Does GlobalGAP Certification PromoteAgricultural Exports?

Dela-Dem Doe Fiankora*, Insa Flachsbartha, Amjad Masooda, andBernhard Brummera

aUniversity of Goettingen, RTG 1666: GlobalFood, Department of Agricultural Economics andRural Development, Germany

July 27, 2017

19th Annual European Study Group Conference, Florence

Abstract

This paper responds to calls for further empirical research on the nexus betweenprivate food standards and agricultural trade flows. Specifically, the analysis focuseson GlobalGAP certification which is one of the most visible private standards inglobal agricultural trade and almost de facto mandatory for producers to gain andmaintain access to EU markets. Thus, we investigate whether and to what extentGlobalGAP certification affects agri-food exports to the EU. Empirically, we estimatea structural gravity model using a novel dataset on the number of certified producersand area of land cultivated to grapes, apples, and bananas from 2010 to 2015. Whileour results generally confirm the trade enhancing effect of GlobalGAP certificationon exports to the EU, we shall see that the effects vary across products. Our findingsemphasize to a large extent the “standards-as-catalyst” hypothesis for apples andgrapes, but for bananas, certification has no significant effect on trade flows.

Keywords: Agricultural trade, GlobalGAP, Private food standards, Gravity model

JEL Classification: F14, L15, Q17, Q18

*Corresponding author: dfianko@agr.uni-goettingen.de

1 Introduction

Food standards and certifications1 have gained prominence in the governance of globalfood supply chains, and continue to play significant roles in agricultural production,processing, marketing, and trade. Unlike public standards, that are (in many cases) legallymandatory, private standards are voluntary. Yet, the proliferation of private standardsmeans they are now seen by many producers as de facto mandatory requirements (Hensonand Humphrey, 2010) to gain and maintain access to high-value markets.

Underlying this spread of private food certification is the growing dominance ofretailers who are often using grades and standards to influence the international marketfor agri-food products. With decreasing use of tariffs and quantitative restrictions ininternational trade, large retail chains source their produce from various destinations.Thus, the growing importance of private food standards is due in part to efforts byretailers to control entire production processes—from farm to fork—and facilitate supplychain management within an increasingly globalised and competitive international foodmarket (Clarke, 2010). This ensures limiting the associated risks of working with variousspatially dispersed actors and activities in the supply chain (Dolan and Humphrey, 2000),ensuring due diligence and protecting their reputations (Subervie and Vagneron, 2013).It allows them to differentiate their products hence decreasing consumers uncertainty andincreasing their demand (Vandemoortele and Deconinck, 2014). For producers—especiallythose from developing countries who often face negative reputation effects with respectto product quality—targeting export markets, it is difficult yet important to send foodsafety and quality signals to consumers and retailers. By adopting certification, theydemonstrate to potential buyers a commitment to quality, environmental sustainabilityand decent labour conditions (Goedhuys and Sleuwaegen, 2016).

However, the existing literature has not provided a definite direction of the effects offood standards on agricultural trade. Increased trade costs required to meet standardsmay reduce trade flows (Shepherd and Wilson, 2013), but the associated improvementin information asymmetry and the reduced consumer search cost may increase consumerconfidence and boost trade (Henson and Jaffee, 2008). Standards may also have noeffects on trade (Schuster and Maertens, 2015) or have different short and long runeffects (Maertens and Swinnen, 2009). This ambiguity creates room for more empiricalevidence to come to more general conclusions (Honda et al., 2015). The existing empiricalliterature reveals a predominant focus on public standards (e.g. Anders and Caswell,2009; Ferro et al., 2015), possibly because data on private standards are often confidentialand inaccessible. It is nevertheless, important to deal with the trade effects of voluntarystandards in major markets because in many cases they are more stringent than publicstandards (Vandemoortele and Deconinck, 2014). Our paper represents another effortin response to this call for further empirical analysis of private standards on agri-food trade.

When faced with standards, exporting countries either exit to markets with laxregulations, voice out their concerns at the WTO or upgrade and comply (Lee et al.,2012). In this paper, our objective is to provide an ex post examination of the effect ofprivate food standards on trade flows. Thus, as a first distinguishing feature, this paperfocuses on countries who chose to comply and test whether and to what extent compliance

1In this paper standards and certifications are used interchangeably.

1

enhances their exports to high value markets. In high-value agri-food markets, privatestandards are ubiquitous, hence, any such study is made complex by the myriads ofprivate food safety standards available. Recently, Latouche and Chevassus-Lozza (2015)showed that French firms who adopt the British Retail Consortium (BRC) standardsbut not the International Featured Standards (IFS) benefit from better access to certainEU markets. Ehrich and Mangelsdorf (2016) also find that on average IFS increasesexports of manufactured agricultural commodities to the EU. As a second distinguishingfeature, our study departs from these and other related studies in that they focus onpost-farm gate standards, whiles we seek to study the effects at the pre-farm gate level.To this end, we focus on perhaps the foremost private agri-food industry standardworldwide and arguably one of the most important in Europe (Henson et al., 2011)— the Global Partnership for Good Agricultural Practices (GLOBALG.A.P.)2. It is abusiness-to-business (B2B) retail industry farm gate level process standard that indicatesat every stage of production—from soil management, plant protection to non-processedend product—how products must be produced and handled. By the year 2010, morethan 40 European retail chains required their suppliers to be GlobalGAP certified (Colenet al., 2012). As a result, many producers in both developed and developing countries areembracing GlobalGAP certification as an entry ticket to high-value EU markets.

How GlobalGAP certification affects trade flows is far from a proven fact. Thoughit provides a good case to study how private standards affect international trade flows,few such studies exist. Where they do, they are mostly country-specific and with mixedfindings. In Kenya, GlobalGAP certified farmers obtain significantly higher net incomesfrom export vegetable production than their non-certified counterparts (Asfaw et al.,2010b). Surveying fresh produce exporting firms in 10 Sub-Saharan African (SSA)countries, Henson et al. (2011) find that GlobalGAP certified firms have appreciablyhigher export revenues. GlobalGAP certified lychee farmers in Madagascar receivedhigher premiums and sold higher quantities (Subervie and Vagneron, 2013) and, in Ghana,GlobalGAP investments to gain access to export markets yield on average positive returns(Kleemann et al., 2014). On the other hand, Schuster and Maertens (2015) cannotconfirm that GlobalGAP certification has any impact on the export performance ofasparagus producing firms in Peru. While these studies are informative, they are limitedin explaining whether the effects are due to sectoral and country characteristics (Beghinet al., 2015). As a third distinguishing feature, this paper adopts a cross-country perspec-tive. Using a gravity model allows us to study the effect of GlobalGAP certification onagri-food exports to the EU from all producing countries. By considering a wide range ofcountries, our approach has the advantage of providing more generalized analytical results.

We accommodate heterogeneity across different agri-food products by studying thetrade effect at the disaggregated HS06 product level. Hence, ours is the first to assess theeffect of GlobalGAP certification on trade flows using multiple but specific agriculturalproducts. Closely related to our work is that of Masood and Brummer (2014) who useda three year panel dataset to show that increasing intensity of GlobalGAP standardsaffects positively EU imports of banana from certified countries. We employ a uniquedataset that covers worldwide GlobalGAP certification of apples, bananas, and grapesover the period 2010 to 2015. Together with potatoes, these products constitute the mostGlobalGAP certified open field crops (GLOBALGAP, 2015). Empirically, we specify

2The brand name is GLOBALG.A.P., but we use GlobalGAP through out this paper for readability.

2

a structural gravity equation, account for the inherent heteroskedastic nature of tradedata and the incidence of zero trade flows using the Poisson pseudo-maximum-likelihoodestimator and address a largely ignored issue in the literature; the potential endogeneityof GlobalGAP certification and trade flows. While our results generally confirm the tradeenhancing effect of GlobalGAP certification on exports to the EU, we shall see that theseeffects vary across products and income distributions.

The rest of the paper proceeds as follows. The next section provides backgroundinformation on GlobalGAP, discusses its potential trade impacts, and introduces the data.Section three explains the empirical framework and econometric specification. The resultsare presented and discussed in Section four. Section five concludes.

2 Background and Data

2.1 Spread of GlobalGAP Certification

GlobalGAP certification is currently one of the most visible private standards in globalagricultural trade and de facto mandatory for producers to gain and maintain access tomarkets especially in the EU. It is aimed at assuring retailers of product safety and qualityrather than providing quality signals directly to consumers. These product attributes may,however, not be directly observable and acquiring such information involves transactioncosts. To reduce these costs, members of the Euro-Retailer Produce Working Groupreacting to consumer concerns—e.g. product safety—and technical regulations—e.g. duediligence—harmonised their own often very different agri-food standards (van der Meulen,2011) to form GlobalGAP in 1997. Certification is based on food safety, traceability,environmental sustainability and worker occupational health and currently includes agri-cultural management practices such as Integrated Crop and Pest Management, QualityManagement System, Hazard Analysis and Critical Control Points (GLOBALGAP, 2015).

It is the most widely used certification program in the agri-food export subsector inmany SSA countries (Colen et al., 2012). Many other countries have taken steps to developtheir own domestic standards and benchmark them fully to the GlobalGAP standards(e.g. ChileGAP, KenyaGAP, MexicoGAP, ChinaGAP). Hence, over time we observe anexponential growth in both the number of certified producers and the area cultivated toall fruits and vegetables (Figure 1). From 2008 to 2015, the number of certified producersincreased by more than 300% in Africa, 70% in Asia and the Pacific, 200% in America and,57% in Europe. In terms of certified land area, we observe a 72% increase in Africa, 160%in America, 82% in Europe and, 61% in Asia and the Pacific. Therefore, its increasingrole as a major private standard linking developed and developing country farmers tointernational retailers cannot be overstated.

3

Figure 1: Development of GlobalGAP certified producers and land area by region

60

70

80

90

10

0

Num

ber

of cert

ifie

d p

roducers

(’0

00)

51

01

52

0

Num

ber

of cert

ifie

d p

roducers

(’0

00)

2008 2009 2010 2011 2012 2013 2014 2015

Year

Africa America Asia and Pacific Europe

(a) Producers

40

60

80

10

01

20

Cert

ifie

d land a

rea (

’0000 H

a)

10

15

20

25

30

35

Cert

ifie

d land a

rea (

’0000 H

a)

2008 2009 2010 2011 2012 2013 2014 2015

Year

Africa Asia and Pacific America Europe

(b) Land area

Source: GlobalGAP data, own graph. (Note: Dotted lines refer to the secondary axis)

2.2 GlobalGAP standards - effects on trade

As noted earlier, one reason why it is important to study how private food standardsaffect trade flows is because certifications involve compliance costs. Standards can hindertrade if their associated costs are too high (Maertens and Swinnen, 2009), i.e., if theysignificantly raise setup and production costs. For example, to get GlobalGAP certified,producers need to pay a registration fee (charged per product and per hectare, andsubject to annual renewal) and the associated costs of implementing the standard. Theseassociated costs will vary depending on the nature of the country (or specifically the farm)i.e. the level of economic development and the quality of existing domestic food safetyregulations. Some may need to implement new policies, processes, and installations tocomply with the standard (GLOBALGAP, 2015). In developing countries where existingdomestic standards are usually low, the initial cost of upgrading to developed countryrequirements may be higher (Martinez and Poole, 2004). On the other hand, for countrieswith stringent domestic standards, producers already bear higher costs to comply, butthese also allow them to assess markets with tighter requirements (Drogue and DeMaria,2012). Initial GlobalGAP certification and compliance costs range from US$ 6000 to US$8000 (Jaffee, 2003) and continuous compliance and renewal of certificates is estimated tocost US$ 200 per month (Kariuki et al., 2012).

When standards are not harmonised, compliance costs can be even higher as this willrequire asset-specific investments by producers to adapt to the idiosyncratic productionprotocols of different buyers. The result is, producers most likely facing wide divergencebetween their domestic and international food safety standards (Maertens and Swinnen,2009). However, by harmonising different agri-food standard requirements, GlobalGAPallows producers to export to all markets in the EU without having to adopt country orretailer-specific production processes. A related argument is that high compliance costsmarginalise small businesses and poor farmers in developing countries from high-valuemarkets leaving the rents in the supply chain for large companies to extract (Colenet al., 2012). This is because the additional fixed costs necessary to adjust productionto international standards deter unproductive firms—which are mostly in developing

4

countries—from exporting to markets with stricter standards (Ferro et al., 2015). Tofacilitate this, GlobalGAP introduced group certifications. There have also been reportedcases of technical and financial support from donors and trade facilitators (Subervie andVagneron, 2013).

Once these costs are overcome, there are potential benefits from compliance i.e.the catalyst effect. Standards can reduce transaction costs by providing a commonlanguage within supply chains. This presents, on the one hand, a potential link betweenincreasingly demanding consumer requirements and on the other, the participation ofdistant suppliers while raising consumer confidence in product safety (Henson and Jaffee,2008; Ferro et al., 2015). They lower coordination costs, reduce information assymetriesalong supply chains and reduce the cost of solving moral hazard problems for buyers facingheterogeneous suppliers (Carlo et al., 2014). Standards help in reducing market failures;for consumers, they allow comparison of products on a common basis and for producers,production of goods subject to recognized standards help achieve economies of scale andreduce overall costs. However, there is the possibility that even after bearing the costs ofcertification, standards have neutral effects on trade (see e.g. Schuster and Maertens, 2015).

As a B2B standard, the GlobalGAP system provides a cost-effective way for retailers toidentify farmers producing according to industry-accepted standards, i.e., those who havevoluntary GlobalGAP certification. For producers, aside access to international markets,following GlobalGAP protocols ensure improved input control, record keeping, traceabil-ity system, and improved farm management which results in improved profitability andincreased exportable yields (Graffham, 2006). Clearly, certification directly affects costs,profits and/or market access for producers and retailers. In practice, however, GlobalGAPcertification redistributes some of the food safety costs—e.g. soil and water testing, em-ployee training and annual audits—away from retailers to producers. These mandatoryinitial investments, recurrent expenditure, and better-trained employees are nevertheless,likely to result in increased productivity and/or enhanced product quality (Colen et al.,2012), and reduced incidence of acute illnesses (Asfaw et al., 2010a). Consequently, oncecertification is achieved we a priori expect positive trade effects to dominate for Glob-alGAP standards. The intuition behind this expectation is clear, as a B2B standard,GlobalGAP certified products benefit from retailer networks, which in turn reduce searchand transaction costs and, enhance exports.3 Nevertheless, it is reasonable to expectvariations across products, farm size, and productivity.

2.3 GlobalGAP Data and Descriptives

This section describes our main dataset. We employ a unique Integrated Farm AssuranceStandard (i.e. the GlobalGAP certificate) dataset supplied by the GlobalGAP Secretariatin Cologne, Germany. Currently, the standard setting body offers 16 standards for threescopes—crops, livestock, and aquaculture—in over 100 countries. We limit our studyto the scope of crops, specifically fruits and vegetables, where producers seek the most

3For instance, the GlobalGAP Chain of Custody certification ensures that for market agents handlingcertified products, there is proper segregation of certified and non-certified produce in packing, processing,handling, and operation units (GLOBALGAP, 2015). In their study on GlobalGAP certified lycheeproducers in Madagascar, Subervie and Vagneron (2013) also find that local treatment plants providedseparate sorting lines for certified and non-certified products. This guarantees certified, but not non-certified farmers, the opportunity to sell larger quantities.

5

certification. There are currently about 153,461 certified fruits and vegetable producers(out of a total of 160,452 certified producers) in 120 countries covering a total land areaof 2,928,775 Ha (GLOBALGAP, 2015).

Given the available data, our initial GlobalGAP dataset includes 42 certified appleproducing countries, 38 certified banana producing countries, and 42 certified grapeproducing countries for the years 2010 to 2015. Since GlobalGAP is a pre-farm gateprocess standard, we extend the dataset to include as exporters, all producing countries(i.e. 75 apple producing, 76 banana producing and 66 grape producing countries).GlobalGAP standards originated from and are widely required by EU retailers, hence,we include as importers members of the EU-27 (excluding Croatia). Table A1 in theappendix provides a detailed list of included countries. The first novelty of our datasetis the multiplicity of certified products; it allows us to study the effect of GlobalGAPcertification on exports of apples (HS 080810 and HS 081330), bananas (HS 080300)and grapes (HS 080610 and HS 080620) to the EU. It also allows us to assess how thetrade effects vary across income distributions; whiles developing countries dominate theEU market for banana, the reverse is mostly the case for apples and grapes (see Table A3).

A bit more background about the different GlobalGAP certification schemes may helpmotivate our choice of interest variables. There are four GlobalGAP certification options;of interest to the present study are Option one (individual certification) and Option two(group certification). The remaining options are the single producer and group certifica-tion benchmarked schemes. For successive years and for each country, our dataset containsextensive data on (1) the number of product specific certificates issued and (2) the num-ber of certified producers per product. Group certifications help to achieve economies ofscale, but they obscure the individual number of certified producers in a country (whichwe consider a better measure of the spread of certification). As we observe in Table 1, theobscuring effect is highest for banana producing countries. Here, we observe almost fourtimes as many certified producers as number of certificates issued. It appears that spe-cific investments associated with certification lead smallholder farmers, who predominatedeveloping countries, to pursue group certifications. For example, in Kenya Mausch et al.(2009) provide evidence that smallholders are mostly group certified, whereas mediumand large scale farms opt for individual certification. In Morocco, irrespective of farm size,farmers that are not affiliated with any cooperative system are unlikely to implement anyform of food safety standard. And even for those cooperative members, they only complywhen a collective approach is undertaken (Aloui and Kenny, 2005). Therefore, to cap-ture the effect of certification, we use the total number of certified producers per product.Our dataset also contains data on the area of land—measured in hectares—cultivated tospecific GlobalGAP certified product per year.4 We employ these two indicators, as ameasure of the spread of certification in a country.

4Note, however, that the number of hectares for countries or products with less than 10 producers arenot provided by GlobalGAP.

6

Table 1: Number of certified producers and certificates per product (2010 - 2015)

Mean Total

Apple Banana Grape Apple Banana Grape

Producers 281·41 48.74 112.67 3,413,178 605,795 1,194,488Certificates 84·47 11.15 35.09 1,024,623 138,605 372,034Producers per certificate 3·33 4.37 3.21 3.33 4.37 3.21

Source: GlobalGAP data and own calculations

As an initial exploratory analysis of our dataset, Figure 2 plots graphically in panel(a) the relationship between GlobalGAP certification and development measured as percapita GDP and, in panel (b) the relationship between certification and exports. Theobserved correlation is positive in both cases. Richer countries have on average morecertified producers of all products. However, when this is studied at the product level, theslope of the fitted lines is steepest for apple producing countries and less so for bananaproducers. This is not surprising because, in terms of national incomes, apple and grapeproduction is dominated by countries with higher incomes relative to bananas.5 Certifiedcountries also enjoy on average higher exports to the EU.

Figure 2: GDP per capita, exports and spread of GlobalGAP certification

ARG

ARM

AUS

AUTBEL

BGR

BIH

BRA

CAN

CHE

CHL

CHN

CIV

CMR

COL

CRI

CYPCZE

DEU

DNK

DOMECU

EGY

ESP

ETH

FRA

GBR

GHA

GRC

HND

HRV

HUN

IDN

IND

ISR

ITA

JOR

JPNKEN

KOR

LBN

LCA

MAR

MDA

MEX

MKD

MOZ

NAM

NIC

NLD

NZL

PAK

PAN

PER

PHL

POL

PRI

PRT

ROU

SAU

SEN

SUR

SVK

SVN

THA

TJK

TUN

TUR

UKR

URY

USA

VCT

ZAF

46

81

01

21

4

To

tal n

um

be

r o

f ce

rtifie

d p

rod

uce

rs (

log

s)

7 8 9 10 11log Average per capita GDP (USD)

lnprod_tot Fitted values

(a) Certification and GDP per capita

ARG

ARM

AUS

AUTBEL

BGR

BIH

BRA

CAN

CHE

CHL

CHN

CIV

CMR

COL

CRI

CYPCZE

DEU

DNK

DOM ECU

EGY

ESP

ETH

FRA

GBR

GHA

GRC

HND

HRV

HUN

IDN

IND

ISR

ITA

JOR

JPNKEN

KOR

LBN

LCA

MAR

MDA

MEX

MKD

MOZ

NAM

NIC

NLD

NZL

PAK

PAN

PER

PHL

POL

PRT

ROU

SAU

SEN

SUR

SVK

SVN

THA

TJK

TUN

TUR

UKR

URY

USA

VCT

ZAF

46

81

01

21

4

To

tal n

um

be

r o

f ce

rtifie

d p

rod

uce

rs (

log

)

2 4 6 8 10 12 14 16

Total exports to the EU in USD (logs)

ln Certified Producers Fitted values

(b) Certification and exports

Source: GlobalGAP, UNComtrade and, World Bank data, own graph

Given the nature of our dataset, we proceed to test the hypothesis that increasingspread of GlobalGAP certification within a country enhances exports to EU markets.That is, we test the extent to which the catalyst argument dominates for each productacross different countries.

5The mean per capita GDP for banana producing countries in our sample is USD 12,298.53 comparedto USD 22,652.53 for apple and USD 20,008.21 for grapes.

7

3 Empirical Application

3.1 The Gravity Model

The gravity model has over the years developed into the preferred workhorse for tradepolicy analysis. Aside from being intuitive, it has solid theoretical foundations, representsa realistic general equilibrium environment, and has good predictive power (Yotov et al.,2016). It is favoured among empirical researchers estimating the impact of standards ontrade flows (e.g. Anders and Caswell, 2009; Ferro et al., 2015). Earlier applications ofthe model were naıve and flawed as they ignored what we now know as the multilateralresistance (MR) terms (Anderson and van Wincoop, 2003). Intuitively, MR implies thattrade flows between two countries do not only depend on bilateral trade costs betweenthem but trade costs prevailing with all their other trade partners. It is important toaccount for these MR terms else we commit the “Gold Medal Mistake” (Baldwin andTaglioni, 2007) and are unable to predict accurately how GlobalGAP standards affecttrade flows in the gravity framework. Following Anderson and van Wincoop (2003) wespecify a modified structural gravity model as:

lnXijt = lnEjt + lnYit − lnYt + (1 − σ) ln tijt − (1 − σ) lnPjt − (1 − σ) ln Πit + εijt (1)

Where Xijt is trade flows from exporting country i to importing country j in year tin current US dollars. Ejt is nominal GDP, which proxies the import demand of j in t.Yit is the annual level of domestic production6 of product l in i in current US dollars. Ytis aggregate world production, σ is the elasticity of substitution and Pjt and Πit are theimporting and exporting country MR terms respectively. εijt is the error term. tijt aretrade costs, which we define as:

tijt = Dβ1ij τ

β2ijtGAPβ3it exp

K∑k=4

βkΩij (2)

As conventional in gravity models, Dij is the bilateral distance between the capitalcities of i and j, and Ωij is a vector of traditional gravity covariates including dummies forcommon language, colonial ties, contiguity, and the presence of a regional trade agreement(RTA). τijt is product-specific ad valorem tariffs. We augment the trade cost componentof the model with a variable, GAPit, which is our measure of GlobalGAP standard.

3.2 Estimation Issues

Estimating equation (1) is not without econometric and modelling issues. First, the MRterms are theoretical constructs and not directly observable. Following much of the recentliterature, we use country fixed effects as proxies (Baldwin and Taglioni, 2007). In paneldata settings, these proxies must be time-varying. However, given that our GlobalGAPmeasure (GAPit) in equation (2) is time varying only in the exporter dimension, it iscollinear with the Πit terms in equation (1). We need to establish a proper identificationstrategy that allows us to combine our variable of interest with the MR terms. This isbecause including the structural time-varying fixed effects will absorb all time-varying

6We argue that using GDP as a proxy for the mass of exporting countries is less suitable in thecontext of agricultural trade, hence, we use sector-specific annual production. This is a better measure ofthe supply-side capacity of the exporting countries.

8

country characteristics (including our variable of interest) that may influence bilateraltrade flows.7 We use instead time-invariant importer (γj), exporter (λj) and year (ψt)fixed effects (see e.g. Czubala et al., 2009; Drogue and DeMaria, 2012; Disdier et al., 2015).We consider this adequate because, over the study period, GlobalGAP requirements areunlikely to have changed much. In fact, the GlobalGAP Integrated Farm Assurancestandards are only reviewed every four years (GLOBALGAP, 2015). While acknowledgingthat this might be insufficient to capture any time-varying multilateral resistance terms, webelieve the potential bias in our case is limited. Following Disdier et al. (2015) we includethe size terms Ejt and Yit to control for time variation in country pair demand and supply.

Secondly, because we study sectoral trade flows, zeroes are ubiquitous in our bilateraltrade dataset. Log-transforming the dependent variable as in equation (1) makes itimpossible to properly account for any informative zero trade flows. Common prac-tices in the literature employed to deal with zeroes in trade data are truncation andcensoring. These methods are arbitrary and without strong theoretical or empiricaljustification and can distort results significantly (Burger et al., 2009). Consequently,we employ more appropriate estimation techniques to deal with the issue of zeroes.We eliminate a number of excess zeroes by limiting our samples to only producingcountries. It is intuitive to assume that countries that are not producing probablydue to climatic or biological reasons are either not exporting or only re-exporting.8 Allremaining zeroes will be informative to our study and dropping them will bias our findings.

Thirdly, if the inherent heteroskedastic nature of trade data is not dealt with, estimatescan be biased and inconsistent. Using the Poisson pseudo-maximum-likelihood (PPML)estimator, we simultaneously overcome jointly the issues of zero trade flows and het-eroskedasticity (Santos Silva and Tenreyro, 2006, 2011). It allows us to specify the gravitymodel in its multiplicative form and use the dependent variable in levels. Even in microsettings such as agriculture, which are likely to include a lot of disaggregated trade data,the use of the PPML estimator is justified (Prehn et al., 2012). The estimator has alsobeen employed by Shepherd and Wilson (2013) to study the effects of product standardson trade flows. The PPML estimator is, however, not without criticism. Particularly,Will and Pham (2008) find that the estimator is biased when a substantial fraction of theobservations is zero.9 The estimator also disregards the existence of another data generat-ing process that produces excessive zeros in the trade matrix caused by self-selection intono trade (Crivelli and Groschl, 2015). As a further check, we will employ the Heckmanestimation method that accounts for zero trade flows in a two-step procedure. First, aProbit (i.e. selection) equation is estimated on whether country pairs in our sample engage

7To circumvent this problem Heid et al. (2015) and Yotov et al. (2016) recommend introducing intra-national trade flows and interacting the unilateral trade policy measure with a dummy which equals onefor international trade flows and zero otherwise. This generates a new variable that varies over time inboth exporter and importer directions and allows us to identify the trade effect in the presence of Πit andPjt. We are unable to employ this approach because their assumption that the trade policy measure doesnot affect intra-national trade flows will not apply in our case.

8Re-exporters are not interesting in our study because GlobalGAP certification is a farm level processstandard. We identify producing countries using the FAO production dataset. Because of limited data onproduction values, we use instead data on production quantities. Our results are however, not responsiveto changes in the data sample that comes with moving from values to quantities.

9Zeroes make up about 69.5%, 80.8%, 65.6% of the apple, banana and grape trade matrices, respec-tively. However, running the PPML model with only positive trade values, we show that the presence ofzeroes do not bias our results in this case (Table A6).

9

in bilateral trade in a given year or not. Second, the expected values of the trade flows,conditional on trade are estimated by OLS (i.e. the outcome equation). This propertyof the Heckman model allows distinguishing between the effect on both extensive (i.e.the probability of trading) and intensive (i.e. the value of export conditional on trading)margins of trade. The Heckman model is, however, not robust to heteroskedasticity andrequires an exclusion variable that is assumed to impact the decision to trade but not theactual value traded.

3.3 Model Specification

One additional econometric issue that arises is the potential endogeneity of GlobalGAPcertification to agricultural trade flows. Put differently, while certification will affect trade,existing trade is also likely to enhance the decision to seek certification. To reduce thispotential reverse causality bias, we consider a one-year lag of GlobalGAP certification.This is because while past and current certification status are highly correlated, we do notexpect past certifications to influence current trade flows (see e.g. Shepherd and Wilson,2013; Ferro et al., 2015). Taking into account possible heterogeneity across different agri-cultural products, we run separate estimations for each product, and specify our baselinemodel as follows10:

Xijt = exp

(γj + λi + ψt + β0 + β1 ln GAPit−1 + β2 ln Yit + β3 ln Ejt + β4 ln Dij

+ β5 ln (1 + τ)ijt + β6Languageij + β7Colonyij + β8Contiguityij + β9RTAijt

)εijt

(3)

Similar variable definitions hold as in equations (1) and (2). Although our baselinemodel specification includes country and time fixed effects, there could still remain en-dogeneity concerns due to omitted variable biases. A natural extension is to follow anInstrumental Variable approach. Addressing this in a gravity framework is not easy dueto a lack of relevant and valid instruments. Among the few studies that embark on thatpath, however, many observe that while endogeneity may be present, it does not seemto qualitatively affect their findings (see e.g. Moenius, 2004; Vigani et al., 2012). Lack-ing plausible instruments, a promising approach is to include country-pair (dyadic) fixedeffects (Head and Mayer, 2014). In auxiliary estimations we include country-pair fixedeffects (φij). These control for all time-invariant trade cost proxies in our benchmarkspecification (i.e. distance, common language, colonial ties and contiguity), reduce pos-sible biases resulting from the omission of any such variables and control for most of thelinkages between our GlobalGAP trade policy variables and the error term (Yotov et al.,2016). The resulting estimation equation is given as:

Xijt = exp

(φij + ψt + β0 + β1 ln GAPit−1 + β2 ln Yit + β3 ln Ejt + β4 ln (1 + τ)ijt + β5RTAijt

)εijt

(4)

To ensure we do not lose observations in situations where there are zero certificationsper country-product-year pairs, we add a constant (i.e. the value of one) to our variablesof interest before taking logarithms. In both estimation equations, our variable of interestis GAPit−1. Giving the nature of our data, which measures the spread of certification

10For robustness, we will also run and present results with a different gravity structure where we poolour dataset over all three products.

10

within countries (a situation that is only possible once the cost of certification is borne),we expect a positive and significant β1, implying that certification increases trade flows.

The remaining gravity model data are derived from different sources. GDP and agricul-tural production data are from the World Bank World Development Indicators databaseand, FAOSTAT of the Food and Agricultural Organisation respectively. Bilateral tradeflow data are assessed from the United Nations Commodity Trade (UNComtrade) databasevia the World Integrated Trading System (WITS). Country-specific data on distance, colo-nial ties, common language, and contiguity are derived from the Centre d’Etudes Prospec-tives et d’Informations Internationales (CEPII), data on bilateral tariffs are from theInternational Trade Centre, and data on Regional trade agreements come from De Sousa(2012). Detailed summary statistics on all included variables are presented in Table A4.

4 Results

4.1 GlobalGAP and trade flows (Benchmark Results)

Benchmark regression results are shown in Table 2. To allow for comparison, we alsorun estimations with a different gravity structure, where observations are pooled over allthree products (columns 1 and 2). Columns 3 - 8 contain our preferred product specificestimates. All models are estimated using the PPML estimator with importer, exporterand time fixed effects (which are omitted for brevity) except for columns (1) and (2) wherewe include country-product fixed effects. The odd numbered columns present results usingthe number of certified producers, while the even numbered columns use certified land area.

All standard gravity variables have their expected signs and almost all of them aresignificant. In most cases, the size terms (i.e. importer GDP and production) havepositive effects on bilateral trade. The level of domestic production, which measures thesupply side capacity of exporting countries has a positive and statistically significanteffect on banana exports to the EU. The effects are statistically insignificant for apples,and grapes. Importing country GDP has a positive and statistically significant effecton imports of apple and grapes. The effects are positive but insignificant for banana.A expected, distance hinders exports of all products; with a one percent increase indistance decreasing trade in apples, bananas, and grapes by 0.9%, 3.1%, and 0.8%respectively. Speaking a common language increases apple and banana export flows.The effects are statistically insignificant for grape. Sharing a common border andpast colonial relationships increase exports of all products. The effects in both casesare not statistically significant for apple export. Trade agreements enhance exports ofapples and grapes but not banana. Higher tariffs, on the other hand, hinder grape exports.

Focusing on our variable of interest, the coefficient estimate of β1, we observe a positiveeffect of certification on exports in all model estimations, except for banana. Specifically, aone percent increase in the number of certified producers increases apple and grape exportsby 0.17%, and 0.16% on average, respectively. For certified land area, a percentage in-crease enhances apple exports by 0.10% and grape exports by 0.04%. For banana exports,the trade effects are indistinguishable from zero. This implies that whiles GlobalGAP cer-tification has not enhanced banana exports to the EU, increasing GlobalGAP certificationin exporting countries does not reduce export volumes. In all cases, however, the coeffi-

11

cients of our estimated elasticities are larger for number of certified producers comparedto certified land area, i.e., a percentage increase in the number of certified farmers has alarger positive effect on exports compared to a percentage increase in certified land area.This finding may also be driven by the distribution of land across countries. Interestingly,the positive trade effects of GlobalGAP standards found using the pooled data thoughconsistent with our a priori expectation also reveal how nuanced the effects can be whenstudied at aggregate levels. We show here that whiles different research questions mightnecessitate the use of different levels of data disaggregation, in seeking to make accuratepolicy recommendations, the level of product aggregation matters.

Table 2: The effect of GlobalGAP certification on agri-food exports to the EU (PPMLmodel)

Pooled data Apple Banana Grape

(1) (2) (3) (4) (5) (6) (7) (8)

ln GAPProducerst−1 0.027*** 0.173** -0.035 0.156**

(0.010) (0.088) (0.065) (0.079)ln GAPHectares

t−1 0.004 0.101** -0.012 0.043**(0.006) (0.046) (0.010) (0.019)

ln Production 0.001 0.002 0.019 0.019 0.312** 0.310** -0.012 -0.016(0.012) (0.012) (0.039) (0.038) (0.148) (0.147) (0.012) (0.011)

ln GDP 0.410* 0.416* 0.992*** 0.985*** 0.122 0.118 0.596* 0.619*(0.219) (0.219) (0.272) (0.256) (0.443) (0.446) (0.359) (0.350)

ln Distance -0.964*** -0.964***-0.895*** -0.896*** -2.994***-2.994***-0.768*** -0.768***(0.178) (0.178) (0.177) (0.177) (0.887) (0.887) (0.274) (0.274)

Language 0.179 0.179 0.554* 0.553* 0.700* 0.700* -0.280 -0.280(0.195) (0.195) (0.324) (0.324) (0.391) (0.391) (0.247) (0.247)

Contiguity 0.669*** 0.669*** 0.283 0.282 2.641*** 2.641*** 1.338*** 1.338***(0.253) (0.253) (0.291) (0.291) (0.852) (0.852) (0.351) (0.351)

Colony 0.784*** 0.784*** 0.358 0.358 1.269*** 1.269*** 0.812*** 0.813***(0.182) (0.182) (0.299) (0.299) (0.427) (0.427) (0.208) (0.208)

RTA 0.118* 0.124* 0.763*** 0.832*** -0.023 -0.034 0.444*** 0.516***(0.069) (0.071) (0.197) (0.182) (0.079) (0.078) (0.074) (0.064)

ln Tariff -0.058*** -0.059***0.041 -0.009 0.035 0.098 -0.027 -0.056***(0.015) (0.015) (0.108) (0.098) (0.258) (0.257) (0.020) (0.015)

Constant -17.521***-8.043***-19.366***-13.680***-0.096 -0.031 -17.707***-17.821***(1.560) (1.830) (3.061) (3.324) (8.984) (9.007) (3.797) (3.629)

Observations 30,245 30,245 9,860 9,860 9,192 9,192 8,835 8,835R-squared 0.808 0.808 0.735 0.735 0.867 0.867 0.796 0.797

Notes: Robust country-pair clustered standard errors in parentheses. ***, **, * denote significance at 1%, 5% and 10% respec-tively. Importer, exporter, and year fixed effects included in all regressions. Columns (1) and (2) include in addition country-product fixed effects. Measure of GlobalGAP standard: columns (1), (3), (5) and (7) use number of certified producers andcolumns (2), (4), (6) and (8) use certified land area.

4.2 Introducing country pair fixed effects

Next, we introduce dyadic fixed effects as described in equation (4) to confirm the robust-ness of our benchmark model estimates to possible endogeneity. The results presented inTable 3 confirm that our main findings remain largely unchanged. The estimated coeffi-cients are qualitatively and quantitatively very similar to the estimates in the benchmark

12

models; an indication of the robustness of our findings. GDPs of the importing countrieshave positive effects on apple and grape imports. The coefficient estimates of grapes gainin significance compared to the baseline estimates. Level of domestic production has nostatistically significant effect on exports of all products when we control for country-pairfixed effects. RTAs increase trade but bilateral tariffs hinder grape trade. Our GlobalGAPvariables of interest retain their positive signs and magnitude for apple and grapes, but forbanana exports, the effects remain statistically insignificant. These findings lend supportto the standards-as-catalyst hypothesis on the one hand, but on the other hand show thatstandards can have neutral trade effects.

Table 3: Robustness check: including country pair fixed effects

Pooled data Apple Banana Grape

(1) (2) (3) (4) (5) (6) (7) (8)

ln GAPProducerst−1 0.027*** 0.165* -0.036 0.164**

(0.010) (0.088) (0.066) (0.079)ln GAPHectares

t−1 0.003 0.095** -0.013 0.044**(0.006) (0.045) (0.010) (0.019)

ln Production 0.001 0.001 0.019 0.020 0.267 0.264 -0.012 -0.016(0.012) (0.012) (0.039) (0.038) (0.166) (0.164) (0.011) (0.011)

ln GDP 0.514** 0.517** 0.963*** 0.954*** 0.240 0.243 0.687** 0.672**(0.204) (0.204) (0.274) (0.257) (0.412) (0.416) (0.346) (0.337)

RTA 0.120* 0.125* 1.637*** 1.635*** -0.020 -0.031 0.450*** 0.524***(0.070) (0.072) (0.155) (0.151) (0.078) (0.078) (0.075) (0.065)

ln Tariff -0.058*** -0.058*** 0.033 -0.012 0.046 0.107 -0.025 -0.055***(0.015) (0.015) (0.108) (0.098) (0.254) (0.252) (0.020) (0.015)

Constant -20.861***-23.369***-32.846***-36.711***-28.392***-17.683***-21.261***-26.822***(2.320) (2.298) (3.040) (3.256) (4.991) (2.710) (3.401) (3.533)

Observations 13,044 13,044 4,625 4,625 2,915 2,915 4,145 4,145R2 0.978 0.978 0.978 0.979 0.976 0.976 0.980 0.981

Notes: Robust ountry-pair clustered standard errors in parentheses. ***, **, * denote significance at 1%, 5% and 10% respectively.Year and country-pair fixed effects included in all regressions. Columns (1) and (2) include in addition country-product fixed ef-fects. Measure of GlobalGAP standard: columns (1), (3), (5) and (7) use number of certified producers and columns (2), (4), (6)and (8) use certified land area.

4.3 Results from the Heckman two-step estimation

Table 4 checks the sensitivity of our results to the choice of estimation technique usingthe two stage Heckman estimator. All estimations include importer, exporter and yearfixed effects. The first two columns which present results estimated using pooled dataalso include product fixed effects. The Heckman two-step estimations require an exclusionvariable that is assumed to have an impact on the decision to trade but not on the actualvalue of trade. We use common religion as the exclusion restriction. This variable entersthe selection equation but not the outcome equation.

13

Table 4: Robustness check: outcome equation of the 2-stage Heckman model

Pooled data Apple Banana Grape

(1) (2) (3) (4) (5) (6) (7) (8)

ln GAPProducerst−1 0.649*** 0.119 -0.028 0.625***

(0.024) (0.124) (0.122) (0.166)ln GAPHectares

t−1 0.322*** 0.029 0.001 0.024(0.013) (0.048) (0.037) (0.047)

ln Production 0.186*** 0.210*** 0.040 0.038 0.011 0.012 -0.035 -0.048(0.018) (0.018) (0.038) (0.038) (0.062) (0.062) (0.049) (0.050)

ln GDP 0.508 0.528 -0.380 -0.395 0.480 0.480 0.788 0.750(0.568) (0.574) (0.873) (0.873) (1.306) (1.305) (0.959) (0.961)

ln Distance -1.543*** -1.519*** -1.650*** -1.648*** -2.057*** -2.061*** -1.537*** -1.534***(0.089) (0.090) (0.103) (0.103) (0.387) (0.387) (0.143) (0.143)

Language 0.229* 0.219 0.170 0.173 0.461 0.457 0.387* 0.385*(0.133) (0.135) (0.180) (0.180) (0.286) (0.286) (0.204) (0.204)

Contiguity 1.507*** 1.521*** 1.199*** 1.198*** 5.074*** 5.067*** 1.932*** 1.938***(0.134) (0.135) (0.150) (0.150) (0.705) (0.705) (0.218) (0.218)

Colony 0.559*** 0.551*** 0.801*** 0.800*** 0.662** 0.665** 0.491** 0.493**(0.138) (0.139) (0.186) (0.186) (0.296) (0.296) (0.216) (0.216)

RTA 0.430* 0.981*** 0.188 0.174 0.128 0.128 0.118 0.457(0.247) (0.252) (0.721) (0.721) (0.324) (0.345) (0.423) (0.413)

ln Tariff -0.350*** -0.417*** 0.230 0.206 0.387 0.386 0.263** 0.200*(0.061) (0.062) (0.215) (0.213) (0.280) (0.280) (0.111) (0.111)

lambda 0.625*** 0.625*** -0.129 -0.129 1.605*** 1.601*** 1.653*** 1.660***(0.129) (0.131) (0.142) (0.142) (0.282) (0.282) (0.198) (0.199)

Constant -15.070*** -15.373*** 1.598 2.193 -11.330 -11.301 -7.077 -3.444(4.317) (4.363) (5.697) (5.643) (9.088) (9.085) (6.343) (6.315)

Observations 31,649 31,649 10,130 10,130 10,299 10,299 8,835 8,835

Notes: ***, **, * denote significance at 1%, 5% and 10% respectively. Dependent variable is the logarithm of the value of trade.Measure of GlobalGAP standard: columns (1), (3), (5) and (7) use number of certified producers and, columns (2), (4), (6) and(8) use certified land area. Importer, exporter and year fixed effects included in all regressions. Columns (1) and (2) also includeproduct fixed effects.

In this paper, we are mostly interested in the results of the outcome equation. Theseare presented in Table 4. In order to save space, the results of the first stage probitselection equation are presented in the appendix (Table A5). The coefficient estimate oflambda (shown at the bottom of Table 4) which is a measure of sample selection biasis statistically significant in all model estimations but apple. This implies that omittingzero trade observations would bias our estimates. The standard gravity variables are inmost cases well behaved and consistent with existing gravity estimates except for tariffswhich show a unexpected signs in columns (7) and (8). Common religion has a significantpositive effect in the Probit equation (Table A5) for all three products and justifies inpart its use as an exclusion variable in the outcome equation. Our variables of interestare positively signed in all cases but for banana in column (5). This confirms the mainfindings in our benchmark model. GlobalGAP certifications have positive effects on tradeflows of apples and grapes. The coefficient estimates are much larger than in our baselinemodel and only statistically significant for grapes in column (7). Here again, we observethat considering the trade effects where the products are pooled together presents a farmore nuanced picture than when studied at the product level.

14

4.4 Differences in trade effects by development status

As a final exercise, Table 5 investigates how the effects of certification vary across devel-opment level of the exporting countries. Because food standards and certifications arelikely to be a particular constraint on small and medium scaled farmers, this makes it avital issue for developing countries. To ascertain if there are any such differences acrossincome classifications, we introduce and interact a binary variable (DC)—which takes thevalue of 1 if a country is developing and 0 otherwise—with our GlobalGAP variables. Ourdefinition of developing includes all countries not listed as high income in the World Bankincome classifications.

Table 5: The effect of GlobalGAP certification on agri-food exports to the EU (By incomestatus)

Apple Banana Grapes

(1) (2) (3) (4) (5) (6)

ln GAPProducerst−1 0.205** -0.049 -0.049

(0.095) (0.239) (0.039)ln GAPHectares

t−1 0.164* -0.047 0.015(0.089) (0.240) (0.031)

ln Production 0.038 0.016 0.267 0.267 -0.015 -0.018*(0.039) (0.041) (0.165) (0.165) (0.011) (0.011)

ln GDP 0.993*** 0.953*** 0.260 0.258 0.698** 0.675**(0.277) (0.276) (0.433) (0.438) (0.333) (0.337)

RTA 1.621*** 1.637*** -0.023 -0.032 0.368*** 0.524***(0.157) (0.155) (0.079) (0.078) (0.082) (0.065)

ln Tariff -0.009 0.034 0.044 0.108 -0.019 -0.048***(0.109) (0.109) (0.257) (0.258) (0.019) (0.018)

DC -6.830*** -0.707 -11.544*** -11.765*** -4.605*** -10.414***(1.443) (2.127) (3.304) (3.295) (1.140) (0.825)

ln GAPProducerst−1 * DCit -0.425*** -0.050 0.289***

(0.147) (0.068) (0.101)ln GAPHectares

t−1 * DCit 0.067 -0.013 0.047**(0.122) (0.010) (0.022)

Constant -20.327*** -19.436*** -16.611*** -16.610*** -13.700*** -14.238***(2.565) (2.523) (1.852) (1.869) (2.735) (2.772)

Observations 4,625 4,625 2,898 2,898 4,145 4,145R2 0.979 0.978 0.976 0.976 0.981 0.981

Notes: Robust country-pair clustered standard errors in parentheses. ***, **, * denote significance at 1%, 5% and 10%respectively. Year and dyadic fixed effects included in all regressions. Measure of GlobalGAP standard: columns (1), (3)and (5) use number of certified producers and columns (2), (4) and (6) use certified land area. Developed countries aredefined here as high income countries, whiles middle and low income countries constitute developing countries.

In most cases, the developing country dummy is negative and significant; implyingthat relative to their developed counterparts, developing countries have on average lowerexports to the EU. However, once certified, the effects of GlobalGAP standards on devel-oping countries’ exports to the EU is rather mixed. The effects are captured by summingup the direct effect of the GlobalGAP proxy and the interaction term. Thus, empirically,a trade enhancing effect of certification becomes even more positive when the interactionterm is also positive, and vice versa. The interaction terms are negative and significantfor apple, negative but insignificant for banana, and positive and significant for grapes.

15

Specifically, conditional on being a developing country, exports are reduced by 0.22% ifnumber of certified apple producers increase by 1%. This surprising finding is possiblydriven by the distribution of production; as shown in Table A3, apple production is dom-inated mainly by developed countries. On the other hand certification increases exportsby 0.21% in developed countries. For an additional certified land area, the effects are pos-itive but marginally significant. For banana producing developing countries, an additionalcertified farmer or certified land area has no significant effect on banana exports. Lastly,conditional on being a developing country, the effects of certification, measured both bythe number of producers or land area, are positive and significant for grape exports tothe EU. Given the nature of our dataset, this significant positive effect is not surprising.It is driven in most part by upper middle income countries (who per our definition aredeveloping) who dominate the production of grapes (i.e. almost 38% of the producingcountries in our sample). If we consider this vis-a-vis, the insignificant effect we observefor banana—which is a preserve for mainly developing countries—it becomes clearer whenwe note that banana production, unlike grape production, is dominated by lower middleincome and low income countries who together make up 39% of production. Keeping inmind the dynamics in the concentration of production, our results show that relative todeveloped countries, increasing GlobalGAP certification has enhanced mainly developingcountries’ exports of grapes to the EU.

5 Conclusions

This paper assesses how private GlobalGAP standards affect agricultural exports. Glob-alGAP is at present one of the most visible certification schemes in global agriculturaltrade and almost de facto mandatory for producers to gain and maintain market accessto the EU. Using a novel dataset on the annual number of certified producers andcultivated land area, this paper investigates the effects of GlobalGAP certification onexports of apples, bananas, and grapes to the EU-27. Many related studies are countryor product specific, which brings into question their generality. We take a multi-countryand product-specific approach and use trade data on all producing countries from 2010to 2015. We control for potential endogeneity of GlobalGAP certification and trade flowsusing lags and dyadic fixed effects, and for zero trade flows using the PPML estimator.

Our results show that GlobalGAP certification enhances exports of apples andgrapes—emphasizing to a large extent the “standards-as-catalyst” argument—but has nosignificant effect on banana exports. The results are robust to two different certificationmeasures, controls for reverse causality, omitted variable biases. Specifically, our estimatessuggest that a percentage increase in the number of certified producers increases appleand grape exports by 0.17%, and 0.16% on average, respectively. For certified land area,a percentage increase enhances apple exports by 0.10% and grape exports by 0.04% onaverage. For banana exports, the effects of certification on exports are insignificant.These findings are consistent with several strands of the existing literature. First, thepositive trade effects for apples and grapes coincide with findings that the returns onGlobalGAP investments are considerable in terms of export growth (Henson et al.,2011) and impacts positively on quantities sold on international markets (Subervie andVagneron, 2013). Second, our insignificant results for banana coincide with the findingsof Schuster and Maertens (2015) that GlobalGAP standards have no effects on exportperformance, but also contradict findings that GlobalGAP certification promotes EU

16

banana imports (Masood and Brummer, 2014). The latter study, however, does notaccount for zero-valued trade flows.

This economically insubstantial point estimate effect of certification for bananadeserves further discussion. First, banana production is dominated by developingcountries which are often affected by inefficient domestic capacities, inter alia, missinginfrastructure, low or non-existent domestic food safety standards and, inadequatetechnical capacities to manage food quality and safety. So, even if certification allowsproduction according to industry-accepted standards, these domestic inefficiencies mayhinder the expected export growth. Secondly, while certification grants market access toEU retail networks, there are other behind-the-border issues, e.g. delayed shipping times,that could lead to import rejections. Thus, even with increasing banana certification,country level inefficiencies mean GlobalGAP standards may not constitute a sufficientcondition per se for increased banana exports to the EU. Thirdly, competing voluntarycertification schemes for banana are gaining in prominence. For instance, though Global-GAP still has the largest banana certified area globally, Fairtrade, Organic and RainforestAlliance/SAN certified banana area increased by almost 60%, 18% and, 28%, respectivelysince 2008, while GlobalGAP certified banana area declined by 6% since 2012 (Lernoudet al., 2015). Given the low level of banana production in the EU, and its reputationas the number one banana consumption market globally, exports of banana productscertified to these other standards may be just as important as GlobalGAP. One otherreason that possibly justifies the insignificant effect for banana may arise from a particularcharacteristic of the sector; i.e. the historic presence of big companies (Dole, Chiquita,Fyffes, Del Monte, Compagnie fruitiere, etc.) that have always strongly structured thesupply to the world market (UNCTAD, 2016). Because they sometimes possess theirown production units in producer countries, it is possible these large vertically integratedfirms were already ensuring high standards. Therefore, the introduction of GlobalGAPcertification may not have made a huge difference in their export volumes. Lastly, it isalso possible that increasing certification marginalizes non-certified production or bet-ter still non-certified producers target markets with relatively lax food safety requirements.

Our findings suggest that in most cases GlobalGAP standards do not constitute sig-nificant barriers to existing trade flows; providing further evidence that food standardsthat are harmonised to a common international level enhance trade. However, becausethe effects differ across products, product specific policy designs are recommended totake account of this differences. There are also obvious heterogeneities across incomeclassifications, with insignificant trade effects dominating for banana which is producedpredominantly by very low-income countries. Thus, in its current form, certification doesnot seem worthwhile for small-scale farmers. This calls for further technical and finan-cial support from donors and trade facilitators to improve supply-side abilities of LeastDeveloped Countries (LDCs).

17

References

Aloui, O. and Kenny, L. (2005). The Cost of Compliance with SPS Standards for MoroccanExports: a Case Study. Agriculture and Rural Development Discussion Paper 47843,The World Bank.

Anders, S. M. and Caswell, J. A. (2009). Standards as barriers versus standards as cata-lysts: assessing the impact of HACCP implementation on US seafood imports. AmericanJournal of Agricultural Economics, 91(2):310–321.

Anderson, J. E. and van Wincoop, E. (2003). Gravity with gravitas: a solution to theborder puzzle. American Economic Review, 93(1):170 – 192.

Asfaw, S., Mithofer, D., and Waibel, H. (2010a). Agrifood supply chain, private-sectorstandards, and farmers’ health: evidence from Kenya. Agricultural Economics, 41(3-4):251–263.

Asfaw, S., Mithofer, D., and Waibel, H. (2010b). What impact are EU supermarketstandards having on developing countries’ export of high-value horticultural products?Evidence from Kenya. Journal of International Food and Agribusiness Marketing, 22(3-4):252–276.

Baldwin, R. and Taglioni, D. (2007). Trade effects of the euro: a comparison of estimators.Journal of Economic Integration, 22(4):780–818.

Beghin, J. C., Maertens, M., and Swinnen, J. (2015). Nontariff measures and standardsin trade and global value chains. Annual Review of Resource Economics, 7(1):425–450.

Burger, M., van Oort, F., and Linders, G.-J. (2009). On the specification of the gravitymodel of trade: zeros, excess zeros and zero-inflated estimation. Spatial EconomicAnalysis, 4(2):167–190.

Carlo, R., Maria Angela, P., and Antonella, D. F. (2014). Using private food safetystandards to manage complexity: a moral hazard perspective. Agricultural EconomicsReview, 15(2):113–127.

Clarke, R. (2010). Private food safety standards: their role in food safety regulation andtheir impact. Paper prepared for presentation and discussion at the 33rd session ofthe codex alimentarius commission, Food And Agriculture Organization Of The UnitedNations.

Colen, L., Maertens, M., and Swinnen, J. (2012). Private standards, trade and poverty:GlobalGAP and horticultural employment in Senegal. The World Economy, 35(8):1073–1088.

Crivelli, P. and Groschl, J. (2015). The impact of sanitary and phytosanitary measureson market entry and trade flows. The World Economy.

Czubala, W., Shepherd, B., and Wilson, J. S. (2009). Help or hindrance? The impact ofharmonised standards on African exports. Journal of African Economies, 18(5):711–744.

De Sousa, J. (2012). The currency union effect on trade is decreasing over time. EconomicsLetters, 117(3):917–920.

18

Disdier, A.-C., Fontagne, L., and Cadot, O. (2015). North-South standards harmonizationand international trade. The World Bank Economic Review, 29(2):327–352.

Dolan, C. and Humphrey, J. (2000). Governance and trade in fresh vegetables: the impactof UK supermarkets on the African horticulture industry. Journal of DevelopmentStudies, 37(2):147–176.

Drogue, S. and DeMaria, F. (2012). Pesticide residues and trade, the apple of discord?Food Policy, 37(6):641–649.

Ehrich, M. and Mangelsdorf, A. (2016). The role of private standards for manufacturedfood exports from developing countries. GlobalFood Discussion Papers 85, RTG 1666GlobalFood. Georg-August-University of Goettingen.

Ferro, E., Otsuki, T., and Wilson, J. S. (2015). The effect of product standards onagricultural exports. Food Policy, 50:68–79.

GLOBALGAP (2015). GLOBALGAP Annual Report 2015. Technical report.

Goedhuys, M. and Sleuwaegen, L. (2016). International standards certification, institu-tional voids and exports from developing country firms. International Business Review,25(6):1344–1355.

Graffham, A. (2006). European standards affect African growers. Pesticide news, 71.

Head, K. and Mayer, T. (2014). Gravity Equations: Workhorse,Toolkit, and Cookbook,volume 4 of Handbook of International Economics, chapter 3, pages 131–195. ElsevierB.V.

Heid, B., Larch, M., and Yotov, Y. V. (2015). A simple method to estimate the effects ofnon-discriminatory trade policy within structural gravity models. Unpublished paper.Available at http://www.etsg.org/ETSG2015/Papers/439.pdf.

Henson, S. and Humphrey, J. (2010). Understanding the complexities of private stan-dards in global agri-food chains as they impact developing countries. The Journal ofDevelopment Studies, 46(9):1628–1646.

Henson, S. and Jaffee, S. (2008). Understanding developing country strategic responses tothe enhancement of food safety standards. The World Economy, 31(4):548–568.

Henson, S., Masakure, O., and Cranfield, J. (2011). Do fresh produce exporters in sub-Saharan Africa benefit from GlobalGAP certification? World Development, 39(3):375–386.

Hobbs, J. E. (2010). Public and private standards for food safety and quality: internationaltrade implications. The Estey Centre Journal of International Law and Trade Policy,11(1):136–152.

Honda, K., Otsuki, T., and Wilson, J. S. (2015). Food safety standards and internationaltrade: the impact on developing countries? export performance. In Food Safety, MarketOrganization, Trade and Development, pages 151–166. Springer.

Jaffee, S. (2003). From challenge to opportunity: Transforming Kenya’s fresh vegetabletrade in the context of emerging food safety and other standards in Europe. World Bank.

19

Kariuki, I. M., Loy, J.-P., and Herzfeld, T. (2012). Farmgate private standards and pricepremium: Evidence from the GlobalGAP scheme in Kenya’s French beans marketing.Agribusiness, 28(1):42–53.

Kleemann, L., Abdulai, A., and Buss, M. (2014). Certification and access to exportmarkets: adoption and return on investment of organic-certified pineapple farming inGhana. World Development, 64:79–92.

Latouche, K. and Chevassus-Lozza, E. (2015). Retailer supply chain and market access:evidence from French agri-food firms certified with private standards. The World Econ-omy, 38(8):1312–1334.

Lee, J., Gereffi, G., and Beauvais, J. (2012). Global value chains and agrifood standards:challenges and possibilities for smallholders in developing countries. Proceedings of theNational Academy of Sciences, 109(31):12326–12331.

Lernoud, J., Potts, J., Sampson, G., Voora, V., Willer, H., and Wozniak, J. (2015). TheState of Sustainable Markets–Statistics and Emerging Trends 2015. ITC, Geneva.

Maertens, M. and Swinnen, J. (2009). Food standards, trade and development. Review ofBusiness and Economics, 54(3):314–326.

Martinez, M. G. and Poole, N. (2004). The development of private fresh produce safetystandards: implications for developing mediterranean exporting countries. Food Policy,29(3):229–255.

Masood, A. and Brummer, B. (2014). Impact of GlobalGAP certification on EU bananaimports: a gravity modeling approach. GlobalFood Discussion Papers 49, RTG 1666GlobalFood. Georg-August-University of Goettingen.

Mausch, K., Mithofer, D., Asfaw, S., and Waibel, H. (2009). Export vegetable productionin Kenya under the EurepGAP standard: is large ‘more beautiful’ than small? Journalof Food Distribution Research, 40(3):115–129.

Moenius, J. (2004). Information versus product adaptation: the role of standards intrade. International Business and Markets Research Center Working Paper, Evanston:Northwestern University.

Prehn, S., Brummer, B., and Glauben, T. (2012). Structural gravity estimation & agri-culture. Diskussionspapiere, Department fur Agrarokonomie und Rurale Entwicklung1209.

Santos Silva, J. and Tenreyro, S. (2011). Further simulation evidence on the performance ofthe poisson pseudo-maximum likelihood estimator. Economics Letters, 112(2):220–222.

Santos Silva, J. M. C. and Tenreyro, S. (2006). The log of gravity. The Review of Economicsand Statistics, 88(4):641–658.

Schuster, M. and Maertens, M. (2015). The impact of private food standards on devel-oping countries’ export performance: an analysis of asparagus firms in Peru. WorldDevelopment, 66:208–221.

Shepherd, B. and Wilson, N. L. (2013). Product standards and developing country agri-cultural exports: the case of the European Union. Food Policy, 42:1–10.

20

Subervie, J. and Vagneron, I. (2013). A drop of water in the Indian Ocean? The impact ofGlobalGAP certification on lychee farmers in Madagascar. World Development, 50:57–73.

UNCTAD (2016). BANANA: an INFOCOMM Commodity Profile. Technical report,United Nations Conference on Trade and Development.

van der Meulen, B. M. (2011). The anatomy of private food law. In van Der Meulen,B., editor, Private food law: governing food chains through contract law self-regulation,private standards, audits and certification schemes, chapter 3, pages 75–111. WageningenAcademic Publishers.

Vandemoortele, T. and Deconinck, K. (2014). When are private standards more stringentthan public standards? American Journal of Agricultural Economics, 96(1):154–171.

Vigani, M., Raimondi, V., and Olper, A. (2012). International trade and endogenousstandards: the case of gmo regulations. World Trade Review, 11(03):415–437.

Will, M. and Pham, C. (2008). Estimating the gravity model when zero trade flows areimportant. Policy Research Working Paper 7308, The World Bank.

Yotov, Y. V., Piermartini, R., Monteiro, J.-A., and Larch, M. (2016). An Advanced Guideto Trade Policy Analysis: The Structural Gravity Model. WTO Publications, Geneva.

21

6 Appendix

Table A1: List of importing and exporting countries

Importing countries

EU 27 Austria, Belgium, Bulgaria, Cyprus, Czech Republic, Denmark, Estonia, Finland,France, Germany, Greece, Hungary, Ireland, Italy, Latvia, Lithuania, Luxembourg,Malta, the Netherlands, Poland, Portugal, Romania, Slovak Republic, Slovenia,Spain, Sweden, United Kingdom

Exporting countries by product

Apple Albania, Algeria, Argentina, Armenia, Australia, Austria, Azerbaijan, Belarus, Bel-gium, Bhutan, Bolivia, Bosnia and Herzegovina, Brazil, Bulgaria, Canada, Chile,China, Colombia, Croatia, Cyprus, Czech Republic, Denmark, Ecuador, Egypt,Estonia, Finland, France, Georgia, Germany, Greece, Hungary, India, Iran, Israel,Italy, Japan, Jordan, Kazakhstan, Kenya, Latvia, Lebanon, Lithuania, Luxembourg,Macedonia, Madagascar, Malta, Mexico, Moldova, Morocco, Nepal, Netherlands,New Zealand, Norway, Pakistan, Peru, Poland, Portugal, Republic of Korea, Roma-nia, Russian Federation, Slovakia, Slovenia, South Africa, Spain, Sweden, Switzer-land, Tajikistan, Tunisia, Turkey, Turkmenistan, Ukraine, United Kingdom, UnitedStates of America, Uruguay, Yemen

Banana Angola, Antigua and Barbuda, Argentina, Australia, Bangladesh, Barbados, Be-lize, Bhutan, Bolivia, Brazil, Brunei Darussalam, Burundi, Cabo Verde, Cambodia,Cameroon, China, Colombia, Congo, Costa Rica, Cyprus, Cote d’Ivoire, DominicanRepublic, Ecuador, Egypt, Ethiopia, Fiji, Ghana, Greece, Guinea, Guinea-Bissau,Guyana, Honduras, India, Indonesia, Iran, Israel, Jamaica, Jordan, Kenya, Laos,Lebanon, Madagascar, Malawi, Malaysia, Maldives, Mali, Mauritius, Mexico, Mo-rocco, Mozambique, Nepal, Nicaragua, Oman, Pakistan, Panama, Paraguay, Philip-pines, Puerto Rico, Rwanda, Saint Lucia, Saint Vincent and the Grenadines, Sey-chelles, South Africa, Spain, Sudan, Suriname, Tanzania, Thailand, Togo, Trinidadand Tobago, Turkey, United States of America, Vanuatu, Venezuela, Viet Nam,Yemen

Grape Albania, Algeria, Argentina, Armenia, Australia, Austria, Azerbaijan, Bolivia,Bosnia and Herzegovina, Brazil, Bulgaria, Canada, Chile, China, Croatia, Cyprus,Czech Republic, Egypt, Ethiopia, France, Georgia, Greece, Hungary, India, Iran,Iraq, Israel, Italy, Japan, Jordan, Kazakhstan, Kyrgyzstan, Lebanon, Luxembourg,Macedonia, Madagascar, Malta, Mexico, Moldova, Morocco, Namibia, New Zealand,Pakistan, Peru, Philippines, Portugal, Republic of Korea, Romania, Russian Fed-eration, Saudi Arabia, Slovakia, Slovenia, South Africa, Spain, Switzerland, Tajik-istan, Thailand, Tunisia, Turkey, Turkmenistan, Ukraine, United States of America,Uruguay, Venezuela, Viet Nam, Yemen

22

Table A2: PPML estimates without zero trade flows

Apple Banana Grapes

(1) (2) (3) (4) (5) (6)

ln GAPProducersit−1 0.160* -0.027 0.166**

(0.089) (0.066) (0.079)ln GAPHectares

it−1 0.092** -0.012 0.044**(0.045) (0.010) (0.019)

ln Productionit 0.019 0.020 0.270 0.266 -0.011 -0.015(0.039) (0.038) (0.184) (0.181) (0.011) (0.011)

ln GDPjt 0.980*** 0.970*** 0.259 0.263 0.694** 0.678**(0.274) (0.258) (0.413) (0.415) (0.346) (0.336)

RTAijt 1.834*** 1.814*** -0.025 -0.037 0.449*** 0.525***(0.145) (0.144) (0.078) (0.078) (0.075) (0.065)

ln Tariffijt 0.069 0.024 0.046 0.102 -0.025 -0.055***(0.115) (0.105) (0.255) (0.254) (0.020) (0.015)

Constant -23.030*** -22.711*** -27.424*** -27.477*** -20.652*** -20.637***(1.015) (0.874) (5.021) (4.968) (0.898) (0.876)

Observations 3,266 3,266 1,905 1,905 2,982 2,982R2 0.978 0.978 0.975 0.976 0.980 0.981

Notes: Country-pair clustered standard errors in parentheses. ***, **, * denote significance at 1%, 5% and 10% respectively.Year and dyadic fixed effects included in all regressions. Measure of GlobalGAP standard: columns (1), (3) and (5) usenumber of certified producers and columns (2), (4) and (6) use certified land area.

Table A3: Top exporters to the EU in Million USD, 2010 - 2015

Apple Banana Grapes

Main exporters Value Main exporters Value Main exporters Value

Italy 3.6167 Ecuador 8.0710 Italy 4.1170France 2.8458 Colombia 8.0427 South Africa 3.6064Chile 1.5295 Costa Rica 5.5206 Turkey 2.8622New Zealand 1.4094 Dominican Rep. 2.4295 Chile 2.8463Netherlands 1.2446 Cameroon 1.8170 Spain 1.8549South Africa 0.9671 Cote d’Ivoire 1.6153 Greece 1.2335Germany 0.9579 Panama 1.0940 Peru 1.0283Belgium 0.7508 Spain 0.4403 USA 0.8956Poland 0.6099 Belize 0.4264 Egypt 0.8651Brazil 0.4294 Suriname 0.3918 Brazil 0.8145Spain 0.4259 Ghana 0.3365 India 0.7421Austria 0.3954 Brazil 0.2118 Namibia 0.4060Argentina 0.3696 Mexico 0.1675 Iran 0.3672China 0.1513 Greece 0.0872 Argentina 0.3422USA 0.1259 Saint Lucia 0.0606 China 0.2029United Kingdom 0.1198 Honduras 0.0500 France 0.1455Czechia 0.0978 Cyprus 0.0086 Morocco 0.1303Slovakia 0.0945 Philippines 0.0082 Portugal 0.0457Portugal 0.0930 Nicaragua 0.0052 Israel 0.0410Hungary 0.0863 Saint Vincent 0.0048 Austria 0.0363

Source: UN Comtrade data and own calculations

23

Table A4: Summary Statistics

Apples Bananas Grapes

Variables Mean S.D. Min Max Mean S.D Min Max Mean S.D Min Max

Contiguityd 0.05 0.21 0 0.05 0.03 0.18Languaged 0.04 0.2 0.07 0.26 0.05 0.21Colonyd 0.03 0.17 0.04 0.19 0.03 0.17RTAd 0.58 0.49 0.36 0.48 0.49 0.5Religiond 0.17 0.38 0.16 0.37 0.18 0.38Distance 4225 4014 59 19586 7287 3054 243 18069 4861 4091 59 19586GDP Importer 648 961 8 3879 649 961 8 3879 650 962 8 3879Production 914 18808 0 396826 408 5980 0 120752 300 5821 0 115500GAP Producers 281 1372 0 12678 48 182 0 1052 112 477 0 5634GAP Hectares 2795 7088 0 47027 2079 8687 0 64862 2149 6607 0 49194GAP Certificates 84 254 0 1937 11 66 0 693 35 108 0 808Tariff 5.14 4.88 0 11.41 5.15 6.82 0 18.32 3.56 4.36 0 12.47Trade 1391 9430 0 292224 2506 21399 0 640772 2150 13781 0 317900

Observations 12129 12429 10602

Notes: All variables denoted by superscript d are dummies. Distance is measured in kilometres. Tariffs are measured in per-centages. Production is measured in tonnes.

24

Table A5: Selection equation of the 2-stage Heckman model

All products Apple Banana Grape

(1) (2) (3) (4) (5) (6) (7) (8)

ln GAPProducerst−1 0.205*** 0.051 0.036 0.117

(0.008) (0.056) (0.042) (0.076)ln GAPHectares

t−1 0.098*** 0.015 0.022 -0.009(0.004) (0.023) (0.019) (0.022)

ln Production 0.083*** 0.093*** 0.012 0.012 0.035 0.035 0.006 0.002(0.005) (0.005) (0.016) (0.016) (0.032) (0.032) (0.020) (0.021)

ln GDP -0.329 -0.312 -0.423 -0.420 -0.224 -0.221 -0.513 -0.507(0.221) (0.220) (0.452) (0.452) (0.535) (0.536) (0.497) (0.497)

ln Distance -0.862*** -0.862*** -0.998*** -0.998*** -1.286*** -1.285*** -0.835*** -0.834***(0.038) (0.037) (0.059) (0.059) (0.119) (0.119) (0.073) (0.073)

Language 0.370*** 0.365*** 0.511*** 0.511*** 0.510*** 0.511*** 0.224** 0.225**(0.058) (0.058) (0.121) (0.121) (0.108) (0.108) (0.111) (0.111)

Contiguity 0.911*** 0.903*** 1.043*** 1.044*** 0.816** 0.817** 0.945*** 0.946***(0.085) (0.085) (0.133) (0.133) (0.383) (0.383) (0.144) (0.144)

Colony 0.229*** 0.223*** 0.112 0.113 0.464*** 0.464*** 0.015 0.015(0.066) (0.066) (0.126) (0.126) (0.122) (0.122) (0.136) (0.136)

RTA 0.187* 0.303*** 0.424 0.433* 0.016 0.057 -0.107 -0.024(0.100) (0.101) (0.258) (0.256) (0.149) (0.155) (0.243) (0.235)

ln Tariff -0.067*** -0.061*** -0.065 -0.072 0.023 0.021 0.062 0.055(0.023) (0.023) (0.079) (0.078) (0.098) (0.098) (0.054) (0.055)

Religion 0.293*** 0.286*** 0.357*** 0.357*** 0.388*** 0.389*** 0.198*** 0.198***(0.039) (0.039) (0.067) (0.067) (0.092) (0.092) (0.076) (0.076)

Constant 6.920*** 6.837*** 11.684*** 11.832*** 9.576*** 9.546*** 11.619*** 12.358***(1.483) (1.478) (2.922) (2.912) (3.527) (3.529) (3.242) (3.231)

Observations 31,649 31,649 10,130 10,130 10,299 10,299 8,835 8,835

Notes: ***, **, * denote significance at 1%, 5% and 10% respectively. The dependent variable equals one if there is positivetrade between countries. Measure of GlobalGAP standard: ccolumns (1), (3), (5) and (7) use number of certified producersand columns (2), (4), (6) and (8) use certified land area. Importer, Exporter and Year fixed effects included in all regressions.Columns (1) and (2) also include product fixed effects.

25