller ones, and nearby clusters attract each other more than far-off ones" (van Bergeijk and Brakman, 2010, p.1). It has become the workhorse model for trade policy analysis. Our modelling approach is similar to Tadesse and White (2010) and Dimitrova et al. (2017) who assess the pro-export effect of immigrants on cultural distance, and the relationship between bilateral country reputation and export volume, respectively. Following Anderson and van Wincoop (2003), our augmented gravity model assumes a constant elasticity of substitution (σ) and product differentiation by place of origin. In addition, prices differ among locations due to asymmetric bilateral trade costs. In its log-log reduced form, the structural gravity model is specified as: where X ijkt is exports of product k from exporting country i to importing country j in year t. E jt is nominal GDP, which proxies the import demand of j in t. Y ikt is the level of domestic production of k in i. Y kt is aggregate world production and P jkt and Π ikt are the inward and outward multilateral resistance terms respectively. ε ijkt is the error term, which we cluster by product and country-pair at the same time. τ ijkt are trade costs, which we define as the following multiplicative log-linear function: ln τ ijkt = γ 1 ln Distance ij + γ 2 ln GovDist ijt + γ 3 GlobalGAP ikt + γ 4 GovDist ijt × GlobalGAP ikt + =79γ Θ As we highlight in the conceptual discussion, institutional quality differences between countries affect trade costs. Simultaneously, compliance with retailer-led standards like GlobalGAP are costs of doing business -that may, or may not, enhance profitability through improved market access (Lockie et al., 2015) -especially for producers targeting high-value export markets. Thus, we argue that the effects of both institutional quality differences and GlobalGAP certification on trade is via the trade cost channel. We augment the trade cost component of our model with GovDist ijt which proxies institutional quality differences between country pairs and a dummy variable, GlobalGAP ikt , which is our measure of the certification status of the exporting country. GovDist ijt × GlobalGAP ikt is the interaction of the two variables.
Distance ij is the bilateral distance between country-pairs, RTA ijt is a dummy that denotes membership in a regional trade agreement, and Tariff ijkt is product-specific ad valorem tariff.
θ ij is a vector of time-invariant traditional gravity covariates including dummies for sharing a common language, colonial ties, and a common border.
For estimation purposes, we introduce the trade cost component, τ ijkt into equation (1) and specify a standard augmented gravity model in its log-linear form as: where α t , ψ i , ρ j, and φ k are year, exporter, importer, and product fixed effects, respectively.
Production ikt is the domestic production of product k in the exporting country and GDP jt is the Gross Domestic Product of the importing country. These variables measure the supply-side capacity of the exporting country and the demand-side capacity of the importing countries, respectively. All other variables remain as defined in equation (2). To deal with the potential endogeneity of institutions and certifications due to reverse causality, we use a one-year lag of both variables (see, e.g., Dimitrova et al., 2017;Álvarez et al., 2018).
The model as specified in equation (3) is at best atheoretical because it does not account fully for the theoretical multilateral resistance terms P jkt and Π ikt in equation (1) , 2007). What this means is that the country fixed effects in equation (3) must vary with product and time. To that effect, our theoretically specified ordinary least squares (OLS) model is: where ψ ikt and λ jkt are the exporter-product-time and importer-product-time fixed effects respectively. Apart from being consistent with the gravity theory, the inclusion of these terms account for the size terms (i.e., GDP jt and Production ikt ) and the certification measure (GlobalGAP ikt ). 7 They also account for unobservable variables that have the country-product- 7 We do not include the main effect for GlobalGAP ikt in equations (4) and (5) because they are accounted for by the exporter-product-time effects. time dimension (e.g., non-tariff measures, infrastructure, domestic institutions), thus mitigating any further omitted variable biases that may lead to endogeneity in our model specification.
Furthermore, the specification in equations (3) and (4)  where zero valued trade flows are ubiquitous. Since we estimate our gravity model at the very disaggregated HS6 digit level, the issue of zeroes is even more pronounced-i.e., 81% of our observed trade flows are zero. As an alternative to the OLS specification, we adopt the Poisson pseudo-maximum-likelihood (PPML) estimator à la Santos Tenreyro (2006, 2011) in equation (5). The estimator's log-linear objective function allows us to specify the gravity equation in its multiplicative form without log-transforming the dependent variable, and is consistent under heteroskedasticity.
Similar variable definitions hold as in equation (2). Our hypotheses are confirmed when the coefficient on the governance distance measure is negative (i.e., β 5 < 0), but we expect a positive coefficient on the interaction term (i.e., β 6 > 0).

Data
Growing interest in studying the quality of governance institutions has given rise to quantitative governance indicators from different sources. These include data from the  (Kaufmann et al., 2011).
Each of these indicators, measured in units ranging from -2.5 (worst) to 2.5 (best), represents a different dimension of governance in a country which can potentially affect trade. Since our interest is in how these measures vary across country-pairs, we transform the country-varying WGIs into country-pair varying variables using an index defined in equation (6). There is an added advantage to this approach; it respects the structural properties of the gravity model by allowing estimates with the proper set of country-time fixed effects. Recent work that has followed this approach include Martínez-Zarzoso and Márquez-Ramos (2018) where GovDist ijt is the bilateral governance distance between exporter i and importer j in year t, WGI jnt and WGI int are the annual values for the n th WGI indicator for i and j, respectively, and V nt is the variance of the n th WGI across all countries in the dataset. The indicator is minimised at zero for countries with similar institutional qualities and maximised for countries that are institutionally furthest apart. In our sample, the average ranges from 0.014 (i.e., the Netherlands -Canada) to 17.69 (i.e., Finland -Somalia). Using the case of Germany as an importing country, Figure ( To test the second hypothesis, we use GlobalGAP certifications as our preferred private voluntary standard. We premise this on the observation that GlobalGAP has become the most widely applied retailer-led quality assurance scheme for agrifood production since its inception in 1997. 9 As we show in Table 2, the number of producers seeking certification has increased over time. The choice of GlobalGAP also makes the EU and EFTA ideal export destinations because GlobalGAP is considered a minimum requirement to access their agrifood markets. In 2007, in an attempt to mark their global relevance, they effected a name change from EUREPGAP to GlobalGAP. Hence, while GlobalGAP still wields a growing global influence, we expect their effects to be stronger for exports targeting the EU and EFTA. The dataset was provided by the GlobalGAP Secretariat in Cologne, Germany. GlobalGAP certifies both crops, livestock, and aquaculture. We limit our study to the fruit and vegetable crops sector, specifically, apples, bananas and grapes, given their relative importance among certified products. Together with potatoes, these products are the most GlobalGAP certified open field crops by area (GLOBALGAP, 2012), representing more than 30% of all GlobalGAP certified fruits and vegetable production (Table 2). Moreover, to test the generality of our findings to the broader high-value agrifood sector, we use aggregate GlobalGAP certified fruits and vegetable production in a country as a robustness check. Unlike the productspecific data on apples, bananas, and grapes, this dataset does not allow us to identify the specific crops certified within a country. It aggregates data on all products that can be certified within the GlobalGAP sub-scope 'Fruits and Vegetables'.
Our dataset covers exports from 134 non-EU/EFTA countries to 31 EU/EFTA countries over the period 2010 to 2015. We omit re-exports from non-producing countries. To match the available product-specific GlobalGAP data, our set of exporters is limited to apple, banana and grape producing countries (see Table A1 of the online appendix). Over the study period, 25 exporting countries had at least one certified apple producer, 36 exporting countries had at 9 In international agri-food trade, private standards are, ubiquitous nevertheless, GlobalGAP standards are more widespread, e.g. Mohammed and Zheng (2017) show that for the 131 countries they study, the number of GlobalGAP certified sites is normally several times larger than that certified to other private standards (i.e. BRC, FSSC 22000, ISO 22000, PrimusGFS, SQF).
least one certified banana producer and 30 exporting countries had at least one certified grape producer. Hence, for each of these countries, the GlobalGAP ikt dummy takes the value of 1.
The remaining gravity model data come from different sources. The bilateral trade data on fresh and dried apples (HS 080810 and 081330), fresh and dried bananas (HS 080300) 10 and, fresh and dried grapes (HS 080610 and 080620) comes from the United Nations Comtrade database and is downloaded at the six-digit HS2007 level. 11 Data on distance, colonial ties, common language, and contiguity are from the Centre d'Etudes Prospectives et d'Informations Internationales. Data on effectively applied tariffs are from the International Trade Centre, and data on regional trade agreements are from De Sousa (2012). Detailed summary statistics on all included variables are presented in the online appendix (Table A2).

Main results
To allow for comparison across model specifications and to conclude whether our variables of interest can be estimated reliably regardless of the estimation procedure, we present and discuss the results of both the OLS and PPML models. 12 Table 3 reports the estimated coefficients in equations (4) and (5). In many cases, consistent with the literature the estimates of the PPML model are smaller than in the OLS specification (Santos Silva and Tenreyro, 2006). The signs and magnitudes of the traditional gravity control variables are all consistent with the gravity literature. Bilateral distance and tariffs decrease trade, but linguistic similarity, and countries that share a common border, past colonial ties or are members of a trade agreement are more likely to trade than otherwise. The coefficient estimates on the Tariff and RTA variables are not always statistically significant across our model specifications. This is not surprising because the sample of importing countries contains EU and EFTA countries and therefore there is little variation in the RTA and tariff variables.
10 Banana trade flows recorded in the six-digit HS2007 classification includes plantains. But trade volumes are low compared to bananas, and should not alter our results. 11 We limit the bilateral trade data to six years to match the available product-specific GlobalGAP data 12 To deal with the high-dimensional fixed effects in our model specifications, we use the user-written commands reghdfe (Correia, 2016) and poi2hdfe (Guimaraes and Portugal, 2010) in Stata.
In support of our first hypothesis, the coefficient estimate on our governance dissimilarity measure, GovDist ijt−1 , is negative and statistically significant at any conventional level in both model specifications. Thus, with increasing bilateral governance distance, bilateral trade decreases. This implies that for retailers in the EU and EFTA, when deciding where to source their agrifood products, they prefer countries with institutional qualities similar to those existing in the EU and EFTA.
Next, we test the effect on the interaction of governance distance and GlobalGAP standards.
We enter the interaction term GovDist ijt−1 × GlobalGAP ikt−1 , and the constitutive terms of the interaction into the models in columns (2) and (4). The GlobalGAP ikt−1 term is omitted from the tables as they are accounted for by the exporter-product-time fixed effects. In support of our hypothesis, the coefficient estimate on the interaction term is positive and statistically significant. Hence, the more distant the governance gap between country pairs, the more effective the use of certification.
Based on these findings, we assess the differential effect of bilateral governance distance on trade flows depending on the GlobalGAP certification status of the exporting country. From equations (4) and (5), the effect for certified countries includes the direct effect of the governance gap proxy and the coefficient on the interaction term (i.e., < + E * 1M3NB ). Thus, empirically based on our a priori expectation, a negative governance gap effect becomes less negative if the interaction term is positive. Specifically, for noncertified countries, the effects on trade are the direct GovDist ijt−1 effects (i.e., -0.599 in column 2 and -0.450 in column 4). For certified countries, the trade-inhibiting effect of governance distance is about half the magnitude for non-certified countries (i.e., -0.296 in column 2 and -0.185 in column 4).
Our results imply that even though bilateral governance distance has a trade impeding effect on trade flows, the negative effects are smaller for certified compared to non-certified countries. This suggests that product certification, which signals product quality, is important in enhancing exports even for country pairs with big differences in institutional quality. This is because where public food safety regulations are missing or, when available, institutions to enforce them are weak, the GlobalGAP standard provides the retailer with an instrument to manage their risks (Lockie et al., 2013). However, because the coefficient on the interaction term is smaller in magnitude than the direct effect of GovDist ijt−1 (i.e., |β 6 | < |β 5 |), the GlobalGAP certification effect is not sufficiently large to completely eliminate the negative effects of governance distance.
To put the findings in perspective, we use the results of the PPML specification. For the average effect in column (3) We also explore the heterogeneity of the effect across products and the development level of the certified exporting country. We define two levels of development based on the World Bank's income classification: "developing" (all countries listed as middle and low income) and "developed" (all countries listed as high income). The results of the analysis presented in columns (1) and (2) of Table 4 show that our findings are driven by developing country exports.
The almost 60%, 18% and, 28%, respectively since 2008 (Lernoud et al., 2015). Given the relevant for exporting firms trying to signal quality via certification than issues concerning the quality of public services (part of Government Effectiveness), or some forms of corruption; all of which are closely related to "behind the border" trade barriers affecting trade flows.
importance of banana in the EU, bananas certified to other standards may be just as important as GlobalGAP.

Robustness checks
In this section, we conduct a series of robustness checks to confirm the reliability of our findings. For comparative purposes, we extend the analysis to include all producing countries as exporters and all importing destinations (Table 5). This sample includes bilateral trade flows between 163 producing countries and 157 importing countries (see Table A2 of the online appendix). All estimated coefficients remain consistent with the gravity theory. In the OLS case, the coefficients on colonial ties and common language become statistically significant compared to the estimates in Table 3. Moreover, membership of a trade agreement increases trade by about 70% 16 , while a 10% increase in bilateral tariffs decreases trade by 42% in column (1). Focusing on our variables of interest, the trade inhibiting effect of bilateral governance distance and the pro-export effect of the interaction term remains robust. The magnitudes are nevertheless smaller than in our main specification and the coefficient on the interaction term is statistically insignificant in the PPML specification in column (4). This latter finding is due to the significant heterogeneity in the sample of importers and the fact that for some developing country importers certification may not be that important as for importers in developed countries. 17 As further checks of the generality of our findings, we extend our analysis to all GlobalGAP certified fruits and vegetables. Hence this part of the analysis re-estimates our baseline specification, but considers an aggregate of products listed under HS07 (i.e., edible vegetables, and certain roots and tubers), HS08 (i.e., edible fruits and nuts) and the spices listed in HS09. 18 To ensure theoretical consistency, we control for the multilateral resistance terms using 16 Dummy variables are interpreted as [exp(β) − 1] × 100%. 17 To test this empirically, we estimate equations (4) and (5) splitting the sample between two importer groups: EU/EFTA and non-EU/EFTA. The interaction term remains statistically insignificant and economically small in magnitude for non-EU/EFTA importers. The results are available upon request from the authors. 18 In this dataset, it is not possible to identify the number of product k specific producers in country i. It aggregates data on all products that can be certified within the GlobalGAP sub-scope 'Fruits and Vegetables'.
importer-time and exporter-time fixed effects. The findings reported in Table 6 confirm our main findings. Hence, our product-level findings in Table 3 can be extended to the broader agrifood sector.
As another exercise, we re-estimate the product-level specifications, but use instead of a GlobalGAP certification dummy, the number of certified producers in each exporting country.
The results presented in the online appendix (Table A5) (Table A6) show that our hypotheses are confirmed regardless of the measure of institution we use.

Conclusion
Much of the existing literature has shown that governance and institutions are important drivers of trade and economic growth. Similarities in governance and institutional quality measures across countries enhance bilateral trade flows. Hence, retailers in countries with good institutions will choose to source their products from countries with similar or better domestic institutions. Aside from the reputational damage associated with potential food scares, institutional dissimilarities also impose significant costs for trade. Hence, the more dissimilar country-pairs the less trade will be observed. Much less attention has, however, been paid to how exporting countries in low-quality institutional regimes can overcome these differences.
This paper evaluates first, the effect of bilateral differences in governance and related institutions across countries on agrifood trade. Retailers, especially in high-value markets such as the EU and EFTA, are increasingly becoming concerned about traceability, quality of production processes and final products. Thus, second, we argue that private food standards and certifications act as surrogate institutions that help to overcome these differences at the country level. We are not aware of any existing studies that test this hypothesis empirically in the agricultural trade literature.
Empirically, our gravity model estimates confirm the trade reducing effect of bilateral governance distance on trade flows. But in addition, we also find that the trade impeding effects vary depending on whether the exporting country is certified to GlobalGAP standards or not.
For certified exporting countries, the trade impeding effects are much lower compared to their non-certified counterparts, especially for exports to the EU and EFTA. Hence, we show that certification exerts a pro-export effect that partially offsets the trade-inhibiting effects of bilateral governance distance at the country level. Our findings are robust to the productspecific analysis of apples, bananas, and grapes but also the aggregate fruits and vegetable sector, and to different measures of institutional quality.
These findings have important policy implications. For producers and firms targeting highvalue markets but are located in countries with low quality of existing domestic public institutions, getting certified to a standard that is accepted in the importing country can help overcome the negative reputation effects associated with their geographical locations.
Undoubtedly, certification in itself is not enough to overcome the total bilateral governance distance at the country level. Nevertheless, it is a viable alternative to reduce trade costs and enhance trade. We leave for further research the evaluation of the effect of private standards on exports from developing to developed countries using firm-level data.  Notes: Robust country-pair-product clustered standard errors in parentheses. ***, **, * denote statistical significance at 1%, 5% and 10% respectively. Importer-product-time and exporter-product-time fixed effects included in all regressions. Intercepts included but not reported. Notes: Robust country-pair-product clustered standard errors in parentheses. ***, **, * denote statistical significance at 1%, 5% and 10% respectively. Importer-product-time and exporter-product-time fixed effects included in all regressions. All standard gravity controls have their expected signs but are omitted from the table for brevity. Intercepts included but not reported. Notes: Robust country-pair-product clustered standard errors in parentheses. ***, **, * denote statistical significance at 1%, 5% and 10% respectively. Importer-product-time and exporter-product-time fixed effects included in all regressions. Intercepts included but not reported. Notes: The dependent variable is aggregate exports of fruits and vegetables from country i to country j in year t. Robust country-pair-product clustered standard errors in parentheses. ***, **, * denote statistical significance at 1%, 5% and 10% respectively. Intercepts included but not reported. Importer-time and exporter-time fixed effects included in all regressions. Fig. 1. Average bilateral governance distance: 2010 -2015 (using Germany as the importer)