Policy Research Working Paper 10581 Trading in Clusters and the Future of Small-Scale Trade in the Borderlands of the Great Lakes Region of Africa Célestin B. Bucekuderhwa Charles Kunaka Nyembezi Mvunga Douglas Amuli Ibale Finance, Competitiveness and Innovation Global Practice October 2023 Policy Research Working Paper 10581 Abstract This paper presents a coping strategy that small-scale income analysis and poverty decomposition highlight an cross-border traders adopted in response to the shock of increase in income, even within the depth and severity of the COVID-19 pandemic and examines its impact. Using poverty. Thus, although access to capital is important for a cross-sectional data set of 1,159 traders from the border- small-scale cross-border traders joining a cluster according lands of the Democratic Republic of Congo and Rwanda, to the literature, the results show that the level of capital is the paper assesses the impact of adopting the Cluster Trad- less important for income increase once there. As the results ing Approach on trade outcomes, household income and are robust to competing explanations and heterogeneity of poverty reduction. When applying a local average treatment the sample, the paper concludes that the Cluster Trading effect approach, the findings reveal that adoption of the Approach is a poverty-reducing strategy and discusses the CTA causes at least 21 and 31 percentage point increases challenges for its sustainability. in traders’ turnover and profit respectively. The household This paper is a product of the Finance, Competitiveness and Innovation Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at ckunaka@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Trading in Clusters and the Future of Small-Scale Trade in the Borderlands of the Great Lakes Region of Africa∗ elestin B. Bucekuderhwaa , Charles Kunakab , Nyembezi Mvungac C´ and Douglas Amuli Ibalea a ee au D´ Laboratoire d’Economie Appliqu´ e Catholique de Bukavu (UCB) eveloppement (LEAD)/Universit´ b Lead Transport Specialist, World Bank, Washington DC c Economist, World Bank, Lusaka, Zambia Keywords: Cluster, Small scale Traders, Impact, LATE, Income, Poverty. JEL codes: D02, D31, I32, C92 , F18, P25 ∗ The paper benefitted from support provided by the Umbrella Facility for Trade trust fund that receives contributions from the governments of the Netherlands, Norway, Sweden, Switzerland and the United Kingdom. Correspondence: C´ elestin B. Bucekuderhwa (Bucekuderhwa.Bashige@ucbukavu.ac.cd); Charles Kunaka (ckunaka@worldbank.org); Douglas AMULI Ibale (amuli.ibale@ucbukavu.ac.cd) 1 Introduction The COVID-19 pandemic has had significant effects on small-scale trade activities in the borderlands of the Great Lakes Region (GLR) of Africa (Mvunga and Kunaka, 2021). Yet small-scale trade plays an important role in the region, as a mechanism to meet basic re- quirements through its contribution to food supply, food access, sustenance of livelihoods, addressing local economic and social vulnerabilities, and thus, reducing some reasons for population displacement and conflict. The pandemic, and especially the measures that were adopted by governments to control its spread, had a profound impact on the organi- zation of trade Bouet et al. (2021), especially in the borderlands of the GLR (Keyser et al., 2022; Murhi and Mesa, 2021; Mvunga and Kunaka, 2021). The most visible measure from a trade perspective was the closure of some national borders to the cross-border move- ment of people. However, the dependence of communities in border regions, especially in the GLR, on supplies of basic requirements from the other side of the border required responses at the commercial and policy levels as otherwise there would have been severe disruptions to food and other goods supply systems. The compromise solution was a switch from individuals acting in their own self-interest to cooperation and collaboration among groups of traders. By adopting a cluster trading approach (CTA)1 , the numbers of people who needed to cross the border for purposes of trade could be reduced while at the same time allowing commerce to continue to flow, maintaining at the same time, an important source of income for thousands of Small-Scale Cross-Border Traders (SSCBTs). We conduct our analysis on the DRC side of the border with Rwanda, a region where labor market is markedly out of balance (Amuli, 2023). A rapidly growing population is continuously increasing the demand for jobs while, since the 1990s, the demand for labor has been falling steeply as a result of instability, conflict and other shocks to the economy (Amuli, 2023). In a state of high poverty, elevated unemployment and non-existent public provisions for unemployment insurance, cross-border trade increasingly is the main source of income for a large proportion of households. Trade contributes to food security, cre- ation of employment, provision of income and poverty alleviation and opens new markets for domestic products (Njoku et al., 2013). In so doing, it links food surplus areas to food deficit areas, especially in the context of an increasing concentration of people and activ- ities in towns and cities (Brenton et al., 2011). In response to the COVID-19 outbreak in March 2020, policy measures were introduced that resulted in the restricted movement of people, goods, and services across borders in the GLR, which greatly affected cross- border trade and traders (Mvunga and Kunaka, 2021). Within the context of the GLR, 1 Different terms have been used in the literature to name the collective actions conducted by individ- uals. Among others, we have group, association, club, cluster, etc. In this paper, we opt for Cluster to name the traders’ collective action 2 individuals who participate in small scale trade typically face very high costs, limiting the volume and prices of products that they trade. The restrictions that were imposed due to the pandemic meant that the costs of arranging market exchanges became even higher for such traders to the point that trade was now quite expensive for low-income people like SSCBTs. More generally, the context was an enormous source of stress to the economies of border regions in conflict zones, because the communities are trading in products that are time sensitive and yet the general environment is one that is characterized by great uncertainty, poor infrastructure and complex regulatory requirements. Suddenly, thou- sands of households had their sources of income and livelihood under threat, and this contributed to higher levels of household and individual poverty, increased dropouts of school children, health disruptions, etc. Under the circumstances, the formation of clus- ters or associations of traders, either those selling similar products or serving the same markets, emerged as a solution to the problem. Due to potentially important consequences on income, poverty and inequality, the economic literature has intensively explored collective action approaches and their benefits for participants. For example, one panel of studies analyzes the determinants of cluster membership (Bernard and Spielman, 2009; Fischer and Qaim, 2014; Keyser et al., 2022; Weinberger, 2021) and scrutinises structural and institutional aspects of clusters, such as clusters size, stringency of rules, commodity focus and market conditions (Barham and Chitemi, 2009; Hellin et al., 2009; Narrod et al., 2009). Another panel of research analyzes the benefit of joining a group. Some of these studies show that the group trading approach allows members to participate in new market developments (Bacon, 2005; Varangis et al., 2003) and to receive higher prices for their goods (Wollni and Zeller, 2007). Furthermore, the collective action displays the potential to increase per capita consumption (Gelo et al., 2019) and help correct some of the market imperfections, such as high transaction costs and missing credit markets, and fill in coordination gaps (Markelova et al., 2009). The limit of these studies is that they focus on the cause-and-effect relationship, leaving out the underlying mechanism at work. Moreover, these studies do not relate this impact to its spillover into the field of action of group members. In this paper, we provide new insights on the impact of collective actions while analyz- ing the mechanisms at work and relate the impact to its spillover effects. The contribution of this study is threefold: first, in quantifying the impact of CTA adoption on trade out- comes, we empirically test and discuss the mechanisms at work. In so doing, we show that the CTA adoption is linked with economy of scale development. Second, we relate this increase in trade outcomes to the increase in household income. Since cross-border trade is the main source of income for the majority of SSCBTs in GLR (Brenton et al., 2011), a lasting shock to trade outcomes could translate into a shock in household income. 3 Third, for policy implications, we relate changes in household income and poverty status by scrutinizing changes within the poor, and assess the SSCBTs’ willingness to pay for the sustainability of the clusters. Drawing on an empirical study of the borderlands of the Democratic Republic of Congo (DRC) and Rwanda, this paper provides insights on the impact of CTA as a strategy to mitigate the effects of the measures that were adopted to control the spread of COVID-19 on SSCBTs in two provinces of the DRC. To deal appropriately with the self-selection bias due to observables or unobservables and address the problem of non- compliance, we use a Local Average Treatment Effect (LATE). In so doing we assess the impact of CTA adoption on trade outcomes and analyse the changes in household income and poverty reduction as the spillover effect of trade outcome variation. The findings show that first, CTA adoption is associated with 21 and 31 percentage point increases in the turnover and the profit, respectively, of participating SSCBTs. The analysis of the underlying mechanism associates the CTA adoption with the economies of scale creation as the transaction cost decreases even as the volume of traded goods increases. Second, as the daily profit of SSCBTs shapes their total income, the monthly household income increases by more than 20% in per-capita terms regardless of whether cross-border trade is the primary source or not of household income. Finally, the poverty decomposition shows that income of the poor increases with the adoption of CTA. Thus, although sufficient capital is important for getting SSCBTs into clusters, consistent with the literature, our results show that the level of capital is less important for income increase of participants in CTA. As the results are robust to competing explanations and heterogeneity of the sample, we conclude that the CTA is a sustainable poverty-reducing strategy for which SSCBTs themselves are committed to maintaining. The findings should help policy makers to understand the effect of switching to or- ganized groups for SSCBTs in terms of the opportunities that become available but also challenges and risks that they should consider. In the next sections, we outline the liter- ature on the topic and the practical implementation of the CTA in the GLR in section 2. Section 3 describes the data and outlines the estimation strategies before presenting and discussing our results in section 4. Section 5 analyzes the competing explanations and the heterogeneity of our results and section 6 gives the implications and concludes. 2 Literature review. This section first explores the impact of COVID-19 on SSCBT and briefly outlines the policy responses that were introduced to alleviate the negative impact of the pandemic related measures and secondly, it explains the CTA and how it was implemented in the 4 GLR. 2.1 Policy responses to COVID-19 The outbreak of COVID-19 in December 2019 led to all countries adopting specific policy measures to control the spread of the disease. The policy measures ranged from physical distancing, to economic and health restrictions to lockdown that implied cross-country border closures (Gordon et al., 2021; Hale et al., 2021; Miao et al., 2021; Park and Chung, 2021). As a consequence, some rights were curtailed (Rambaree and Nassen, 2020; Zweig et al., 2021), supply chains were disrupted leading to an increase in costs of food and other products (Laborde et al., 2020; Ramsey et al., 2021; Schmidt and Gilbert, 2021; Swinnen and R., 2021), employment decreased (Guerrero-Amezage et al., 2022; Schmidt and Gilbert, 2021), and per-capita incomes reduced (Gordon et al., 2021), affecting in particular, the livelihoods of economically vulnerable groups (Mvunga and Kunaka, 2021). The measures took a heavy toll on informal, small-scale cross-border activities (Murhi and Mesa, 2021; Mvunga and Kunaka, 2021). In an effort to alleviate the negative impact of the pandemic related measures, some governments provided direct financial support to the population (Karim et al., 2021; Sengupta and Jha, 2020) while others introduced different social assistance interventions (Dafuleya, 2020). Some responses focused on structural changes in social and economic security by developing response packages to cushion members pushed into poverty by the pandemic while strengthening financial institutions to support the recovery of businesses in the medium term and ensuring the resilience of food supply chains, particularly, those making available nutrient-dense foods (Kansiime et al., 2021). However, as found by Kansiime et al. (2021), countries relying on international aid were slow to react as they were not able to commit scarce domestic resources towards the poor through cash and food transfers. In the Great Lake regions, the border closures entailed a negative shock on cross border trade that is a main source of income of SSCBTs. To mitigate this negative impact on SSCBTs, the GLR decided to pilot a new approach that could help SSCBTs to cope with their vulnerability and protect their sources of income by continuing their business activities beyond the border closure. The new approach involved grouping SSCBTs in clusters and providing them with support for trade logistics. 2.2 Trading in clusters and its implementation The ”Great Lakes Trade Facilitation Project” (hereafter called GLTFP) is a World Bank project that started in 2015, aimed at facilitating small scale cross border trade in the Great Lakes region. The project involves three countries, namely DRC, Rwanda and 5 Uganda. Its activities include training of government border officials and SSCBT on trade and individuals’ duty and rights to improve cross-border trader livelihood, mitigat- ing gender-based violence at the borders, providing adequate border infrastructure, and enhancing public revenue at the border, etc. The project is implemented by the ministries responsible for Trade with the support of the World Bank Group. The Trading in clusters strategy has been proposed by the GLTFP since 2018 through the process of structuring and professionalizing the different groups and associations of Small Cross-Border Traders (ACT). It started during the appearance of the Ebola Virus and aimed to reduce in- formality and its corollaries in the cross-border exchanges. But its importance became accentuated with the appearance of COVID-19 in December 2019 when restrictions were imposed to crossing international boundaries. This greatly affected communities that participated in trade by headloading, one of the main means of transport of SSCBTs.2 Consequently, the source of income of thousands of households involved in cross-border trade was under threat. To mitigate this sudden negative shock, the GLTFP countries adopted the Cluster Trading Approach (CTA) as an innovative way to allow SSCBTs to continue their business activities beyond the border closure and sustain their livelihoods. The approach, consists of forming different groups of SSCBTs based on defined criteria (like type of product sold, proximity between SSCBTs, ...). Within each cluster, money for the purchase of goods is collected from each member and the cluster (through an elected supervisor) places the purchase order for all its members in the selling country. The sellers, from a neighboring country, after receiving the payment, pack the goods for transport and delivery to a counterpart cluster in the receiving country. Ultimately, the SSCBTs on one side either receive or sell goods without crossing the border and can then continue their business and mitigate the consequences of cross-border restrictions. The aim is to facilitate trading activities through (1) provision of selective services to their members and (2) increasing the collective benefits (Benett, 1998). In addition to the aim of mitigating the negative shock of cross-border restrictions, trading in clusters was expected to reduce the wholesale price of goods and services and other costs related to transport and border payment (Che et al., 2015), offering solutions to the collective action problem (Benett, 1998). In general, collective action leads to greater efficiency through the reduction of transaction costs; time saving, improved marketing coordination, access to market information, better bargaining power, improved quality and quantity of products and better economies of scale, and therefore improving net incomes (Kersting and Wollni, 2012; Markelova et al., 2009; Narrod et al., 2009; Wollni and Zeller, 2007). Nevertheless, trading in clusters as a collective action may introduce unintended consequences such as 2 In reference to the ”Ordonnance n° 20/014 of March 24, 2020 proclaiming a state of health emergency to deal with the epidemic of COVID-19.” 6 the risks of adverse selection, free riding and elite capture (Che et al., 2015; Xiao et al., 2017). Furthermore, there may be difficulties with achieving and maintaining compliance, fluidity of membership with some participants opting out, instability, and a preference for industry interests rather than public interests. Moreover, the adopted facilitation policies can have limits. On the one side, they can require some logistic facilities that can be provided only at relatively high cost, barely affordable by SSCBTs. In fact, the evidence from elsewhere suggests that logistics related services such as transport, warehousing, and marketing can become important non-tariff barriers in trade facilitation (Madkour et al., 2020; Marti et al., 2014; Zaninovic et al., 2021). On the other hand, the prevalence of complex and cumbersome border procedures and other forms of institutional trade costs in Sub-Saharan African economies can inflate the costs of moving goods across borders. Given this, further trade facilitation actions will need to focus on the border documentation, time and real costs of trading across borders (Sakyi et al., 2017) and on providing some logistics services to small traders to develop and afford the cost of new modalities for conducting their operations. 3 Data and methodological approach. This section describes the data used in the paper and the estimation approach that was implemented. 3.1 Data and descriptive evidence. To assess the impact of adopting the CTA on SSCBT economic outcomes, we collected data in two provinces of the DRC, namely, North and South Kivu, from April to June 2022.3 The two provinces include several markets and share four border crossing points with Rwanda. Thus, the sampling frame for the study is made up of active small cross- border traders. During the survey in the field, SSCBTs that were included in the sample were those operating within physical market structures and on roadsides. To collect data, we first list all the markets and mains roads used by SSCBTs to sell their products. We exhaustively consider all these markets within Goma and Bukavu, and 3 The sample includes both CTA’s adopters and non-adopters. Indeed, at the time of data collection, the Government of Democratic Republic of Congo had already eased some restrictions on the movement of people and goods. Indeed, on July 21, 2020, the government announced: (1): As of July 22, 2020, the resumption of commercial activities, stores, banks, restaurants, cafes, bars, and businesses, as well as the resumption of rallies, meetings, and celebrations, and public transportation. (2): The resumption of schools and universities was set for August 3. (3): From August 15, the reopening of churches and places of worship, the resumption of inter-provincial migration movements, the opening of ports, airports and borders. Thus, at the time of the collection, there were two groups of individuals, those who adopted and kept trading through the CTA and those who did not adopt and were still crossing the border for trade. 7 we purposively select roads within each town. After that, SSCBTs were randomly selected within markets and roads. The analysis is based on a sample of 1,159 respondents, after data cleaning. Evidence from Table 1 reveals that the majority of respondents (84.38%) Table 1: SSCBTs socio-economic characteristics by adoption status Characteristics Total sample Adopters Non-Adopters Size 1159 522 637 Proportion of female SSCBT(%) 84.38 83.91 84.77 Age (average) 36.51 37.7 35.52 Number of years as SSCBT (average) 9.04 9.81 8.42 Education: No formal education (%) 12.94 12.26 13.50 Primary school (%) 38.40 33.91 42.07 Secondary school (%) 45.04 51.15 40.03 Tertiary school (%) 3.62 2.68 4.40 proportion of Maried (%) 68.00 72.80 65.62 Living in urban (%) 97.50 98.66 96.55 Trade as mains source of income (%) 94.82 96.55 93.41 Have a fix place in the market (%) 94.39 97.51 91.84 Member of an association (%) 62.81 91.00 39.72 Household size (average) 6.61 6.84 6.42 Nord Kivu 516 229 287 South Kivu 643 293 350 Source: Our own computation based on survey data. and 83.91% of the adopters of CTA were female. At the time of the survey, the average age of the SSCBT was 36 years. The average household size of respondents (both adopters and non-adopters) was 6 people per family; about 95% of respondents were living in urban areas and had spent an average of about 9 years in cross-border trade. The educational level of the SSCBTs was different between adopters and non-adopters. Whereas 55.5% of the non-adopters had at least primary school level of education, only 46% of adopters had a similar level of education. However, at least 54% of adopters reached secondary education against 44% of non-adopters. In addition, the majority of respondents, both adopters and non-adopters have a fix place in the market where they sell their goods. The adoption rate in the two provinces is 44.3% and 45.5% in North Kivu and South-Kivu, respectively. To implement the CTA, the GLTFP worked with mass media (radio and television), officials, existing SSCBTs’ associations and groups of interest to spread the awareness and sensitize SSCBTs in order to organize themselves in groups. In DRC, the dissemination of CTA was implemented in only a few selected markets and regions. This means that the overall population of SSCBTs was not equally exposed to the CTA (the instrument for the policy intervention was not randomly distributed). On the other hand, SSCBTs exposed to the CTA had full control over their decision to adopt or not to adopt (i.e. the receipt of the treatment is endogenous). According to the impact assessment literature, the most plausible assumption to make in this case is that of selection on 8 the unobservable (Diagne et al., 2009; Imbens and Wooldridge, 2009). This is because SSCBTs decided to adopt the CTA based on the anticipated benefit they would derive by adopting it and this anticipated benefit cannot be observed. Hence, to identify and estimate the impact of CTA adoption, we need an instrument that is independent of this unobserved anticipated benefit and can affect income and poverty only through the act of adoption. 3.2 Estimation strategies. The study first uses impact evaluation methods to assess the impact of the CTA adoption on SSCBTs’ trade outcomes and household income. Thereafter, it decomposes the SS- CBTs into various poverty statuses to assess how the adoption affects poverty and income distribution and analyzes the SSCBT’s Willingness To Pay (WTP) for the sustainability of the clusters with an interval regression. The following subsections briefly explain the methodological approach. The Local Average Treatment Effect (LATE) Under the potential outcome framework developed by Rubin (1974), each SSCBT has ex-ante two potential outcomes: when adopting the CTA; denoted by y1 and when not adopting it; y0 . If we let the binary outcome variable d stand for CLUSTER adoption status, with d = 1 meaning adoption and d = 0 non-adoption, we can write the observed outcome y of any SSCBT as a function of the two potential outcomes: y = dy1 + (1 − d)y0 (1) For any SSCBT, the causal effect of the adoption on their observed outcome y is simply the difference between two potential outcomes (y1 − y0 ). From equation (1), one can measure the mean effect of adoption on a population of SSCBT (such parameter is called the average treatment effect (ATE)) and the mean effect on a subpopulation of adopters (called the average treatment effect on the treated (ATT)). To obtain con- sistent estimators of causal effects, two broad categories of models based on the types of assumptions they require are proposed (see Imbens (2004); Imbens and Wooldridge (2009)). First, there are the methods designed to remove overt bias only. Based on the ‘ignorability’ or conditional independence assumption (Rosenbaum and Rubin (1983); Rubin (1974)), they provide estimators via a two-stage estimation procedure where the conditional probability of treatment P (d = 1|x) ≡ P (x) (called the propensity score), is estimated in the first stage and ATE and ATT are estimated in the second stage by parametric regression-based methods or by non-parametric methods. The latter include 9 various matching method estimators (see Dontsop et al. (2011) for a non-exhaustive list). For this study, the conditional independence-based estimators of ATE and ATT are the so-called Propensity Score Matching estimators (PSM). In that case, our model takes the following form : Yi = βCT Ai + λXi + ϕt + θb + γp + εitbp (2) Where i indexes the individual, t the type of product, b the border and p the Province. In Equation 2, Y is the outcome, CTA, a dummy variable for whether or not individual i has adopted the CTA, X a vector of covariates, ϕ is the product dummies, θ Border dummies and γ province dummy. The vector of controls, Xi , includes variables such as gender, age, education, living in urban, marital status, experience as SSCBT, whether the SSCBT has a fix place in the market and the logarithm of purchased volume amount. Secondly, there are instrumental variable (IV)-based methods (Abadie (2003); Heck- man and Vytlacil (1999, 2005,?); Imbens (2004); Imbens and Angrist (1994); Manski and Pepper (2000)), which are designed to remove both overt and hidden biases and deal with the problem of endogenous treatment. The IV-based methods assume the existence of an instrument called Z , that explains treatment status but is redundant in explaining the outcomes y1 and y0 , once the effects of the covariates x are controlled for. Following Abadie (2003), Imbens and Angrist (1994), we use the IV-based estimators to estimate the Local Average Treatment Effect (LATE) of the CTA adoption on trade outcomes, on income and poverty. Within the general impact model, our models from Equation 2 take the following form: Yi = βCT Ai + λXi + ϕt + θb + γp + εitbp ′ CT Ai = αZi + ν (3) where Y , X , ϕ, θ and γ are as in Equation 2. Z is a vector which includes a binary instrument and β is interpreted as the return to CTA adoption. Following the Imbens and Angrist (1994) LATE estimator and that of Abadie (2003), we note, as in Dontsop et al. (2011), that an SSCBT’s exposure status4 to the CTA (i.e. his awareness of the existence of CTA) is a ‘natural’ instrument for the CTA adoption status variable. This is because, first, awareness does cause adoption. Indeed, an SSCBT cannot be part of the CLUSTER without being aware of it and its mechanism. Secondly, it is natural to assume that it is only through adoption that the exposure to CTA affects the overall SSCBT income and poverty outcome indicators.5 4 For the 637 Non-adopters, 52.28% were aware of the existence of the CTA. For the 522 Adopters, 100% was aware of the existence of the CTA (see Table 10 in the appendices). Thus the monotonicity condition (P (d1 ⪰ d0 | x) = 1) necessary for LATE application was fulfilled (Dontsop et al., 2011). 5 The mere awareness of the existence of a CTA without adopting it does not affect the outcome 10 Hence, the two requirements for the CTA exposure status variable to be a valid in- 6 strument for the CTA adoption status variable are met. Now, let z be a binary outcome variable taking the value 1 when a SSCBT is exposed to the CTA and the value 0 oth- erwise. Let d1 and d0 be the binary variables designating the two potential adoption status of the SSCBT with and without exposure to the CTA trading mechanism. When applying Abadie (2003) LATE estimator, which requires the conditional independence assumption (i.e, The instrument z is independent of the potential outcomes d1 , y1 and y0 conditional on a vector of covariates X determining the observed outcome y ) the result f (x, d) ≡ E (y/x, d, d1 = 1) holds for the conditional mean outcome response function for potential adopters and any function g of (y, x, d) (Abadie, 2003; Lee, 2005). The function f (x, d) is called a Local Average Response Function (LARF) by Abadie (2003). Consid- ering some threats to identification of the CTA effect, we discuss in the robustness section (Section 5) First; the positive selection of individuals within the analytical sample due to attrition in Subsection 5.1, second; the positive selection of SSCBTS within clusters due to possible reverse causality in Subsection 5.2 and lastly, Third; the heterogeneity of our analytical sample in Subsection 5.3. Diagnostic test of the validity of the instrument In testing the relevance and validity of instruments we essentially test that the instru- ments are correlated with the included endogenous variable(s) and orthogonal to the error process. The degree of correlation to the endogenous variables is easily tested by examin- ing the fit of the first-stage regression after the included instruments are ‘partialled-out’ (Bound et al., 1995) and the F-test associated with it. If the explanatory power in the first stage is positive but, nevertheless, weak conventional asymptotics fail; this can happen even when the first stage tests show significance at conventional levels of significance and a large sample (Baum et al., 2003; Staiger and Stock, 1997). One rule of thumb that can be used is that an F-test below 10 is a cause of concern. In order to ascertain the independence of instruments from the unobservable error process, with an overidentified equation and given L instruments, one needs to test the corresponding orthogonality conditions arising from the L instruments. Tests such as Sargan (1958) statistic – a special case of the J statistic of Hansen (1982) – are used here, and in the presence of heteroscedastic errors, a ‘robust’ Sargan’s statistic7 . indicators of a SSCBT. 6 The usual third requirement that the instrument be “uncorrelated with the unobserved error term” made in classical IV can be weakened by the Abadie (2003) generalization of the LATE identification estimation through the Local Average Response Function (LARF). 7 This statistic is numerically identical to the J statistic derived from a two-step GMM for that equation (see, Baum et al. (2003)). 11 Finally, in testing for the consequences of employing the estimation method of instru- mental variables (test of the endogeneity or exogeneity of regressors), one acknowledges the trade-off between a possibly biased and inconsistent OLS estimator and the higher asymptotic variance of the IV estimator. The test employed here is the Durbin (1954); Wu (1973) and Hausman (1978)’s version of the Hausman statistic, using the OLS estimate of the error variance, as opposed to the IV estimate.8 The Poverty Decomposition Model We use the Foster, Greer and Thorberke (1984) (therafter called FGT) poverty model to decompose SSCBT into various poverty statuses. The FGT class is based on the normalized gap gi = (z − yi )/z of a poor person i, which is the income shortfall expressed α as a share of the poverty line (z ). Viewing gi as the measure of individual poverty for a α poor person, and 0 as the respective measure for non-poor persons, Pα = (1/(n))gi is the average poverty in the given population. The procedure entails estimating the Incidence of individual Poverty (α = 0), the depth (α = 1) and severity (α = 2) of poverty among SSCBT. Income changes resulting from adoption of CLUSTER trading approach and changes in the number of SSCBT and depth of poverty were also estimated. The interval regression for wilingness to pay(WTP) Willingness to pay (WTP) is a key component of consumer demand. It helps to set the prices at a level that allows firms to be sustainable over time by maximizing profits and customer satisfaction. In the case of the SSCBT studied, a monthly contribution set by the CLUSTER is paid by the members to unsure its sustainability in the future. Given this, we want to study the variables (especially the gender, as women are the core component of SSCBT) that determine the SSCBT’s WTP for the sustainability of the CLUSTERS over time. As the values surveyed on the WTP for the sustainability of the cluster are not continuous but consist of intervals and censoring observations. The Interval Regression Approach is the main technique to tackle such a data structure (Jin-Long and Zhang Yi, 2012; Louinord et al., 2013; Riccioli et al., 2019). The interval regression model assumes that WTP is a function of a series of explanatory variables. Thus, we assume the following function: W T P = βxi + ξi (4) where WTP represents people’s true WTP level; xi represents a vector of independent variables, b is the corresponding vector of estimated parameter and ξi is an error term 8 Under the null hypothesis both estimates are consistent, but the OLS estimate is more efficient. Furthermore, the chosen flavor of the test has the additional advantage of performing better when the instruments are weak (Baum et al., 2003; Staiger and Stock, 1997). 12 following a normal distribution with a mean of zero and variance of δi . Although the true WTP level cannot be observed directly, respondents state their lower and upper bound of WTP for the sustainability of the cluster. The study estimated the mean WTP using maximum likelihood methods based on the “intreg” procedure of STATA with a robust error estimation. 4 Estimation results. The section presents four sets of results. First, it presents the baseline results on the impact of CTA adoption on SSCBT’s trade outcomes that are the daily turnover and profit, using the LATE approach. Thereafter, it describes the potential mechanisms using OLS and Logit techniques. Second, it relates change in trade outcomes with its spillover effect by assessing the effect of CTA adoption on household incomes and poverty reduction. Third, it assesses SSCBT’s willingness to pay for the sustainability of the Clusters with an interval regression approach. Fourth and ultimately, it conducts robustness check for competing explanations and assesses the heterogeneity analysis using quasi-homogeneous sub-samples. 4.1 Mean difference in trade and household outcomes Table 2 presents the mean difference analysis of the impact of CTA adoption in terms of daily profit, daily turnover, daily volume of purchase, monthly spending and monthly income. A straightforward comparison between both household per capita income and per capita expenditure of adopters and non-adopters was considered. Table 2: SSCBTs socio-economic characteristics by adoption status Characteristics Total sample Adopters Non-Adopters difference n=1159 n=522 n=637 Test Daily Profit (average in CDF) 16881.9 21799.1 12852.4 8946.6*** Daily Turnover (average in CDF) 181992. 265110.6 113880.0 151230.6*** Daily Volume purchace (average in CDF) 123843.8 167462.8 88099.5 80309.8*** Monthly spendings (average in CDF) 414270.1 479050.0 361185.1 121805.1*** Monthly per capita spendings (average in CDF) 70674.5 78512.5 64224.3 14446.9*** Monthly income (average in CDF) 471532.0 602929.7 363855.9 250018.5*** Monthly per capita income (average in CDF) 87898.4 115477.9 65201.8 52127.2*** Headcount ratio (%) 82.55 77.68 86.55 Poverty gap ratio (%) 56.74 51.99 60.65 Extreme Poverty Headcount ratio (%) 49.69 45.64 53.02 Source: Our own computation based on data collected ***,**,* Denote significance at 1%, 5% and 10% levels. While household income indicates the ability of the household to purchase its basic needs, per capita expenditure reflects the effective consumption of households and there- 13 fore provides information on the food security status within households. Furthermore, as is evident from the Table 2, the incidence of poverty was higher among non-CTA adopters (86.55%) than CTA adopters (77.685%). In addition, both the depth and severity of poverty were also higher (60.65% and 53.02%) among non-adopters than the adopters (51.74% and 45.64%). All three poverty measures indicate that poverty was more preva- lent and severe among non-adopters compared to adopters. The result reported in the fifth column of Table 2 shows that there is a significant difference between all trade and house- hold characteristics of adopters and non-adopters. The mean differences in per capita income and expenditure of adopters and non-adopters indicates that adopters of CTA are better off than the non-adopters. However, the differences in observed mean outcomes between adopters and non-adopters cannot be attributed entirely to CTA adoption due to the problem of self-selection and non-compliance (Heckman and Vytlacil, 2005; Imbens and Angrist, 1994). The impact of the adoption of CTA on trade and household outcomes is discussed in the next section 4.2 The impact on trade outcomes Table 3 reports estimates of Equation 3 for the two trade outcomes. The dependent variables are continuous and log transformed while the independent variable of interest is binary, indicating that the CTA adoption would increase the turnover of SSCBT (first panel) and in turn would increase their profit (second panel). Column 1 reports uncon- ditional LATE estimate. This gives descriptive relationship between the CTA adoption and trade outcomes. Column 2 controls for Socio-Demographic characteristics that can influence both the turnover and the profit of SSCBT, Column 3 controls for borders fixed effects and column 4 add the type of product sold fixed effects and finally, the column 5 controls for provinces fixed effects. This allowa us to control non-parametrically for all potentially omitted variables that can vary across borders, type of products and province levels. All Columns of Table 3 show a positive and statistically significant effect of CTA adoption on trade outcomes, both the daily turnover and the daily profit. The most conservative estimates in Column 5 restrict all variation to within-type of product sold, within borders and within-province. Taking it as the benchmark specification, we find that trading in a cluster leads to a 21% increase in the daily turnover and 31% increase in the daily profit of SSCBT. Given that the mean level of these variables are 181,992.5 CDF (Congolese francs) and 16,881.9 CDF respectively as reported in Table 1, the effects of CTA adoption are sizable. Importantly, CTA adoption not only has an impact on the daily turnover but also shape the daily profit of traders which represents the share of trade activities in the SSCBT’s total income. This suggests that, SSCBT’s shift to CTA 14 should result in an increase in total income of SSCBTS’ households. Table 3: LATE estimates of the Effects of CTA adoption on SSCBT’s trade outcomes (1) (2) (3) (4) (5) Outcome Daily Turnover CTA adoption 0.956*** 0.275*** 0.269*** 0.211*** 0.212*** (0.122) (0.092) (0.091) (0.101) (0.101) Observations 1159 1145 1145 1145 1145 F Statistics (excluded instruments) 1337.96*** 885.98*** 879.89*** 629.14*** 631.05*** Under identification test p value 0.0000 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 61.24*** 8.86*** 8.59*** 4.19** 4.20** Cragg-Donald Wald F 475.718 392.85 400.19 347.12 336.16 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) (16.38) Outcome Daily Profit CTA adoption 0.782*** 0.360*** 0.358*** 0.310*** 0.313*** (0.101) (0.088) (0.087) (0.093) (0.093) Observations 1159 1145 1145 1145 1145 F Statistics (excluded instruments) 1337.96*** 885.09*** 879.89*** 629.14*** 631.05 Under identification test p value 0.0000 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 59.43*** 16.87*** 17.02*** 10.83*** 11.05*** Cragg-Donald Wald F 475.7 392.85 400.19 347.12 347.15 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) (16.38) √ √ √ √ Socio-demographic Controls √ √ √ Borders fixed effects √ √ Type of Product fixed effects √ Province fixed effects Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Socio-demographic Controls includes variable such as gender, Age, Education, Living in urban, Marital status, Experience as SSCBT, whether the SSCBT has a fix place in the market and the logarithm of purchased volume amount. To corroborate this significant impact, we applied two additional treatment effect models, namely the propensity score matching (PSM)9 and the nearest-neighbor match- ing techniques (NNM).10 In comparison to LATE method, PSM and NNM are free of instruments use and impose fewer constraints on the functional form of the treatment model, as well as fewer assumptions about the distribution of the error term. The estima- tion of Equation 2 shows that, though the coefficients provided are different in magnitudes, the result reported in Table 6 give an ATE and an ATT for the two methods (PSM and NNM) applied on the two trades outcomes (Turnover and Profit), are positive and signifi- cant, which confirm the results of LATEs estimators. The results are in line with Kariuki 9 For the two approaches, ATT and ATE are estimated first, by propensity-score matching (when using PSM). In this case, It does not need bias correction, because PSM matches on a single continuous covariate. Second, by nearest-neighbor matching (when using NNM), and it is based on a weighted function of the covariates for each observation. Furthermore, it uses a bias-correction term when matching on more than one continuous covariate (STATA, 2017). 10 Both NNM and PSM commands have a nneighbor(#) option in STATA. But this only specifies the number of matches per observation. The criteria used in matching differs for both commands. teffects nnmatch, for the NNM model, uses a ”distance” metric criteria, while teffects psmatch, for the PSM model, uses the criteria of the estimated predicted probabilities of treatment. 15 and Place (2005), Verhofstadt and Maertens (2014) and Mutunga (2015) who found that clusters that work as collective actions contribute to increased income for members. We also follow the method proposed by Oster (2019) to investigate whether our re- sults could be driven by unobservable factors.11 Using the methodology presented in Altonji et al. (2005) and recently refined by Oster (2019), we check how large selection on unobservable would need to be to explain away our main effect. The technique relies on comparing the coefficient of interest and the R-squared between regressions with and without control variables to gain insights into the importance of omitted variable bias. Here, we focus on the calculation of the delta, which indicates the degree of selection on economic unobservables, relative to observables, needed for our results to be fully ex- plained by omitted variable bias. Table 8, Panel 1 shows the results from our analysis. From Table 8, columns 1 and 2 show that in a regression of daily turnover on the treat- ment dummy, adding controls reduces the coefficient in absolute value while increasing the R-squared. The corresponding delta indicates that selection on unobservables would have to be more than three times as large as the selection on observed controls to make the effect in column 2 go to zero, a value well beyond the threshold of one.12 Columns 3 and 4 present the results for the daily profit. As in the two first column, the inclusion of controls reduce the coefficient on the treatment dummy while increasing the R-squared. The corresponding delta implies that to explain away the impact in column 4, unobserv- ables would have to move the coefficient in the opposite direction as observables, and their influence would have to be more than four times as large. Taken together, the re- sults in Table 8, Panel 1 thus strongly suggests that omitted variable bias is not driving our results. 4.3 The Mechanism In this section, we describe potential mechanisms that can explain the relationship be- tween CTA adoption and the trade outcomes of SSCBTs. To do so, we combine a quantita- tive exploration with qualitative insights from fieldwork13 conducted in the two provinces. As regards, we identified four trade channels, namely transport cost, transaction time, volume of goods and access to credit that appears to significantly explain this relationship between CTA adoption and trade outcomes. 11 To test this, we use OLS regression as the formalized test by (Oster, 2019) only applies to OLS estimates. 12 The rule of thumb to be able to argue that unobservables are unlikely to fully explain the treatment effect is for Oster’s delta to be over the value of one Oster (2019). 13 We undertook qualitative fieldwork in the two provinces with focus group and interviews from April to June 2022 to collect data on cross-border trade related to trade in cluster. Because of the high frequency of SSCBT, we performed random sampling for participants in 30 interviews and two focus group with SSCBT. In addition, we conducted 10 interviews with officials at the border. 16 a. Transport Cost Higher transport costs significantly deter trade (Martinez-Zarzoso et al., 2008). Indeed, without clusters, traders pay individually the transport costs and the informal fees as- sociated with crossing the border. With CTA, the cost of transport is shared among members. Furthermore, since they are no longer physically involved in the logistics of getting goods across the border, the informal costs are also significantly reduced. This reduction in transportation costs and informal fees paid is likely to increase the return of the SSCBT’s trade activities. To test this, we regress our benchmark model (Column 5 of Table 3) on the perceived level of transport cost of SSCBT. As the variable is captured on a scale of one to five, we use both OLS and logit regressions (where the variable is dummy transformed, see explanation in 11) to check how it is affected by CTA adoption. The result reported in Table 11 Panel 1 show a negative and significant effect for both OLS (-0.279) and Logit (-0.172) regressions. This implies that CTA adopters are more likely to bear low transport cost. b. Volume of transactions If SSCBT face a lower cost of transport, enjoy trade transactions that are less time consuming and can access to credit within the group, they are likely to buy more quantity of goods. The greater the quantity of goods traded, the higher is the sales turnover and the larger the profit. We thus investigate the role of CTA in influencing the volume of good traded. To test this we regress our benchmark model on the perceived level of the volume of good bought for trade. As the variable is captured on a scale of one to five, we use both OLS and logit (where the variable is dummy transformed) regressions to check how it is affected by CTA adoption. The result reported in Table 11 Panel 3 show a positive and significant effect for both OLS (0.309) and logit (0.128) regressions. The result implies that CTA adopters are more likely to deal with a larger volume of goods. Our results are in line with Markelova et al. (2009)’s conclusion. Acting collectively enable participants to operate on a larger scale. The combined effects of the previous result that supports a decrease in transport cost with the increase in volume of transaction, shows that CTA adoption is associated with economy of scale. c. Time of trade Transactions When they are not in a cluster, SSCBTs are forced to start the day early to join queues while waiting for the border to open. Waiting time can be more than three hours as the registration processes are manual and therefore very slow. However, some individuals who came late can pay informal fees to the officials and are able to skip the queue, 17 making crossing times longer for everyone else. The same phenomenon can also happen on return. The non-cluster traders therefore can face high costs (Poulton et al., 2010). However, when trading in clusters, SSCBTs claim to have more time to spend at the market, which improves the profitability of their activities. To test this, we regress our benchmark model on the perceived transactions time. As the variable is captured on a scale of one to five, we use both OLS and logit regressions (where the variable is dummy transformed) to check how it is affected by the CTA adoption. The result reported in Table 11 Panel 2 show a negative and significant effect for both OLS (-0.151) and logit (-0.102) regressions. In line with Gelo et al. (2019), our result implies that CTA adopters are more likely to spend few transactions time, which increase the time allocated to selling on the market. d. Acess to credit Given the law level of their income, the lack of guarantees and the informal status of their activities, SSCBTs are excluded from access to credit with classic financial institutions. With CTA, SSCBTs are able to save money on a regular basis. The savings are used to provide members with credit within the clusters. According to SSCBTs, the credit enables them to increase their purchases and to mitigate the effects of unexpected events such as illness or similar shocks. To test this, we regress the CTA adoption on the dummy variable, access to credit, which is equals to one when SSCBTs has access to credit and zero otherwise. The results in Table 11, Panel 4 show that CTA adoption increases significantly the probability to get a credit by 0.061 percentage point. CTA adopters have more access to credit, which increases their purchasing power and in turn their turnover and profit. Acting collectively correct some of the market imperfections such as missing credit markets (Markelova et al., 2009). 4.4 The impact on household income As seen from Section 4.2, CTA adoption shapes the daily profit of traders which represents the share of trade activities in the SSCBT’s total income.14 In fact, changes in household income could be result of a spillover effect of the change in trade outcomes. In this section we check whether CTA adoption affects household income. To do so, we restrict our sample to SSCBTs with a dependency ratio greater than zero.15 We then analyze this subsample on the one side and on the other side a sub-sub-sample of SSCBTs household 14 Data on income were collected using different items that reflect possible sources of current household income. These items are listed in Table 9 in the appendices. 15 Dependency ratio is the ratio between dependent individuals and the household size. For this purpose, single SSCBTs with dependency ratio equal to zero are removed from the sample. 18 for which cross-border trade is the main source of income.16 We then check whether CTA adoption influences monthly per-capita spending17 and monthly per-capita income. The results reported in Table 4 show that the impact of CTA adoption on Household income is positive and significant. The CTA adoption is associated with an increase in monthly household expenditures of more than 20% and an increase in monthly household income of more than 30%, regardless of whether cross-border trade is the primary source of household income. These numbers are the average change in per capita expenditure and per capita income of households that belongs to a change in trade status. To corroborate this significant impact, we applied two additional treatment effect models, namely the propensity score matching (PSM) and the Nearest-neighbor matching techniques (NNM). The result reported in Table 7 give an ATE and an ATT for the two methods (PSM and NNM) applied on the two household outcomes (Spendings and Income), are positives and significant.This confirms the results of LATEs estiators from 4. We also follow the method proposed by Oster (2019) to investigate whether these results on household outcomes could be driven by unobservable factors. Using the technique explained in the section 4.2, we find that the corresponding deltas in Table 8, Panel 2 imply that to explain away the impact in columns 2 and 4, unobservables would have to move the coefficient in the opposite direction as observables, and their influence would have to be more than three (δ in columns 2) and four ( δ in columns 4) times as large for spendings and income respectively. Taken together, the results in Table 8, Panel 2 thus strongly suggests that omitted variable bias is not driving our results. Table 4: LATE estimates of the Effects of CTA adoption on SSCBT’s household Income Outcome Monthly per capita spendings Monthly per capita income Full sample Main source Full sample Main Source of income income CTA adoption 0.221*** 0.217*** 0.307*** 0.328*** (0.084) (0.085) (0.104) (0.104) Observations 1084 1029 1089 1029 F Statistics (excluded instruments) 792.25*** 885.98*** 793.48*** 791.34*** Under identification test p value 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 6.90*** 6.51*** 8.64*** 9.84*** Cragg-Donald Wald F 358.89 351.10 360.73 352.13 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Socio-demographic Controls includes variable such as Head of Household, gender, Age, Education, Living in urban, Experience as SSCBT, Dependency ratio, whether the partner has a Job, the logarithm of purchased volume amount (a proxy of SSCBT’s capital) and the dummy for Province 16 Descriptive evidence in Table 1 shows that for 94.82% of our sample, the cross-border trade is the main source of household income. 17 These items include all food consumption; non-food consumption items were restricted to direct con- sumables (matches, soap, linen, and clothes), school and health expenditure, as well others contributions. 19 4.5 The impact on household poverty Given the positive and significant effect of the CTA adoption on the household outcomes (namely spending and income), the change in household poverty could be another spillover effect of the changes in trade outcomes. We evaluate it in this section by checking the extent to which CTA adoption affects the household poverty.18 As previously in the section 4.4, we exclude from the sample the single SSCBTs with dependency ratio equal to zero and we report the result in the Table 5. We assess the effect of CTA adoption on: the income of the poor (people falling below the poverty line) (column 2); the individual poverty gap (column 3); the income of SSCBTs classed within the poverty depth (column 4); the individual poverty severity (column 5) and the income of SSCBTs classed within the poverty severity (column 5).19 Table 5: LATE estimates of the Effects of CTA adoption on Poverty Status Poverty Depth of Poverty Svererity of the Poverty (2) (3) (4) (5) (6) Outcome Income Poverty income within Poverty income within of the poor depth poverty depth Severity poverty severity CTA Adoption 0.180* -0.027 0.163* -0.062* 0.219** (0.095) (0.017) (0.098) (0.025) (0.101) Observations 900 507 507 443 443 F Statistics (excluded instruments) 668.74 396.40 396.40 325.44 325.44 Under identification test p value 0.0000 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 3.58* 2.48 2.75* 6.04** 4.56** Cragg-Donald Wald F 307.52 200.10 200.10 168.51 168.51 Stock-Yogo : critical values 16.38 16.38 16.38 16.38 16.38 Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Socio-demographic Controls includes variable such as Head of House- hold, gender, Age, Education, Living in urban, Experience as SSCBT, Dependency ratio, whether the partner has a Job, the logarithm of purchased volume amount (a proxy of SSCBT’s capital) and the dummy for Province The result shows that CTA adoption is negatively correlated with the ratios of poverty depth (-0.027 but not significant) and severity of poverty (-0.062 and significant). Conse- quently, CTA adoption increases significantly; by 18% the income of SSCBTs classed as poor, by 16,3% the income of SSCBTs classed within the depth of poverty and by 21.9% the income of SSCBTs classed within the severity of the poverty. Therefore, although the CTA might exclude SSCBTs with meagre capital (Keyser et al., 2022; Murhi and Mesa, 2021), their level of capital does not hinder the increase in their income once they are participants. In that regard, CTA can be taken as a pro poor policy as it does not affect only the SSBTs initially endowed with significant income but also those classified as poor 18 Currently measured as people living on less than 1.90 a day. 19 Individual poverty depth and severity are obtained by the FGT poverty formulas to decompose poverty which are respectively; (z − yi )/z and ((z − yi )/z )2 , see Section 3.2 for more explanation. 20 or classed as failing within the poverty depth and severity. Thus the CTA can be used as a poverty-reducing strategy for SSCBTs. In any event, our result supports the prediction of Brenton et al. (2011) who state that, removing constraints and supporting cross-border traders should support poverty reduction. 4.6 Willingness to pay of SSCBT: Does gender matter ? The determinants of WTP are presented in Table 12. These estimates provide evidence on sociodemographic variables that significantly explain the WTP among SSCBTs who adopted the CTA. The first column considers only sociodemographic withous fixed effects. The second column control for the type of product fixed effect and the last columns add the Province fixed effect. The most conservative estimates in Column 3 restrict all variation to within-province and within-type of product sold. Taking it as the benchmark specification, we find that that among other variables, gender, female head of household, trade related training and the profit are the variable that significantly affect the amount of contribution one is willing to pay. More specifically, females are less willing to pay for the CTA survival compare to their counterpart male. This is explained by the fact that females are less endowed in terms of income and they often have to seek their partner’s permission for transactions involving money. They are therefore less able to contribute than men. However, when this variable is combined with a female being head of household, the results show that female traders who are responsible for their household are more likely to contribute than those who are not the head of household. This is probably because when they are in charge of the entire household, they have greater freedom in the choices they make (Kanyurhi et al., 2018). The result is in line with (Adnew and Abadi, 2011; Narang, 2012) who found that collective actions are effective approaches for women empowerment because they create opportunities and make them economically empowered within their communities. Access to training about trade related activities increases significantly the WTP among SSCBTs. This is due to raised awareness about fees, the ease of crossing borders and the advantage of bargaining power of groups relative to individuals. Access to information makes one more willing to contribute to the survival of clusters. As could be expected, profit increases the WTP for cluster survival. The more the profit one has, the more they are willing to contribute. This is line with Fischer and Qaim (2014)’s conclusion that benefits received through a group positively influences intensity of participation in the collective action, suggesting that reciprocity motives play a role. As seen in section 4.2, the CTA adoption increases the profit, the increase in profit increases the WTP. As a consequence, CTA adopters are more willing to pay for the CTA survival, making the future of cross-border trade in the GLR more likely to be conducted through clusters. 21 5 Robustness check In this section, we check for the robustness of our estimates on the relationship between CTA adoption and the trade related outcomes that have been established above. We first explore the plausibility of competing explanations of the relationships, then we analyse robustness when characteristics are balanced among CTA and non-CTA adopters and finally, we explore heterogeneity in the results with respect to various socio-demographic characteristics 5.1 Competing explanations One threat to identification of the CTA effect relates to attrition. This is due to the fact that the analytical sample includes only traders who persisted in their trading activities until May 2022, which is when we collected our data. Given the cross-section nature of our data, we are not able to measure the attrition rate but we know, from Mvunga and Kunaka (2021), that the daily average number of SSCTBs crossing the border was following a decreasing trend in GLR due to the COVID-19 pandemic. If so, the most vulnerable SSCBTs could be the first to exit the activity, which could imply plausible positive selection in our analytic sample (i.e., SSCBTs predicted to have relatively suf- ficient trade outcomes are more likely to be observed in the sample). We then rely on our qualitative approach combined with the empirical literature (Muuls, 2015; Valerievna et al., 2020; Wong, 2004), to highlight three facts that might have helped those SSCBTs who persisted in their trading activities during the crisis period and which could be at the basis of change in trade outcomes. First, the access to credit as within provinces in DR Congo, one can observe the proliferation of credit-based groups, in different provinces of the country, that offer credit to members. Many individuals with low income have joined them (Kanani et al., 2016) and this has likely increased their transactions (Muuls, 2015), resulting in increases in terms of trade outcomes . Furthermore, as described in Section 4.3, CTA adoption improves access to credit among SSCBTs. Second, according to Wong (2004), telephone traffic has a quantitatively larger effect on income per worker. We then consider access to technology of communication which is said to be the core of trade during the lockdown. Indeed, some traders from one side used to contact their partners of the other side and make trade operation bilaterally without being in the group. And this has been said to reduce drastically the cost and might increase the trade outcomes which in turn can affect income. Third and last, assistance with logistics can affect trade outcome as logistics cost and performance can be an important source of competitive advantages in trading enterprises (Valerievna et al., 2020). Thus, we consider assistance with logistics that some traders have received. Indeed, the GLTFP has distributed some 22 vehicles (tricycle) to help traders to carry their goods to and from markets across the borders. We checked whether our findings pick up the effect of such competing explanations by increasing the traders’ capital (with access to credit) and by drastically decreasing the time and transaction cost due to the availability of the communication technology and the logistic assistance. We do so by including, in a parsimonious way, the three variables namely access to credit, having an active telephone and having received the tricycle from the GLTFP in our benchmark specification (Column 5 of Table 3). The results are reported in Table 13 and show that our coefficient of interest, CTA adoption, remains significant for trade outcome increase. 5.2 Mahalanobis metric matching A second threat to identification of the CTA effect relates to potential reverse causality. This arises when SSCBTs predicted to have relatively sufficient trade outcomes are more likely to be in the treated group. Indeed, in their study on the borderland of the Great Lakes, Keyser et al. (2022) have shown that formal SSCBT (i.e wealthier SSCBTs) are more likely to join a cluster. Empirical literature has tried to cope with reverse causal- ity issue (Debbie et al., 2006; Derhard and Andreas, 2018; Lilah et al., 2021; S. et al., 2010) and among various analytic approaches proposed to solve it there is instrumental variable analyses (Collischon, 2021; Djankov et al., 2010; Lilah et al., 2021; Ranjan and Lee, 2007). That is why this paper used an instrumental variable approach (see Section 3). Another approach to deal with reverse causality is to work on a matched sample such as SSCBTs in the analytical sample are equal based on characteristics in the treated and the control group. That is what we did when using PSM and NNM approaches in the paper. In this subsection we go further in details and investigate whether our results are driven by differences in the composition of the samples between CTA adopters and CTA non-adopters and whether the threat of biased estimation due to these potential differ- ences can be mitigated. In line with comparisons between treated and control groups and as in Docquier et al. (2020), we use the Mahalanobis Metric Matching technique to iden- tify samples of CTA adopters and non-adopters that are balanced in terms of covariates, purchasing capacity included.20 We will then run LATE regressions, using a sample of individuals that are similar in terms of observed covariates (apart from their Adoption status) and potentially similar in terms of unobservable if the latter correlate with ob- served characteristics. The matching procedure minimizes the Mahalanobis metric. For 20 Indeed, the wealth of SSCBTs can be proxied by their purchasing capacity. If we remove the purchas- ing capacity and other variables based differences among the two groups, we can test the CTA adoption effect on individuals of the same profile in the treated and control group. 23 each covariate x, we compute the normalized difference: −1 2 2 Sx, CTA + Sx,NonCTA 2 ∆x = (xCTA − xNonCTA ) (5) 2 Where the difference between the mean value of the covariates for CTA adopters and non-CTA adopters, xCTA − xNonCTA , is divided by the mean of the standard deviations of the covariate over the whole sample (Sx,CTA + Sx,NonCTA ). The results of the matching technique are described in Table 14 and Figure 1 in the appendix. For each sample, we report the difference in terms of covariates before and after the matching procedure. Before matching, the distribution of covariates is unbalanced for both samples; differences in characteristics are always statistically different from zero, especially for purchasing capacity which proxied the income level of SSCBTs. By contrast, the matching technique allows us to generate a matched sample exhibiting a balanced distribution of covariates (i.e. where differences in characteristics are removed or are statistically not different from zero). Mainly, the difference in purchasing capacity between CTA and Non-CTA adopters is no longer different from zero. We then apply a fixed-effect LATE regressions using the matched samples with balanced characteristics. The results are provided in Table 15. As we can see, these results can be easily compared with those of Table 3 for the non-matched samples. The most conservative estimates in Column 5 of Table 15 restrict all variation to within-type of product sold, within borders and within-province. Taking it as the benchmark specification, we find that Trading in cluster leads to 21% increase in the daily turnover and 29% increase in the daily profit of SSCBT, confirming thus our benchmark results in section 4.2. All conclusions of the benchmark regressions hold when using the matched samples, implying that our benchmark results are less likely to be driven by reverse causality bias and that CTA adoption increases SSCBTs trade outcomes. 5.3 Heterogeneity We explore the heterogeneity in our results with respect to SSCBTs’ characteristics. Un- fortunately, working with the sub-sample reduces the size of the sample that leads to some singleton dummies which in turn might affect the estimates21 . To avoid this, the heterogeneity is explored without fixed effects dummies and only on the SSCBTs’ daily turnover. In Table 16, we explored this heterogeneity of the result with respect to SSCBTs’ (1) gender (Male vs Female) as females constitute the great share of SSCBTs according to Brenton et al. (2011) and the description in Table 1. (2) Education (Illiterate vs literate) 21 As a consequence of singleton dummy variable when present in the regression, the estimated covari- ance matrix of moment conditions are no longer of full rank, overidentification statistic are not reported, and thus standard errors and model tests should be taken with caution in that case. 24 to test whether two polarized skill groups benefit from the CTA. Indeed, the cross-border trade is one of the activities in which illiterate are not neither explicitly nor implicitly excluded. Moreover, education, combined with trade, can raise workers’ ability to adapt and move easily to industries with the greatest productivity (Kim and Kim, 2000). (3) Age (Youth22 vs Adults) as trade shocks affect workers differently because of their age (Artuc, 2012). Finally, we explored the heterogeneity with respect to Province (North Kivu vs South Kivu) as regions in DR Congo are reported to be heterogeneous (Amuli et al., 2022). The result reported in the Table 16 shows that of each of the eight columns, ours estimates survive to the heterogeneity of the sample as they remain positive and significant. This means that, both male and female traders who are either youth or adult, literate or not and leaving in North or South Kivu increase their trade outcomes with the CTA adoption. 6 Conclusion and implications This study examined the impact of the adoption of the CTA, a mitigation strategy based on the consequences of COVID-19, on trade outcomes household income and poverty reduction in two provinces of the DR Congo. Given the non-experimental nature of the data used in the analysis, associated with the biases and non-compliance behavior of some SSCBTs, a local average treatment effect model was used to account for the selection bias issue. Overall, the findings in this study indicate that adoption of CTA helped raise SSCBTs’ income and per capita expenditure, thereby increasing their probability of escaping poverty. This confirms the widely held view that dynamic-enhancing trade innovations can contribute to raising the SSCBTs’ household income and alleviate poverty in developing countries. Given this, the CTA is a poverty reducing strategy that can help policy makers to make headway in the efforts to achieve the first Sustainable Development Goal that is to end poverty in all its forms everywhere. The underlying mechanisms relate the CTA adoption to the development of economies of scale. Indeed, with CTA adoption, the transaction cost decreases meanwhile the volume of good traded increases. In the context of GLR, the aggregation of volumes however presents operational challenges especially with logistics and clearance at the border. In terms of logistics, whereas when acting individually, each trader can headload or use a small cart for their consignment, the clusters generate much larger volumes that can be moved only with motorized transport. There is also a need for staging areas, for consolidation or distribution of goods. The services and infrastructure could change the structure of the trading costs. At the border one of the effects is a tension between 22 In the paper, Youth refers to an SSCBT with a maximum of 36 years. 25 facilitation and border clearance processes and procedures. Consolidated consignments of clusters or associations can exceed the thresholds for duty- and tax-free treatment that small scale traders often benefit from, potentially excluding the very communities that are supposed to benefit. There is also a risk of free riding and capture of the new social structures, a phe- nomenon that is common in conflict zones. Some of the members might enjoy the ad- vantage of the CTA but escaping at the same time the costs (regular contribution) of maintaining the group. One interviewee expressed this clelary when they agrued that: ’Given the advantage of the CTA, some members order goods through the CTA mecha- nism in their names but for the benefit of others who are not members (and who don’t contribute)’. On the other hand, some elected group leaders engage the cluster in ac- tivities that benefit themselves personally, and not necessarily all the members. This was also captured in the interviews where one respondent argued that ’Our former group leader used his own money to order goods and sell them to group members at a higher price than the purchase price, and make a profit on us’.. That is why to prevent SSCBT to revert back to individual transactions, it would be ideal policy to consider first, the determination of the optimal number of members within clusters. Second, to prevent the elite capture phenomenon, Third, to enforce the law (as some traders try to sell wholesale in their country, then cross the border to sell in retail in the countries of others, which is prohibited by law because it causes harm to small national traders ). Finally, logistics matters for the sustainability of clusters, mainly in terms of transport and information and communications technology as these are basic ingredients for success. A mechanism that provides means of transport or connects transporters with SSCBTs would be prefer- able. This is because, on the side of transporters, the consolidation of small quantities of cargo might be seen as not particularly profitable and might endanger the long-term perspective of the CTA initiative. The results of the research suggest that intervention programs to help extend the CTA can produce important development outcomes in areas with high poverty rates. The approach is a worthwhile policy instrument to raise incomes in poor areas, although complementary measures such as effective extension services, support to establish clusters, an effective rule of law and appropriate economic incentives are needed to enhance the utility of trade for the sustenance of livelihoods in the GLR . 26 References Abadie, A. (2003). Semi-parametric instrumental variable estimation of treatment re- sponse models. Journal of Econometrics, 113(2):231–263. Adnew, B. and Abadi, Z. (2011). Researching women’s collective action. Ethiopia Report. GB: Oxfam. Altonji, J. G., Elder, Todd, E., and Taber, Christopher, R. (2005). Selection on observed and unobserved variables: Assessing the effectiveness of catholic schools. Journal of Political Economy, 113(1):51–184. Amuli, Ibale, D. (2023). Earning structure and heterogeneity of the labor market; evidence from democratic republic of congo. Journal of African Economies, pages 1–22. Amuli, Ibale, D., Docquier, F., and Iftikhar, Z. (2022). Spatial inequality, poverty and informality in the democratic republic of the congo. CEPR Discussion Paper No. DP, 17195. Artuc, E. (2012). Workers’ age and the impact of trade shocks. World Bank Policy Research Working Paper, 6035. Bacon, C. (2005). Confronting the coffee crisis: can fair trade, organic, and specialty cof- fees reduce small-scale farmer vulnerability in northern nicaragua? World Development, 33(3):497–511. Barham, J. and Chitemi, C. (2009). Collective action initiatives to improve marketing performance: Lessons from farmer groups in tanzania. Food Policy, 34:53–59. Baum, C., Schaffer, M., and Stillman, S. (2003). Instrumental variables and gmm: esti- mation and testing. Stata Journal, 3:1–31. Benett, R. (1998). Business associations and their potential to contribute to economic development: reexploring an interface between the state and market. Environment and Planning, 30:1367–1387. Bernard, T. and Spielman, D. (2009). Reaching the rural poor through rural producer organizations? a study of agricultural marketing cooperatives in ethiopia. Food Policy, 34:60–69. Bouet, A., Kurtz, J., and Traor´e, F. (2021). Covid-19 impact on informal trade: Dis- ruptions to livelihoods and food security in africa. IFPRI Trade Brief, Washington DC. Bound, J., Jaeger, D., and Baker, R. (1995). Problems with instrumental variables esti- mation when the correlation between the instruments and the endogenous explanatory variable is weak. Journal of the American Statistical Association, 90:443–450. Brenton, P., Bucekuderhwa, B, C., Hossein, C., Nagaki, S., and Ntagoma, J. B. (2011). Earning structure and heterogeneity of the labor market evidence from dr congo. Africa Trade Policy Notes, 60112(11). 27 Che, T., Peng, Z., Lim, K. H., and Hua, Z. (2015). Antecedents of consumers’ intention to revisit an online group-buying website: A transaction cost perspective. Information & Management, 52(5):588–598. Collischon, M. (2021). Methods to estimate causal effects-an overview on iv, did and rdd and a guide on how to apply them in practice. Institut fur Arbeitsmarkt-und Berufsforschung (IAB), Research Department PASS. Dafuleya, G. (2020). Social and emergence assistance ex-ante and during covid-19 in the sadc region. The International Journal of Community and Social Development, pages 251–268. Debbie, A., Carole, L., David, J., and George, D. (2006). Reverse causality and confounding and theassociations of overweight and obesity withmortality. Obesity, 14(12):2294–2304. Derhard, K. and Andreas, E. (2018). What explains the negative effect of unemploy- ment on health? an analysis accounting for reverse causality. American Journal of Epidemiology, 55:25–39. Diagne, A., Adekambi, S., Simtowe, F. P., and Biaou, G. (2009). The impact of agri- cultural technology adoption on poverty: the case of nerica rice varieties in benin : Paper contributed to the 27th conference of the international association of agricul- tural economists. August 16-22, Beijing, China. Djankov, S., Freund, C., and Pham, C. (2010). Trading on time. The review of Economics and Statistics, 92(1):166–173. Docquier, F., Aysit, T., and Turati, R. (2020). Do emigrants self-select along cultural traits? evidence from the mena countries. International Migration Review, 54(2):388– 422. Dontsop, P., Diagne, A., Olusegun, V., and Ojehomon, V. (2011). Impact of improved rice technology (nerica varieties) on income and poverty among rice farming households in nigeria: A local average treatment effect (late) approach. Quarterly Journal of International Agriculture, 50(3):267–291. Durbin, J. (1954). Errors in variables. Review of the International Statistical Institute, 22:23–32. Fischer, E. and Qaim, M. (2014). Action: What determines the intensity of participation? Journal of Agricultural Economics, 65(3):683–702. Gelo, D., Muchapondwa, E., Shimeles, A., and Dikgang, J. (2019). Welfare effect and elite capture in agricultural cooperatives intervention: Evidence from ethiopian villages. IZA Discussion Papers, No. 12495, Institute of Labor Economics (IZA), Bonn. Gordon, V., Graffon, R. Q., and Steinshamn, S. (2021). Cross-country effects and policy responses to covid-19 in 2020: The nordic countries. Economic Analysis and Policy, 71:198–210. 28 Guerrero-Amezage, M.E.and Humphries, J. E., Neilson, C., Shimberg, N., and Ulyssea, G. (2022). Small firms and the pandemic: Evidence from latin america. Journal of Development Economics, 155. Hale, T., Angrist, N., Goldszmidt, R., Kira, B., Petherick, A., Phillips, T., Webster, S., Cameron-Blake, E., Hallas, L., and Tatlow, S. (2021). A global panel database of pandemic policies. Nature Human Behaviour, 5:529–538. Hansen, L. P. (1982). Large sample properties of generalized method of moments estima- tors. Econometrica, 50:1029–54. Hausman, J. (1978). Specification tests in econometrics. Econometrica, 46:262–280. Heckman, J. and Vytlacil, E. (1999). Local instrumental variables and latent variable mod- els for identifying an bounding treatment effects. Proceedings of the National Academy of Sciences, 96(April):4730–4734. Heckman, J. and Vytlacil, E. (2005). Structural equations, treatment effects, and econo- metric policy evaluation. Econometrica, 73(May):669–738. Hellin, J., Lundy, M., and Meijer, M. (2009). Farmer organization, collective action and market access in meso-america. Food Policy, 34:16–22. Imbens, G. (2004). Nonparametric estimation of average treatment effects under exogene- ity: A review. Review of Economics and Statistics, 86(1):4–29. Imbens, G. and Angrist, J. (1994). Identication and estimation of local average treatment effects. Econometrica, 62(2):467–476. Imbens, G. and Wooldridge, J. (2009). Recent developments in the econometrics of pro- gram evaluation. Journal of Economic Literature, 47(1):5–86. Jin-Long, L. and Zhang Yi, S. (2012). Dealing with preference uncertainty in contingent willingness to pay for a nature protection program: A new approach. Transportation Research, 17:124–128. Kanani, A., Kimbu, T., and Wa Bolinda, B. (2016). Les cooperatives d’epargne et de credit: Aubaine pour l’entrepreprise et l’emploi dans la ville de bukavu en rd congo. European Scientific Journal, 12:495–510. Kansiime, M., Tambo, A., Mugambi, I., Bundi, M., Kara, A., and Owuor, C. (2021). Covid-19 implications on household income and food security in kenya and uganda: Findings from a rapid assessment. World Development, 137. Kanyurhi, E. B., Bucekuderhwa, C., B., and Kadurha, H., L. (2018). Les d´ eterminants etude empirique sur les entrepreneurs et investisseurs du crowdfunding hors ligne: une ´ a bukavu. Innovations, 2:187–215. potentiels ` Karim, M., Islam, M. T., and B., T. (2021). Covid-19’s impacts on migrant workers from bangladesh: In search of policy intervention. World Development, 136. Kersting, S. and Wollni, M. (2012). New institutional arrangements and standard adop- tion: Evidence from small-scale fruit and vegetable farmers in thailand. Food Policy, 37:452–462. 29 Keyser, J., Kunaka, C., and Walkenhorstc, P. (2022). Health crisis, mobility restrictions, and group trade: evidence from small-scale cross-border transactions in the great lakes region. Working paper. Kim, S. J. and Kim, Y. J. (2000). Growth gains from trade and education. Journal of International Economics, 50(2):519–545. Laborde, D., Martin, W., and Vos, R. (2020). Impacts of covid-19 on global poverty, food security, and diets: Insights from global model scenario analysis. Agricultural economics, 52(3):375–390. Lee, M.-J. (2005). Micro-econometrics for policy, program and treatment effects. advanced texts in econometrics. Oxford University Press. Lilah, M., Willa, D., Oanh, L., Serena, H., and John, R. (2021). Methods to address self-selection and reverse causation in studies of neighborhood environments and brain health. International Journal of Environmental Research and Public Health, 18. Louinord, V., Claudio, P., and Denis, B. (2013). Dealing with preference uncertainty in contingent willingness to pay for a nature protection program: A new approach. Ecological Economics, 88:76–85. Madkour, A., Mohamed, S., and Dabees, A. (2020). The impact of logistics performance index on trade openness in africa. International journal of management and applied science, pages 2394–7926. Manski, C. and Pepper, J. (2000). Monotone instrumental variables: with an application to the returns to schooling. Econometrica, 68(4):997–1010. Markelova, H., Meinzen-Dick, R., Hellin, J., and Dohrn, S. (2009). Collective action for smallholder market access. Food Policy, 34:1–7. Marti, L., Puertas, R., and Garcia, L. (2014). The importance of the logistics performance index in international trade. Applied economics, 46(24):2982–2992. Martinez-Zarzoso, I., Perez-Garcia, Eva, M., and Suarez-Burguet, C. (2008). Do trans- port costs have a differential effect on trade at the sectoral level. Applied Economics, 40(24):3145–3157. Miao, Q., Schwarz, S., and Schwarz, G. (2021). Responding to covid-19: Community volunteerism and coproduction in china. World Development, 137. Murhi, M. and Mesa, H. (2021). Subverting borders, precarity and vulnerability : The socio-economic impact of covid-19 on informal cross-border traders between rwanda and rdc. 8th EPRN Annual Research Conference. Muuls, M. (2015). Exporters, importers and credit constraints. Journal of International Economics, 95(2):333–343. Mvunga, N. and Kunaka, C. (2021). Eight emerging effects of the covid-19 pandemic on small-scale cross-border trade in the great lakes region. World Bank, Trade, Investment and Competitiveness. 30 Narang, U. (2012). Self help group: An effective approach to women empowerment in india. International Journal of Social Science and Interdisciplinary Research, 1(8). Narrod, C., Roy, D., Okello, J., Avendano, B., Rich, K., and Thorat, A. (2009). Public- private partnerships and collective action in high value fruit and vegetable supply chains. Food Policy, 34:8–15. Njoku, O. A., Kagiso, T., M., Francis, N., O., and Helen, A., A. (2013). Profitability of the informal cross-border trade: A case study of four selected borders of botswana. Applied Economics, 7(40):4221–4232. Oster, E. (2019). Unobservable selection and coefficient stability: Theory and evidence. Journal of Business & Economic Statistics, 37(2):187–204. Park, J. and Chung, E. (2021). Learning from past pandemic governance: Early response and public-private partnerships in testing of covid-19 in south korea. World Develop- ment, 137. Poulton, C., Dorward, A., and Kydd, J. (2010). The future of small farms: new directions for services, institutions and intermediation. World development, 38(10):1413–1428. Rambaree, K. and Nassen, N. (2020). The swedish strategy’ to covid-19 pandemic: Impact on vulnerable and marginalised communities. The International Journal of Community and Social Development, 2(2):234–250. Ramsey, A. F., Goodwin, B. K., Hahn, W. F., and Holt, M. (2021). Impacts of covid-19 and price transmission in u.s. meat markets. Agricultural economics, 52(3):441–458. Ranjan, P. and Lee, J. (2007). Contract enforcement and international trade. Economics and Politics, 19(2):191–218. Riccioli, F., Marone, E., Boncinelli, F., Tattoni, C., Rocchini, D., and Fratini, R. (2019). The recreational value of forests under different management systems. New Forests, 50(2):345–360. Rosenbaum, P. and Rubin, D. (1983). The central role of the propensity score in obser- vational studies for causal effects. Biometrika, 70(1):41–55. Rubin, D. (1974). Estimating causal effects of treatments in randomized and non- randomized studies. Journal of Educational Psychology, 66(5):688–701. S., M., Erica, E., Mourad, D., and W., R. (2010). Breastfeeding and infant size: Evidence of reverse causality. American Journal of Epidemiology, 173(9):978–983. Sakyi, D., Villaverde, J., Maza, A., and I., B. (2017). The effects of trade and trade facilitation on economic growth in africa. African Development Review, 29(2):350–361. Sargan, D. J. (1958). The estimation of economic relationships using instrumental vari- ables. Econometrica, 26:393–415. Schmidt, E.and Dorosh, P. and Gilbert, R. (2021). Impacts of covid-19 induced income and rice price shocks on household welfare in papua new guinea: Household model estimates. Agricultural economics, 52(3):391–406. 31 Sengupta, S. and Jha, M. (2020). Social policy, covid-19 and impoverished migrants: Chal- lenges and prospects in locked down india. The International Journal of Community and Social Development, 49(2). Staiger, D. and Stock, J. H. (1997). Instrumental variables regression with weak instru- ments. Econometrica, 65:557–586. STATA (2017). Stata functions reference manuel release 17. A Stata Press Publication, Stata Corp LLC, College Station, Texas. Swinnen, J. and R., V. (2021). Covid-19 and impacts on global food systems and household welfare: Introduction to a special issue. Agricultural economics, 52(3):365–374. Valerievna, Shtal, T., Ievgenievna, Uvarova, A., Viktorivna, Proskurnina, N., and Leoni- divna, Savytska, N. (2020). Strategic guidelines for the improvement of logistic activities of trade enterprises. Journal of International Economics, 12(3):69–81. Varangis, P., Siegel, P., Giovannucci, D., and Lewin, B. (2003). Dealing with the coffee crisis in central america: impacts and strategies. The World Bank, Policy Research Working Paper No. 2993. Weinberger, K. (2021). What role does bargaining power play in participation of women? a case study of rural pakistan. Journal of Entrepreneurship, 10(3):209–221. Wollni, M. and Zeller, M. (2007). Do farmers benefit from participating in specialty markets and cooperatives? the case of coffee marketing in costa rica. Agricultural Economics, 37:243–248. Wong, W.-K. (2004). How good are trade and telephone call traffic in bridging income gaps and tfp gaps. Journal of International Economics, 64(2):441–463. Wu, D. M. (1973). Alternative tests of independence between stochastic regressors and disturbances. Econometrica, 42:529–546. Xiao, L., Guo, Z., and D’Ambra, J. (2017). Analyzing consumer goal structure in online group buying: A means-end chain approach. Information & Management, 54(8):1097– 1119. Zaninovic, P. A., Zaninovic, V., and Skender, H. P. (2021). The effects of logistics perfor- mance on international trade: Eu15 vs cems. EconomicResearchEkonomskaIstrazivanja, 34(1):1566–1582. Zweig, S., Zapf, A., Beyrer, C., Guha-Sapir, D., and Jaar, R. (2021). Ensuring rights while protecting health: The importance of using a human rights approach in implementing public health responses to covid-19. Health and Human Rights Journal, 23(2). 32 7 Appendix 33 7.1 PSM and NNM estimate of the Effects of CTA adoption on SSCBT’s outcomes Table 6: PSM and NNM estimate of the Effects of CTA adoption on SSCBT’s outcomes PSM NNM ATE ATT ATE ATT Outcome Daily Turnover CTA adoption 0.144** 0.164** 0.349*** 0.313*** (0.063) (0.077) (0.080) (0.077) Outcome Daily Profit CTA adoption 0.110* 0.135* 0.341*** 0.236*** (0.061) (0.074) (0.061) (0.071) √ √ √ √ Socio-demographic Controls √ √ √ √ Borders fixed effects √ √ √ √ Type of Product fixed effects √ √ √ √ Province fixed effects Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Socio- demographic Controls includes variable such as gender, Age, Education, Living in urban, Marital status, Experience as SSCBT, whether the SSCBT has a fixed place in the market and the logarithm of purchased volume amount. 34 7.2 PSM and NNM estimates of the Effects of CTA adoption on SSCBT’s household income Table 7: PSM and NNM estimates of the Effects of CTA adoption on SSCBT’s household income PSM NNM ATE ATT ATE ATT Outcome Per-capita income CTA adoption 0.113** 0.113* 0.124*** 0.100** (0.049) (0.059) (0.044) (0.048) Panel 1: Observation 1084 1084 1084 1084 Full Outcome Per-capita income Sample CTA adoption 0.142** 0.239*** 0.146** 0.243*** (0.063) (0.075) (0.060) (0.072) Observation 1089 1089 1089 1089 Outcome Per-capita spendings PSM NNM ATE ATT ATE ATT CTA adoption 0.144*** 0.111** 0.113** 0.096* Panel 2: (0.047) (0.055) (0.045) (0.049) Trade as Observation 1029 1029 1029 1029 Main Source of Income Outcome Per-capita income CTA adoption 0.041 0.121 0.144** 0.247*** (0.065) (0.077) (0.063) (0.074) Observation 1033 1033 1033 1033 √ √ √ √ Socio-demographic Controls √ √ √ √ Borders fixed effects √ √ √ √ Type of Product fixed effects √ √ √ √ Province fixed effects Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% lev- els, respectively. Standard errors are in parenthesis. Socio-demographic Controls includes variable such as Head of Household, gender, Age, Education, Living in urban, Experience as SSCBT, Dependency ratio, whether the partner has a Job, the logarithm of purchased volume amount (as a proxy of SSCBT’s capital) and the dummy for Province 35 7.3 Judging the importance of selection on unobservables Table 8: Judging the importance of selection on unobservables Daily Turnover Daily Profit No Controls With controls No Controls With controls Panel 1: (1) (2) (3) (4) CTA Adoption 0.619*** 0.111** 0.441*** 0.112** (0.068) (0.052) (0.056) (0.047) Observation 1159 1145 1159 1145 R2 0.0681 0.5636 0.052 0.436 δ -3.77 -5.68 Notes: OLS estimates based on the trade outcomes. Columns 1 and 3 report estimates from regressions of trade outcomes, respectively, on the treatment dummy (CTA Adoption) without further controls. Columns 2 and 4 add controls as in Table 3, column 5. The final row shows the amount of selection on unobservables necessary, relative to the amount of selection on observable controls, to explain away the coefficient in the respective column. For the calculation of this δ , we use the Stata command −psacalc−, setting Rmax to 1.3 times the R2 in the respective column; for details, see text and Oster (2019). ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Household spendings Household Income No Controls With controls No Controls With controls Panel 2: (1) (2) (3) (4) CTA Adoption 0.232*** 0.106** 0.327*** 0.111** (0.044) (0.041) (0.062) (0.054) Observation 1041 1029 1045 1033 R2 0.0254 0.1832 0.0265 0.2979 δ -3.73 -4.74 Notes: OLS estimates based on the household outcomes. Columns 1 and 3 report estimates from regressions of household outcomes, respectively, on the treatment dummy (CTA Adoption) without further controls. Columns 2 and 4 add controls as in Table 4. The final row shows the amount of selection on unobservables necessary, relative to the amount of selection on observable controls, to explain away the coefficient in the respective column. For the calculation of this δ , we use the Stata command −psacalc−, setting Rmax to 1.3 times the R2 in the respective column; for details, see text and Oster (2019). ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. 36 7.4 Household outcomes description Table 9: Household outcomes description Variable Description Household Outcome: Monthly Expenditures Total Expenditure : Sum of household spending on : Food Consumption Dressing Health Education Saving Credit interest payment Other Consumption Household Outcome: Monthly Income Total Income : Sum of household income from : Trade activities Sale of livestock Partner job Rental income (house, lands) Transfers received from third parties Other non-commercial activities Notes: Variable were captured on a monthly basis and in monetary unit 37 7.5 Cross-Tabulation of Main variables Table 10: Cross-Tabulation of the mains variables Awareness vs CTA Adoption CTA adopters Awareness Total Not Adopters Adopters 304 0 304 Not Aware (47.72) (0.00) (26.23) 333 522 855 Aware (52.28) (100.00) (73.77) 637 522 1159 Total (100.00) (100.00) (100.00) Notes: The Table 10 reports numbers and their relative frequency in parenthesis 38 7.6 Analysis of the Mechanisms Table 11: OLS & Logit evaluation of the Mechanisms OLS estimates Logit estimates Panel 1: Dependent variable Perceived Level of transport cost CTA adoption -0.279*** -0.172*** (0.063) (0.027) Observation 1,145 1,133 P rob > F 0.0000 0.0000 Panel 2: Dependent variable Perceived Level of cross border transaction time CTA adoption -0.151*** -0.102*** (0.056) (0.030) Observation 1,145 1,145 P rob > F 0.0000 0.0000 Panel 3: Dependent variable Perceived level of volume of goods per transaction CTA adoption 0.309*** 0.128*** (0.058) (0.030) Observation 1,145 1,145 P rob > F 0.0000 0.0000 Panel 4: Dependent variable Access to credit (Yes=1) CTA adoption − 0.061*** − (0.019) Observation − 1103 P rob > F − 0.0000 √ √ Socio-demographic Controls √ √ Borders fixed effects √ √ Type of Product fixed effects √ √ Province fixed effects Notes: OLS & Logit estimates based on the perceived level of transport cost, border transaction duration and volume of good per transaction. The three variables are collected on a Likert scale of five points (of (1): Very less, (2): less (3): neutral (4): high and (5): very high). For the Logit regression we have transformed these variable into a binary variable with points 3 and above equal to one and points less than 3 equal to zero. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Each model includes controls as the benchmark specification of Table 3, Column 5. The logit regression reports marginal effect 39 7.7 Interval regression estimates of of SSCBT WTP for Cluster sustainability Table 12: Interval regression estimates of of SSCBT WTP for Cluster survival Variables (1) (2) (3) Female -0.195 -0.356** -0.352*** (0.154) (0.162) (0.162) Female Head of household 0.277* 0.321** 0.321** (0.147) (0.148) (0.148) Age 0.011* 0.010 0.010 (0.006) (0.007) (0.007) Education Primary -0.228 -0.229 -0.234 (0.184) (0.186) (0.186) Secondary -0.608*** -0.565*** -0.572*** (0.180) (0.183) (0.183) Tertiary -0.052 -0.014 -0.118 (0.394) (0.401) (0.401) Trade related training 0.311*** 0.272** 0.273** (0.111) (0.113) (0.113) Civil status -0.069 -0,078 -0.078 (0.093) (0.092) (0.092) Living in urban -0.444 -0.447 -0.470 (0.455) (0.458) (0.462) Dependancy ratio -0.082 -0.002 -0.020 (0.210) (0.212) (0.215) Experience -0.020* -0.014 -0.014 (0.008) (0.008) (0.008) Weither the partner has a job 0.175 0.205 0.210* (0.114) (0.114) (0.114) Satisfaction with CTA 0.246 0.199 0.210 (0.273) (0.270) (0.272) Daily profit 0.071 0.101* 0.106* (0.055) (0.056) (0.057) Constant 8.097*** 7.68*** 7.798*** (1.090) (1.110) (1.135) √ √ Type of Product Fixed effects - √ Province fixed effects - - Observation 476 476 476 LR chi2 44.25*** 65.50*** 65.73*** Notes: The dependent variable is an interval type which represents the amount in CDF. (0-999; 1000-1999; 2000-2999; 3000-3999; 4000-4999; 5000-). ***,**,* Denote significance at 1%, 5% and 10% levels, respec- tively. Standard errors are in parenthesis. illiterate is the reference category for education, Dependency ration is the number of dependent person out of the household size. 40 7.8 Competing explanation of the changes in trade outcomes Table 13: Competing explanation of the changes in trade outcomes Outcome Daily Turnover CTA adoption 0.206** 0.214** 0.201* 0.204* (0.102) (0.010) (0.109) (0.114) Access to credit 0.097 0.095 (0.075) (0.079) Technology of communication -0.008 -0.017 0.055) (0.056) Logistic Assistance 0.026 0.019 (0.061) (0.061) Observations 1145 1145 1145 1145 F Statistics (excluded instruments) 625.33*** 567.55*** 545.04*** 490.85*** Under identification test p value 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 3.91** 3.93*** 3.29* 3.10* Cragg-Donald Wald F 344.16 323.98 318.37 298.41 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) Outcome Daily Profit CTA adoption 0.317*** 0.351*** 0.329*** 0.367*** (0.094) (0.097) (0.100) (0.103) Access to credit -0.072 -0.033 (0.072) -0.121* Technology of communication 0.755 -0.121* (0.097) (0.050) Logistic Assistance -0.040 -0.039 (0.042) 0.044 Observations 1145 1145 1145 1145 F Statistics (excluded instruments) 625.33*** 567.55*** 545.04*** 631.05 Under identification test p value 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 11.15*** 12.89*** 10.67*** 11.05*** Cragg-Donald Wald F 344.16 323.98 318.37 347.15 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) √ √ √ √ Socio-demographic Controls √ √ √ √ Borders fixed effects √ √ √ √ Type of Product fixed √ √ √ √ Province fixed effects Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Socio-demographic Controls includes variable such as gender, Age, Education, Living in urban, Marital status, Experience as SSCBT, whether the SSCBT has a fix place in the market and the logarithm of purchased volume amount. 41 7.9 Difference in terms of covariates before and after the Ma- halanobis matching procedure Table 14: Difference in terms of covariates before and after the Mahalanobis matching procedure Variables Mean %bias %bias P value Treated Control reduction Female Unmatched 1.158 1.150 2.2 0.715 Matched 1.158 1.158 0.0 100.0 1.000 Age Unmatched 37.751 35.508 22.7 0.000 Matched 37.751 37.046 7.1 68.6 0.228 marital status Unmatched 2.098 1.987 15.4 0.010 Matched 2.098 2.028 9.6 37.5 0.120 Education level Unmatched 2.443 2.349 12.4 0.037 Matched 2.4432 2.4277 2.1 83.5 0.733 Urban Unmatched 1.986 1.964 14.1 0.020 Matched 1.986 1.986 0.0 100.0 1.000 Experience Unmatched 9.807 8.432 20.2 0.001 Matched 9.807 9.258 8.1 60.0 0.191 Volume of Good Unmatched 11.405 10.84 51.4 0.000 Matched 11.405 11.328 7.1 86.2 0.243 Fixed place Unmatched 2.963 2.845 28.8 0.000 Matched 2.963 2.965 -0.5 98.4 0.903 Notes: Column of P value reports the statistical significance of the difference between CTA adopters and CTA non adopters for each Covariate before and after the matching . procedure 42 Figure 1: Fit of covariates before and after the matching procedure Volume_purchassed Fixed_place_in_market Age Experience Matrimonial_Status Urban Education_level Female Unmatched Matched 0 10 20 30 40 50 Standardized % bias across covariates 43 7.10 LATE estimates of the Effects of CTA adoption on SS- CBT’s trade outcomes using Matched sample Table 15: LATE estimates of the Effects of CTA adoption on SSCBT’s trade outcomes using Matched sample (1) (2) (3) (4) (5) Outcome Daily Turnover CTA adoption 0.671*** 0.211** 0.216** 0.212** 0.212** (0.136) (0.088) (0.088) (0.092) (0.092) Observations 811 811 811 811 811 F Statistics (excluded instruments) 2251.25*** 1816.35*** 1744.38*** 881.97*** 881.97*** Under identification test p value 0.0000 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 27.46*** 12.41*** 13.03*** 10.17*** 10.17*** Cragg-Donald Wald F 457.60 436.19 440.00 375.00 375.00 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) (16.38) Outcome Daily Profit CTA adoption 0.578*** 0.301*** 0.308*** 0.294*** 0.294*** (0.110) (0.085) (0.085) (0.091) (0.091) Observations 811 811 811 811 811 F Statistics (excluded instruments) 1337.96*** 885.09*** 879.89*** 629.14*** 631.05*** Under identification test p value 0.0000 0.0000 0.0000 0.0000 0.0000 Wald test (joint significance) 59.43*** 16.87*** 17.02*** 10.83*** 11.05*** Cragg-Donald Wald F 475.7 392.85 400.19 347.12 347.15 Stock-Yogo : critical values (16.38) (16.38) (16.38) (16.38) (16.38) √ √ √ √ Socio-demographic Controls √ √ √ Borders fixed effects √ √ Type of Product fixed effects √ Province fixed effects Notes: outputs are log-transformed. ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Socio-demographic Controls includes variable such as gender, Age, Education, Living in urban, Marital status, Experience as SSCBT, whether the SSCBT has a fix place in the market and the logarithm of purchased volume amount. 44 7.11 Heterogeneity of the sample Table 16: Heterogeneity of the sample Outcome Daily Turnover Gender Education Female Male Illiterate Literate CTA adoption 0.245*** 0.344* 0.584** 0.246** (0.098) √ (0.192) √ (0.253) √ (0.098) √ Socio-demographic Controls Observations 969 176 147 998 F Statistics (excluded instruments) 732.61 138.38 98.13 720.07 Under identification test p value 0.000 0.000 0.000 0.000 Wald test (joint significance) 6.19** 3.11* 4.84** 6.19** Cragg-Donald Wald F 320.49 73.83 53.84 322.60 Stock-Yogo : critical values 16.38 16.38 16.38 16.38 Age Province Youth Adult North Kivu South-Kivu CTA adoption 0.352** 0.200* 0.283** 0.270* (0.141) √ (0.115) √ (0.131) √ (0.144) √ Socio-demographic Controls Observations 548 597 510 635 F Statistics (excluded instruments) 331.84 580.78 350.88 330.97 Under identification test p value 0.000 0.000 0.000 0.000 Wald test (joint significance) 6.18** 2.93* 4.54** 3.41* Cragg-Donald Wald F 159.75 237.34 159.67 182.61 Stock-Yogo : critical values 16.38 16.38 16.38 16.38 Notes: ***,**,* Denote significance at 1%, 5% and 10% levels, respectively. Standard errors are in parenthesis. Each model includes Socio-demographic Controls as the . benchmark specification of Table 3, Column 5. 45