Business, Brokers and Borders: The Structure of West African Trade Networks

Abstract

The objective of this paper is to show how a formal approach to networks can make a significant contribution to the study of cross-border trade in West Africa. Building on the formal tools and theories developed by social network analysis, we examine the network organisation of 136 large traders in two border regions between Niger, Nigeria and Benin. In a business environment where transaction costs are extremely high, we find that decentralised networks are well adapted to the various uncertainties induced by long-distance trade. We also find that long-distance trade relies both on the trust and cooperation shared among local traders, and on the distant ties developed with foreign partners from a different origin, religion or culture. Studying the spatial structure of trade networks, we find that in those markets where trade is recent and where most of the traders are not native of the region, national borders are likely to exert a greater influence than in those regions where trade has pre-colonial roots. Combining formal network analysis and ethnographic studies, we argue, can make a significant contribution to the current revival of interest in cross-border trade in the policy field.

1. Introduction

Chief Ibrahim Hassan looked out at the new developments being built around his compound and talked about the days when the city of Gaya was little more than a village along the Niger River. ‘There were no houses or warehouses around here,’ he said, pointing at a businessman dressed in a long white robe supervising the loading of a Nigerian truck. ‘I have waited for seven years until the traditional authorities of old Gaya appointed me as District Chief. Nobody lived here on theses wild hillsides.’ That was 30 years ago. A little more than 8,000 people lived in this Nigerien border town facing Benin and Nigeria, most of them engaged in agriculture and fishing. They are now more than 45,000. Every day, hundreds of trucks loaded with food stuff, manufactured products, oil and uranium concentrates find their way along the crowded one-lane road that crosses Gaya, en route to the Gulf of Guinea or to the major cities of the Sahel. The Cotonou corridor, where Gaya is located, accounts for half of the imported goods from Niger and three-quarters of total Nigerien exports in terms of value. In three decades, dozens of warehouses have been built by wholesalers attracted by the strategic location of the city, at the crossroad between three states. The city now stretches out over a length of four kilometres and the urban area – originally concentrated around old Gaya – has been extended more than 10 times.

The spectacular economic development of Gaya – among many other border cities in West Africa – would certainly not have been possible without the existence of national borders. Without the price differences across countries and the bans affecting the imports or exports of certain agricultural or manufactured products, West African traders would have fewer incentives to move their goods across the continent and avoid the taxes collected by customs authorities. Considering that the cost of moving goods in Africa is the highest in the world (Teravaninthorn & Raballand, 2009), second-hand clothes, for example, would probably not be imported through Benin and then stored in a warehouse in Gaya before being re-exported illegally to Nigeria, where they are subject to an import ban – a journey of about 1,800 kilometres across three countries (Brooks & Simon, 2012). However, the existence of national borders is not a sufficient condition for trade to flourish. Not all border cities are important market places. Some of them lack market and road infrastructure or are predominantly dominated by administrative or political rather than business activities. Many border cities also lack one of the most important ingredients of trade: entrepreneurs willing to develop transnational networks.

The organisation and evolution of these networks have been interpreted in conflicting ways. In most of the African studies literature, social networks are seen as a heuristic device or as a metaphor for informal economic relations. Qualitative approaches, it is argued, are best suited to study the ethnographic dimensions of social networks, that is, how their organisational and institutional dimensions result from historical, cultural or geographical factors. Social networks are rarely – if ever – mapped, formally described or modelled. In most of the network science literature, by contrast, new conceptual and computational developments have been used to represent, describe, and model social networks in an increasingly sophisticated way. However, and despite Mitchell’s (1969) advocacy for an analytical rather than metaphorical approach to networks, most formal social network analysis has thus far largely ignored Africa. The objective of this article is to fill this gap by showing how a formal approach to networks can make a significant contribution to the study of cross-border trade and to the current revival of interest in cross-border trade in the policy field. Building on the formal tools and theories developed by social network analysis (SNA) that allows researchers to map the nodes and ties that compose a network, we examine the network organisation of traders in two border market complexes on the Nigeria–Niger–Benin border and on the Nigeria–Niger border.

The first part of the article aims to understand the centralisation of cross-border trade networks. We wish to know whether traders are organised in a hierarchical structure in which a few central actors transmit information, orders and resources, or in a rather decentralised way in which every actor is connected to everyone else. This question is crucial for the efficiency of trade networks, as centralised networks are generally regarded as more efficient, but less resilient to threats, than decentralised structures. Our two case studies show that trade networks tend to be rather decentralised, with few ties between the actors and few highly connected actors. We also study the trade-offs faced by traders between embeddedness and brokerage for the purpose of determining whether traders are strongly embedded within their group of peers or if they play the role of brokers beyond their own business community. Our work shows that combining embeddedness and brokerage is a determinant factor of individual economic performance because each of the two dimensions contributes to the accumulation of social capital among and between groups.

The second part of the article considers the spatial structure of trade networks. We are particularly interested in analysing the influence of national borders on the development of social ties. We study whether traders are organised with a multiplicity of cross-border ties or according to several nationally organised markets with only a few brokers able to bridge the markets. Our aim here is to determine how the social structure of trade networks has adapted to the constraints and opportunities offered by the existence of national borders. In contrast to what may be initially assumed from a setting where markets are located at a very small distance from the border, not all traders are engaged in cross-border trade. Those who carry out a cross-border activity occupy a position that makes them highly important for the functioning of the overall structure.

In a third part, we discuss certain important differences between the formal and the ethnographic approaches, especially the ways in which social network analysis defines brokers and central actors, and argue that formal network analysis should be seen as complementary to conventional ethnographic studies. We also show how combining formal network analysis with ethnographic studies may contribute to support the recent revival of interest in cross-border trade in the policy field, which contrast with the views of the 1990s, during which traders were rarely rewarded for their role of suppliers of an increasingly urbanised continent and informal activities were perceived as a threat to economic growth and political stability (Meagher, 2010; Taylor, 2012).

2. Social Networks and Economic Performance

A certain consensus has emerged among historians, geographers, anthropologists and political scientists to consider that social networks can potentially enhance economic efficiency or undermine economic development, depending on the structural position and underlying strategies of the actors (Bähre, 2012; Goyal, 2012; Meagher, 2005; Urwin, Di Pietro, Sturgis, & Jack, 2008).

In Africa, this fundamental duality of social networks has been demonstrated in several historical studies that looked at the social composition of trade networks. One the one hand, these studies show that the structure of trade networks was reliant on factors such as birth, kin and ethnicity. In many African regions, traders recruited their sons, not their clients, into their businesses as junior partners, with birth providing legitimacy and allowing the creation of a group of obligors bound by family links and alliances (Grégoire, 1992). For example, family ties were particularly important in long-distance kola trade between Asante and Hausa and for Kooroko merchants during pre-colonial times (Amselle, 1977; Lovejoy, 1980). In other cases, the transmission of business knowledge between fathers and sons was disrupted by the brutal decline of trade cities or the establishment of national borders and railway roads (Howard, 2014). In Eastern Niger, for example, economic hardship in the early years of colonial rule, uncertainty in the cattle business, and inheritance law division of estates between heirs prevented the emergence of family dynasties (Baier, 1980).

On the other hand, empirical studies also indicate that kin was frequently not the preferred means of doing business due to the social pressure that resulted from economic transactions between family members. Family ties offer comfort, but social obligations can also undermine prosperous businesses or prevent enterprising individuals from developing innovative activities, pushing traders to develop original ways to accumulate wealth (Whitehouse, 2012). When kin relations constitute a burden, traders can choose to conceal their prosperity from their families by adopting a modest lifestyle or join groups with different internal norms, such as reformist religious movements that promoted austerity and a reduction of social obligations. Numerous large traders from the Niger–Nigeria border, for example, have joined the Society for the Removal of Innovation and Reinstatement of Tradition in the last decades, not only because the Islamist movement provided trust, but also because social obligations were reduced among its members (Masquelier, 2009). Another strategy adopted by traders is to emigrate and establish trade diasporas far enough from their community of origin to escape from the daily pressure from kin, but close enough to play the role of cross-cultural brokers. Since Cohen’s (1969 [2004]) study on the Hausa trade diaspora of Ibadan, many studies have documented how these spatially dispersed communities provided a suitable ground for reducing uncertainty and ensuring trust among their members (see Pellow, 2002; Whitehouse, 2012).

Econometric studies have also documented the duality of social networks. On the one hand, social networks appear to enhance trader’s productivity due to better information on prices, the trustfulness of clients and suppliers, and potential access to commercial credit (Fafchamps & Minten, 2002). In a business environment characterised by weak formal institutions, being connected to numerous others traders, suppliers and clients positively influences the economic performances of the traders and provides the trust that neither the state nor formal business institutions can provide (Fafchamps, 2004). On the other hand, social networks do not benefit all equally. Because they promote the well connected rather than the most qualified, they make it hard for newcomers to enter a market. Being too embedded in a set of interpersonal relationships has disadvantages because it decreases the ability of traders to reach external resources, such as foreign business partners of a different ethnic or religious background.

Another negative aspect of social networks is that their propensity to work across sectors and administrative levels can potentially lead them to destabilise the fiscal and monetary control of nation states. Competing with formal institutions, social networks have frequently been accused of impeding their development, leading to the criminalisation of states, corruption, and instability (Bayart, Ellis, & Hibou, 1999). These social liabilities are especially true when interpersonal networks include not only businessmen but also politicians and civil servants. In this case, market-oriented firms have a hard time competing with network-based firms, unless they also rely on interpersonal links within the state. In the particular case of cross-border trade, the difficulty of predicting which of the formal or informal rules will be applied when dealing with trade agreements, blockade of key ports, or border conflicts, has led many traders to develop clientelist ties with state officials, or engage in politics themselves (Boone, 2006; Kraus, 2002; Tijani Alou, 2012).

In parallel to the ethnographic approach, recent research conducted in a variety of disciplinary and geographical contexts by network scientists has reached similar conclusions regarding the duality of social networks (Burt, 2005; Everton, 2012; Fleming, King, & Juda, 2007; Uzzi, 1996). As shown in Table 1, maximum performance is attained when an actor is simultaneously embedded in a cohesive group while being able to create diverse external contacts between actors that are not themselves connected. A strong degree of embeddedness can be beneficial to the activity of social actors, since it provides trust between peers and reduces the risks associated with business activities. Strongly embedded actors are therefore regarded as very central, in the sense that they are surrounded by a large number of other actors with whom they frequently interact to exchange information, draw resources or communicate orders. This particular form of centrality is known as degree centrality because it refers to the number of ties of each actor or degree. In many rural societies, for example, traditional chiefs have a high degree centrality, because they are the centre of a large network of family, ethnic and political ties within the local community. Degree centrality is a local measure of centrality, in the sense that it only takes into account the immediate connections of an actor. It captures only a partial aspect of the centrality of social actors, which can also arise from their ability to reach other actors beyond their own group and play the role of structural brokers (Scott & Carrington, 2012).

The importance of brokers is formally measured through betweenness centrality, which refers to the importance of bridging actors that would otherwise not be connected. It is a global centrality measure calculated on the entire network and based on the number of shortest paths between actors. In the formal network literature, brokers are defined as actors who bridge several disconnected groups and benefit from the existence of areas of low density in the networks, known as ‘structural holes’ (Burt, 1992). Traders connected to business partners in distant markets or returning immigrants can play such brokerage role between their own community and the outside world.

Table 1. Combining embeddedness and brokerage

	Low degree of embeddedness	High degree of embeddedness
High brokerage beyond group	Divisive group with diverse external contacts	Cohesive group with diverse external contacts (maximum performance)
Low brokerage beyond group	Divisive group with homogeneous external contacts (minimum performance)	Cohesive group with homogeneous external contacts

Source: Burt (2005, p. 139).

As Spiro, Acton, and Butts (2013) demonstrate, brokerage is a fundamentally dynamic process, which can generate value in three different ways. Firstly, brokers can transfer resources between two disconnected parties, a situation known as tertius gaudens or ‘rejoicing third’, as when traders act as a bridge between importers and final consumers along transnational routes. Secondly, brokers can facilitate matchmaking between two actors to the benefit of each, a situation known as tertius iungens, or ‘the third who joins’ (Obstfeld, 2005). Finally, the structural advantage of a broker can come from his ability to coordinate the activities of third parties without creating a direct relationship between them, which reinforce their dependence on the broker.

Some actors may not be aware of their structural role as broker. As Morselli (2009, p. 17) shows, ‘being a broker is not simply a role that can easily be identified, as would be the case in occupations such as a stock broker, a real-estate broker, or even a power broker – in such cases, the occupation or role defines the position’. Rather than a recognised professional occupation, brokers occupy a position that can vary according to the kind of information or resources conveyed by the network. The structural definition of brokers significantly differs from the one commonly found in the African studies literature (Walther, 2014), which regards brokers as a specialised institution whose aim is to work between buyers and sellers on a market (Brooks, 1993; Little, 1992). In their study of the Hausa trade, for example, Grégoire and Labazée (1993, p. 22) identify a number of socially recognised status (such as courtiers, rabatteurs or coxeurs), which all work at facilitating business transactions.

Similarly, the formal network literature defines powerful actors according to their degree centrality. Actors with a high degree centrality are assumed to be powerful because, being at the centre of the network, they can better control information flows, give advice and orders, and influence outcomes than peripheral actors. This definition differs from the ethnographic definition of big men which takes into account the ability of certain actors to achieve economic and political objectives through social patronage and redistribution (Nugent, 1995; Utas, 2012).

3. Case Study and Methodology

3.1. Two Border Regions

Data were collected between January and December 2012 on five border markets located between Niger, Nigeria and Benin: Gaya–Malanville–Kamba (GaMaKa) and Birni N’Konni–Illela (BNI) (Figure 1).

Figure 1. Location of case studies.

Cartography: author.

The two border regions were selected because their historical development of trade is rather dissimilar, which, we assume, should result in two structurally different trade networks. In the GaMaKa region – known as the Dendi – long-distance trade activities have been encouraged by the liberalisation of trade that occurred since the 1980s in Benin and Niger. As mentioned in the introduction, the small city of Gaya, whose estimated population was 45,000 in 2010, has progressively transformed into a regional hub for wholesalers willing to trade with Nigeria, where the imports of numerous goods, such as second-hand clothes, are prohibited. A few kilometres south, the city of Malanville (population 60,000) is a regional centre for agricultural goods produced in the River Niger Valley, such as onion, cassava and cereals. A colonial creation, the market of Malanville is one of the largest in Benin and attracts a considerable crowd of foreign traders from neighbouring countries (Walther, 2009). The busy commercial activities of Gaya and Malanville contrast with the relative decline of the neighbouring Nigerian city of Kamba (population 27,000), whose economic development has been slowed by the diminishing comparative price advantage of Nigerian products, poor road conditions and insecurity.

The Dendi region was historically located on one of the main trade routes developed between Hausaland and Asante in the nineteenth century. Caravans of Hausa traders from Kano, Sokoto and Jega stopped in Gaya before crossing the Niger River on their way to Sansanne Mango, Yendi-Gamaji and Salaga, where kola nuts were purchased. The region was also a production centre of salt, extracted in the fossil valleys of the Niger River and exported regionally by Dendi merchants (Lovejoy, 1986). These merchants exerted such a considerable influence that, by the nineteenth century, Dendi had become the dominant trade language of the northern regions of present day Togo and Benin.

Today, the Dendi population of Gaya, Malanville and Kamba is predominantly engaged in agricultural activities. Up to 80 per cent of the large traders surveyed in this article do not come from the city in which they work. They are alien entrepreneurs of Songhay, Zarma, Tuareg, Igbo, Fulani and Hausa origin attracted by the opportunities of cross-border trade. The reason why trade is currently dominated by foreign traders if the Dendi used to have a pre-colonial history of long-distance trade is not clear from empirical evidence. Trade could have suffered from major disruptions in the nineteenth century, perhaps due to the Fulani jihad of 1804–1808, which has destroyed the business of many traders in Hausaland and in the neighbouring regions (Lovejoy, 1980). The historical decline of the Dendi could also have come from the reorganisation of trade networks, from the interior of West Africa to the Gulf of Guinea, which would have made the commercial route passing through Gaya obsolete.

Three hundred kilometres further north-east, the border area between Birni N’Konni (population 63,000) in Niger and Illela (population 32,000) in Nigeria is characterised by the highly informal nature of economic activities and the historical Hausa ethnic networks that developed before colonial times (Amselle & Grégoire, 1998). Birni N’Konni was the capital of a small Hausa state before becoming a tributary of Gobir in the middle of the eighteenth century, and of the Sokoto Caliphate in the nineteenth century. Both Birni N’Konni and Illela were integrated in the trade route that linked Sokoto to Agades in the nineteenth century, along which livestock, dairy products and salt from the north were traded against millet and cloth from the south. The enforcement of customs regulations between French and British colonies led to a major disruption of trade but local traders quickly adjusted to the new economic order imposed by colonisation. In the post-colonial era, the existence of national discontinuities became even more conducive to the establishment of transnational trade networks and to the rise of a new generation of businessmen enriched by the re-export of cigarettes, textiles and other manufactured and agricultural products (Grégoire & Labazée, 1993). Today, Birni N’Konni and Illela constitute an important border post on the road between Sokoto and Niger, as well as on the east–west N1 highway which cross the Republic of Niger. Contrasting with the GaMaKa region, a large majority of the traders come from the cities where they do business and are of Hausa origin.

3.2. A Network Approach

This study builds on social network analysis (SNA), which is both a paradigm of social interactions based on graph theory and a methodology using statistical tools to formally describe, represent and model structures that are not easily visualised (Newman, 2010). This approach regards social networks as a finite set or sets of actors linked to one another by social ties. It is particularly appropriate to understand trade, which is a fundamentally relational activity in which the constraints imposed on an individual and his outcomes are explained by the structure of the social network (Spiro et al., 2013).

The first step of our survey was to identify a number of products that would be important for the economic development of the border regions. The identification of the relevant products was based on a preliminary macro-analysis of international trade conducted on the basis of longitudinal data from national customs authorities, pre-existing surveys made in the border regions in the mid-2000s and an analysis of the goods banned in Nigeria, which are of high importance for re-export trade in West Africa (Raballand & Mjekiqi, 2010; Walther, 2012). This combination of data led us to identify four products whose trade would be highly important for the traders located in the two border regions under consideration: building materials; cereals and flour; textile; and used clothing. The trade of some of these products is sometimes considered illegal on one side of the border and legal on the other, such as textiles. As in previous studies (Igué & Soulé, 1993), we considered these trade flows as relevant for our survey, insofar as they did not generate criminal activities comparable to those of trafficking weapons, drugs or precious metals, but were merely the result of different legislation within the region.

Once these products were identified, we identified the actors involved in trading them. Social network analysis differs from most of the surveys conducted in Africa in the sense that we are primarily interested in collecting data on the relations – rather than the attributes – of social actors. In contrast to traditional surveys that consider social actors as independent units that can be added until they constitute a representative sample of the population, our data refer to non-independent observations. Sampling a population would not work in our case because we don’t know how the social actors are intertwined with each other before we start our analysis and, by randomly selecting some of them, we would miss a large number of relevant connections. In order to address this issue, we used snow-balling techniques, a recognised alternative that allows identifying new economic agents from among the subjects’ existing acquaintances. Snowball sampling is particularly adapted to the study of cross-border traders, who don’t necessarily belong to a formal professional institution in which insiders could easily be distinguished from outsiders, and whose number and activities are extremely difficult to evaluate from the investigator’s perspective.

With the help of local freight agents, we selected the traders whose annual turnover was over 100 FCFA million (€152,000) in 2010. This threshold is used by freight agents to distinguish between small traders whose activities are predominantly limited to petty business and large traders who can potentially employ a greater number of employees and conduct much more ambitious trade activities. Focusing on large traders, we conducted a first wave of interviews with 43 of them, during which they were asked to nominate whoever they considered as business partners, whatever their age, gender, ethnic group, nationality or religious membership. The interviews were conducted in French and Hausa by the author and several colleagues from the University of Niamey. We collected data on the existence and frequency of business interaction between the actors, as well as on the type of relationships between them (family, neighbour, organisation member, work, other friend). We concluded that a tie existed if two actors had been in business in the last two years (2010–2012). These interviews produced a list of 114 nominees. We conducted two additional waves of interviews with the actors mentioned at least three times by the first wave. After three waves of interviews, the same names started appearing again and again, which meant that population saturation was reached. With a response rate of 87.9 per cent in the GaMaKa trade network and 88.9 per cent in the BNI trade network, we work on an almost complete population and our results are not considered to be negatively affected by missing data. Contrarily to the Gulf of Guinea, where some female traders have acquired prominent positions in the textile business, all the traders interviewed in this article – except for one – are male.

4. The Structure of Trade Networks

This section investigates the social structure of the networks found between Gaya-Malanville, and Kamba (GaMaKa) and Birni N’Konni-Illela (BNI). We start by testing whether the trade networks are more centralised or more decentralised and studying the relative importance of central actors and brokers, before turning to the spatial structure of the networks and look for a border effect.

4.1. Two Decentralised Trade Networks

Centralised networks are composed of a small number of actors with many ties and tend to be more efficient in terms of coordination than decentralised structures, because information, orders and resources can be more easily transferred from central nodes to the rest of the network. The star network in which a single actor occupies the centre of the network and where peripheral actors have no ties between each other is the most extreme example of a centralised network. Its opposite is the fully connected network, a completely decentralised structure where every actor is connected to every other actor. This structure is, however, much more resilient to threats than the star network because of its redundancy of ties.

Our results show that the structure of both trade networks is composed of few actors (or nodes) with few business relations (or ties) (Table 2). The BNI network is composed of 53 nodes and 64 ties, while the GaMaKa network is a little bit larger, with 83 nodes and 104 ties, including two isolated dyads of actors. Both networks have a low density: only 4.6 per cent and 3.1 per cent of the potential ties between the actors are actually present in the network. The clustering coefficients of 0.094 and 0.062, which ranges from 0 (no cluster) to 1 (one single cluster), also indicate that both networks are highly decentralised. Most of the actors have relations with a limited number of contacts: the average number of business partners is 2.4 in both regions. This result is consistent with previous studies that showed that African traders engage in repetitive transactions with existing business partners and that past collaborations exert a considerable influence in shaping the structure of current trade networks (Fafchamps, 2004). Both networks have a relatively short characteristic path length – every actor can be connected to every other by an average of 3.7 steps in the BNI region and 4.2 steps in the GaMaKa region. Short path lengths theoretically designate a rather compact network where information spreads easily, as opposed to networks with long path lengths, which tend to be linear and less efficient in the spread of information. On the basis of these structural characteristics, both networks approximate a random network, a structure with a low degree of clustering and short paths. They differ from a regular network, an ordered structure where every actor has the same number of ties and is composed of long paths and strong clusters. They are also different from a small-world network, where most nodes can be reached from every other by a small number of steps even if they are not neighbours of one another (Watts & Strogatz, 1998).

Table 2. Key metrics for the two networks

	Birni N’Konni-Illela (BNI) network	Gaya–Malanville–Kamba (GaMaKa) network
Number of nodes	53	83
Number of links	64	104
Number of dyadic isolates	0	2
Density	0.046	0.031
Average number of ties	2.385	2.357
Clustering coefficient	0.094	0.062
Average path length distance	3.712	4.203
Average degree	0.046	0.031
Average betweenness	0.053	0.036
Degree centralisation	0.172	0.181
Betweenness centralisation	0.403	0.320

Calculations are the author’s; dyadic isolates are sub-groups of two actors separated from the rest of the network; the clustering coefficient refers to the extent to which the actors tend to cluster together; average measures refer to the average centrality of the actors involved in the network; centralisation measures highlight whether certain nodes are exceptional in terms of centrality. Individual centrality measures for top-scoring actors can be found in the Online appendix.

In addition to being loosely connected, the actors of both networks have relatively little centralisation, as can be seen from their low average measures of centrality. The average degree measures whether nodes are linked to a particularly high number of actors and the average betweenness measures whether a node plays the role of a gatekeeper by controlling access to other nodes. The decentralised nature of the networks is also confirmed by further centralisation measures, which test whether the network contains potentially exceptional nodes whose centrality would strongly differ from the rest of the network. Table 1 presents two measures: degree centralisation and betweenness centralisation. All measures vary from 0, a situation where no node is atypical in terms of centrality, to 1, a situation where the centrality of one node exceeds all nodes. Both networks have low values for degree centralisation (0.172 in BNI and 0.185 in GaMaKa), which makes them less sensitive to fragmentation, since no node seems to be particularly highly connected. The relatively higher measures found for betweenness centralisation – 0.403 in BNI and 0.320 in GaMaKa – confirm, however, the existence of prominent brokers, which makes sense in a border setting where much profit can be expected from bridging different partners from various countries.

4.2. Embeddedness and Brokerage

We now turn to the trade-off between degree centrality (embeddedness) and betweenness centrality (brokerage), two centrality measures which, when combined, produce social capital. As discussed in the previous section, both networks are highly decentralised, with few actors having a high degree centrality, which measures immediate influence. The relative absence of actors with high degree centrality is particularly evident on Figure 2, which maps business relations between traders in the GaMaKa region. The size of the nodes reflects the number of ties developed by each actor and the colour refers to each of the border markets. A part from El Hadj Harouna, a Nigerian textile wholesaler who entertain numerous business ties with relatively minor actors, most of the traders involved in the GaMaKa network are not particularly central. The same applies to the trade network of the BNI region.

Figure 2. Gaya–Malanville–Kamba trade network: degree centrality.

Note: the size of the nodes is proportional to their relative importance in terms of degree centrality. The more ties an actor has, the bigger it appears.

Source: author.

What makes the two networks interesting is how certain actors have managed to occupy a strong brokerage position. In both regions, trade is organised through certain brokers, who bridge several sub-groups of actors. The sociogram presented in Figure 3 provides a visual representation of the betweenness centrality scores in the BNI network. Betweenness centrality is based on the number of shortest paths that pass through a given actor, and therefore gives an indication about its ability to control crucial flows of information and resources between groups. Each actor is represented according to his importance as a broker (size) and according to his country membership (colour). Figure 2 shows that business relations between traders are mediated through a limited number of prominent brokers, such as El Hadj Zainou, a successful Hausa trader from Birni N’Konni, whose professional contacts developed with traders in Kano helped him become the Nigerien representative of a large Nigerian wholesaler. El Hadj Talatu, a native of Sokoto active in the business of sugar and cooking oil, is another example of a prominent broker. Without these actors, the network would break into several isolated components, some of them organised by nationality.

Figure 3. Birni N’Konni–Illela trade network: betweenness centrality.

Note: the size of the nodes is proportional to their relative importance in terms of betweenness centrality. Brokers are characterised by larger nodes.

Source: author.

Brokers can play different structural roles depending on whether they bridge actors from within or beyond their own group. In our case, we hypothesise that borders create several groups of traders according to their country of residence and test whether their brokerage roles vary according to their country membership. To do so, we draw on the typology developed by Gould and Fernandez (1989), which identified several types of brokers based on their structural position: coordinator; itinerant; gatekeeper; representative; and liaison brokers. In a triad composed of a broker (A_b) and two other nodes, coordinators belong to the same group as the nodes they bridge (A–A_b–A). In our case, this would refer to traders coordinating economic activities within Niger, Nigeria or Benin respectively. For their part, itinerant brokers connect two nodes from a different group than their own (B–A_b–B), for example when a trader from Nigeria connects two Nigerien or two Beninese traders. These two categories of brokers are known as within group roles, because the A or B belong to the same group, contrarily to gatekeepers, representatives and liaison brokers, who all connect individual between groups. In a non-directed network such as ours, gatekeepers and representatives have identical values and connect a source or a recipient to a different group (A–A_b–B or B–A_b–A). Liaison brokers connect two nodes from different countries (B–A_b–C). This latter brokerage role is only relevant if there are more than two countries, which is the case for the GaMaKa region, located between Niger, Nigeria and Benin.

Table 3 summarises the relative importance of each of the brokerage scores according to the country of residence of the actors. These are relative scores, which means that the raw scores obtained by counting the number of times each actor occupies a brokerage position within each network have been divided by expected values given group sizes. The results for the within group roles show that coordinators bridging traders who belong to the same country are by far the most represented brokers in both networks and in all countries. Coordinators are particularly prominent in the GaMaKa network, which stresses the importance of national – rather than cross-border – business relations. Itinerant brokers bridging two actors from the same country are rare, except in the Nigerien part of the BNI network (0.87). As for the between group brokerage roles, our results indicate that gatekeepers/representatives are particularly represented in the Nigerian (1.00) and Nigerien (0.80) parts of the GaMaKa network. Liaison brokers, whose role can only be analysed in a trinational setting because they imply three nationalities of actors, are surprisingly weakly represented in the GaMaKa region, where the conditions for bridging nodes that belong to different groups appeared theoretically to have been met. These results suggest that the most frequent brokers in the region bridge people from their own country.

Table 3. Relative brokerage scores by country and case study

	Coordinator	Itinerant	Gatekeeper/representative	Liaison
	(A–Ab–A)	(B–Ab–B)	(A–Ab–B, B–Ab–A)	(B–Ab–C)
BNI network
Niger	2.04	0.87	0.58	–
Nigeria	3.10	0.19	0.42	–
Whole network	2.66	0.47	0.49	–
GaMaKa network
Niger	5.59	0.16	0.80	0.73
Benin	7.12	0.38	0.35	0.04
Nigeria	4.38	0.13	1.00	0.27
Whole network	6.22	0.27	0.59	0.08

Notes: liaison brokers imply at least three groups (countries).

Source: author’s calculation.

Thus far, we have separately described how large traders of the BNI and GaMaKa regions differed in terms of embeddedness and brokerage. In what follows, building on Burt’s (2005) theory of social capital, we combine the centrality scores obtained earlier and show how social capital results from a combination of embeddedness and brokerage. In Figure 4, we plot the degree and betweenness centrality scores of all traders. In each region, we use the mathematical limit of one standard deviation from the mean to split the actors into four quadrants. Because the size and the composition of the two networks vary, the mathematical limits between the quadrants are slightly different: the one concerning actors from the BNI region appears with a continuous line; the one concerning the GaMaKa region with a discontinuous line.

Figure 4. Combining embeddedness and brokerage.

Source: author.

Actors whose degree and betweenness centrality scores are higher than normal values occupy the upper-right quadrant, such as Elh Hadj Yacine, a Zarma trader whose connections with other suppliers in Niamey, China, and Dubai make it one of the most successful businessmen of the region. Actors whose centrality scores are below normal values occupy the lower-left quadrant. Deprived of both dimensions of social capital, these actors struggle to build a dense cluster of close business partners and, at the same time, have few connections beyond their group. Actors who are very central on degree centrality only occupy the lower-right quadrant of the scatter plot. El Hadj Adams is a good example of such actor: his position as president of the traders of Gaya has contributed to reinforce his degree centrality rather than his broker role. Actors who are very central on betweenness centrality only occupy the upper-left quadrant. El Hadj Babangida, for example, is a structural broker with relatively little degree centrality: established in the market of Illela since 1979, this Hausa trader has become the chairman of the cereal market and a wealthy businessman connecting Niger and Nigeria. As the next section shows, the profile of these actors is strongly constrained by the existence of national borders and historical development of the region.

4.3. The Effect of National Borders

The major difference between the two trade networks comes from the fact that the proportion of ties within a group compared with the ties outside of the group – called homophily – is much larger in the GaMaKa network than in the BNI network. Despite the fact that local traders are located in the immediate vicinity of a national boundary, 86.6 per cent of the ties exchanged within the GaMaKa network are with business partners from the same country, which is the sign of a highly homophilous network (Table 4). Between Birni N’Konni and Illela, on the contrary, Nigerien and Nigerian traders are not strictly segmented according to their country of residence: far more cross-border ties have developed and only 67.6 per cent of the ties are within traders from the same country. This fundamental difference between the two networks is confirmed by the fact that the E/I index, calculated as the difference between external (E) and internal (I) ties for each country, divided by the total number of ties, is high and negative in the case of GaMaKa (−0.727**). This indicates a preference for homophilous ties. The E/I index is neutral in the BNI network (0.108), signalling that country membership is a not a relevant attribute for explaining the structure of the network. In sum, most of the large traders in the BNI region are used to travelling across the national border to do business with their foreign correspondents and none of them occupies a strong brokerage position between countries as in the GaMaKa region, where we find few cross-border ties.

Table 4. Cross-border ties and nodes

	BNI network	GaMaKa network
Cross-border ties	31.3%	18.3%
Nodes with cross-border ties	43.4%	31.6%
In Gaya (Niger)	–	40.0%
In Malanville (Benin)	–	34.5%
In Kamba (Nigeria)	–	20.0%
In Birni N’Konni (Niger)	56.0%	–
In Illela (Nigeria)	32.1%	–

Source: author’s calculation.

The difference in the spatiality of the two networks reflects the historical development of both regions. Between Birni N’Konni and Illela, in the heart of Hausaland, historical ethnic networks have constantly evolved since precolonial times, resulting in a fine mesh of business ties. In contrast, the development of the trade activities between Gaya, Malanville and Kamba is much more recent and linked to the liberalisation of international trade in Niger and Benin. In this region, trade policies adopted by Benin in 1973 encouraged re-export trade, by maintaining lower import barriers than Nigeria (Golub, 2012). These favourable customs regulations allowed goods to be imported to Benin and then re-exported to neighbouring countries, stimulating the development of a foreign trade diaspora in Malanville. The situation changed in the mid-1990s, during which a number of Nigerien importers established in Malanville found out that it was more profitable to import goods from the world market through Benin free of tax and stock them in the city of Gaya before eventually re-exporting them to Nigeria. Despite being more circuitous, the route Benin–Niger–Nigeria is preferred to the route between Benin and Nigeria because goods imported duty free into Benin have to be declared as in transit for Niger rather than Nigeria, where the import of textile is banned. As a consequence, Benin imports a very large volume of goods declared as in transit to land-locked countries, when in fact everyone knows the goods are mostly ultimately destined for Nigeria. Many experienced traders that had initially chosen to settle in Malanville have subsequently built large warehouses in Niger and continue to commute frequently between the two markets.

There are a number of interesting differences within each region: the proportion of cross-border ties and nodes is always higher for actors located in Niger, as can be seen in the border markets of Gaya and Birni N’Konni, and the lowest proportion for traders located in Nigeria, as in Kamba and Illela (Table 4). This element suggests macro dependence of Niger towards Nigeria, as traders located in Niger are obliged to develop transnational business interactions.

These results show that social relationships in trade can vary dramatically in terms of structure, which speak to the importance of using social approaches as a complement to other methods of understanding cross-border trade.

5. Discussion

A formal approach to social networks can contribute to a better understanding of cross-border trade in many ways. At the individual level, it can highlight how traders are related to each other, which traders are prominent and which are not, what make some traders different from others, and, if temporal data are available, how the centrality of traders has changed over time. Social network analysis is also particularly well adapted to study the overall architecture of a network (its topology), which has direct effects on individual behaviour. The two case studies discussed above have made clear that cross-border trade was organised in a rather decentralised way. This particularity is in line with previous ethnographic studies, which noted that transnational transit networks that import manufactured goods into Nigeria were less populated and less hierarchised than local networks because they didn’t rely on large physical and human infrastructures (Grégoire & Labazée, 1993). Compared to highly centralised structures, decentralised ones are less efficient, because information and resources flow less easily between the actors. They are, however, well adapted to the constant variations in the volume and direction of business activities due to the seasonal and cyclical food shortages that affect West Africa. Depending on the season or on the availability of supplies, economic flows can abruptly decline or change directions, which make the establishment of a network of business partners strategically located on both sides of national borders indispensable.

A formal approach to networks can also help understand the spatiality of cross-border trade, that is, how social actors are spatially connected and the impact of national borders. Our results show that there is a correspondence between the type of trade (domestic or cross-border) and the structure of the network that underlies it. This is congruent with the existing ethnographic literature, which showed that four types of networks could be observed in West Africa depending on their geography (Grégoire, 2003): (1) the mesh network typical of short distance trade powered by ecological complementarities, the landlocked situation of many Sahelian countries, market size differences, disparities in economic and monetary policy, and currency disparities, which corresponds to the one studied in this article; (2) the star network typical of certain ethnic groups whose trade is centralised in a regional centre or capital city; (3) the string network typical of pre-colonial trans-Saharan trade, which has been modernised to include smuggling of cigarettes, drugs, and migrants; (4) the bipolar network which characterise long-distance relationships between Africa and the Persian Gulf.

Mapping the exact topology of these networks could also contribute to support the recent revival of interest in cross-border trade in the policy field. In recent years, numerous international financial institutions, regional bodies, and aid organisations have aimed to promote informal trade activities, now regarded as ‘the most efficient, organised and deep-rooted system of trade in the region’ (African Development Bank, 2012a, p. 9). The commercial skills of traders are widely seen today as a crucial instrument for alleviating poverty due to income earnings and employment opportunities for poor households and the ability of trade networks to allocate supplies during food crises (CILSS, 2014). This revival is supported by numerous initiatives that aim at improving both soft and hard infrastructures (African Development Bank, 2012b; McLinden, Fanta, Widdowson, & Doyle, 2010; OECD, 2009; World Bank, 2013). Instead of capturing mobile investment and redirecting it towards certain regions, growth-enhancing policies aim at improving the qualifications of the local workforce, supporting existing entrepreneurs, and creating the conditions for the development of new economic activities. A crucial aspect of such policies is the improvement of the body of rules and regulations governing regional trade, which are currently seen as an obstacle to regional integration. The objective of ECOWAS’s Regulatory Informal Trade Programme, for example, is to incorporate current business practices that circumvent the state into the formal economy, by simplifying administrative, tax and custom procedures, and encouraging traders to register their businesses. Other programmes developed by UEMOA, USAID and ECOWAS aim at eliminating corruption and illegal payments at border posts and along trade routes (USAID, 2014), or establish joint border posts, such as the one inaugurated in 2014 between Gaya and Malanville, to accelerate border clearance. Complementary to growth policies, network-enhancing policies currently implemented in West Africa aim at improving both the internal connectivity of economic actors at the local level, and their external connectivity with the rest of the world. The extension and rehabilitation of road, rail and maritime infrastructure is a crucial aspect of such policies. The Trans-African Highway and the African Development Bank’s Aid for Trade Trust Fund, for example, plan to develop the roads connecting the major urban centres of the continent and, hence, reduce transport costs.

A formal approach to trade networks could contribute to both growth-enhancing and network-enhancing policies. It could first help identify the strong ties between economic actors within the region and the relatively weaker ties between these local actors and the rest of the world. As our study shows, the most successful traders are those who can achieve a high level of internal and external ties, whereas those deprived of both dimensions of social capital are disadvantaged. Development initiatives would have a higher impact on trade if they enhance the positive aspects of intracommunity ties in business communities, which are synonymous with solidarity and protection against uncertainty, while simultaneously supporting the creation of external ties between traders and governments, non-governmental organisations, and aid agencies.

Combined with qualitative approaches that provide information on the historical and social nature of the ties, formal approaches to trade could also contribute to the development of trade policies that take into account the local economic, social, political and institutional conditions in which economic actors are embedded. As our article has shown, despite the proximity to the border, regional integration occurs either through a multitude of cross-border ties, or through a limited number of key brokers depending on the historical roots of the networks. Because of the local variety in the structure of trade networks, development policies that aim at facilitating cross-border trade would be more successful if they are place-specific rather than spatially blind. Unlike spatially blind policies that are applied without explicit consideration to space and encourage migration to large and densely populated regions, place-based policies acknowledge that the local actors and institutions shape the development potential of cross-border trade. They appear crucial for the local development of traders and border markets, for which the existence of national borders constitutes a resource.

6. Conclusion

So far, studies devoted to cross-border trade have rarely – if ever – used formal approaches to represent and analyse social networks in West Africa, relying instead on econometric methods that consider the price of commodities as a proxy for evaluating regional integration, or on qualitative approaches that aim at understanding the professional history of traders as well as their sociability. Our article argues that the models describing the organisational structure of social networks developed by social network analysis over the last decades also apply to West African trade networks.

Our first contribution is to study how trade networks vary in terms of centralisation. We find that both networks are rather decentralised, with few business relations between the actors. In a business environment where transaction costs are extremely high, decentralised networks are well adapted to the various uncertainties induced by long-distance trade. As in other regions of the world, combining embeddedness and brokerage, the former referring to the inclusion of social actors within their group and the latter referring to the ties that actors build beyond their group, is crucial for the success of cross-border trade.

Our second contribution is to show that the effect of national borders varies greatly across regions. Between the markets of Birni N’Konni and Illela, large traders have developed numerous cross-border links. This contrasts with the situation between the markets of Gaya, Malanville and Kamba, where, despite the proximity to national borders, trade mainly relies on few cross-border ties that provide opportunity for a limited number of brokers. The article shows that the spatial form of trade networks is constrained by the historical origin of the traders engaged in cross-border activities. In those markets where trade has followed the recent liberalisation of commerce and where most of the traders are from outside the region, borders exert a greater influence than where trade has developed since pre-colonial times.

Our third contribution is to show that future research would benefit from going beyond the current gap between ethnographic and network-based approaches to trade in West Africa. Qualitative approaches are well adapted to capturing the complexity of social ties, the meanings that social actors give to their relations, the historical formation of networks, and their institutional context, but they would be ideally complemented with a more formal approach. This appears especially productive when a great number of actors is involved or when many of them are hidden, as is often the case in border regions. Cutting across sectors and professional status, social network analysis can complement the ethnographic studies and provide a comprehensive picture of cross-border trade by addressing the structure and agency simultaneously.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Acknowledgements

This research was supported by the National Research Fund of Luxembourg (CROSSTRADE Project C10/LM/783313 and WANETS Project MOBILITY/12/4753257). An earlier version of this article was presented at the African Studies Association Annual Meeting in Baltimore in November 2013 and at the World Bank Workshop on Cross-Border Trade in Washington, DC in March 2014. The author thanks Emmanuel Grégoire, Leena Hoffmann, Al Howard, Moustapha Koné, and Bill Miles for their comments. Data can be obtained from the author on request.