lable at ScienceDirect

Socio-Economic Planning Sciences 48 (2014) 198e219

Contents lists avai

Socio-Economic Planning Sciences

journal homepage: www.elsevier .com/locate/seps

Market analysis and transportation procurement for food aid inEthiopia

Marie-�Eve Rancourt a, *, François Bellavance b, Jarrod Goentzel c

a CIRRELT, ESG UQ�AM, case postale 8888 succursale centre-ville, Montr�eal H3C 3P8, Canadab CIRRELT, HEC Montr�eal, 3000 chemin de la Cote-Sainte-Catherine, Montr�eal H3T 2A7, Canadac MIT Center for Transportation & Logistics, 77 Massachusetts Avenue, Cambridge 02139, United States

a r t i c l e i n f o

Article history:Available online 11 July 2014

Keywords:Transportation procurementEconometric modelMarket analysisRoad infrastructureCompetition intensityHumanitarian logistics

List of acronyms: CIRRELT, Interuniversity Researcworks, Logistics and Transportation; ESG UQ�AM, �Ecolel’Universit�e du Qu�ebec �a Montr�eal; MIT, Massachuset* Corresponding author. Tel.: þ1 514 987 3000x530

E-mail addresses: rancourt.marie-eve@uqa(M.-�E. Rancourt), francois.bellavance@hec.ca (F. Be(J. Goentzel).

http://dx.doi.org/10.1016/j.seps.2014.07.0010038-0121/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

Empirical research characterizing transportation markets in developing countries is scarce. By analyzingcontracts between the World Food Programme and private carriers, we identify the determinants oftransportation tariffs in Ethiopia and quantify their relative importance. The econometric models devisedfrom our unique dataset provide insights for shippers to improve procurement processes, for carriers todevelop competitive business models and for government authorities to define effective investments andpolicies. Results indicate that the number of carriers on a given lane significantly reduces transportationtariffs and that policies stimulating competition may be as important as road infrastructure investmentsin Ethiopia's development strategy.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

In Africa, freight transportation can be a costly and time-consuming component of the supply chain due to issues such aspoor infrastructure, security risks and complex regulations. Forbasic goods and commodities, transportation can represent a sig-nificant portion of the overall cost to the consumer. In the privatesector, volatile transportation markets are seen as a challenge formultinational companies looking to Africa as a key region forgrowth. Africa's competitiveness suffers from high transportationcosts [49]. This is particularly true for Sub-Saharan African coun-tries, where the average freight costs are 20 percent higher thanthose of other countries [50]. Indeed Lim~ao and Venables [24], haveshown that a 10 percent increase in transportation costs reducestrade volumes by 20 percent. Increasing the competitiveness andtransparency of transportation markets should enable many

h Center on Enterprise Net-des sciences de la gestion dets Institute of Technology.4; fax: þ1 514 987 3343.m.ca, marieeve@mit.edullavance), goentzel@mit.edu

African countries to better meet the needs of their citizens and bepositioned for economic investment.

The markets for freight transportation in African countries arenot well understood due to the lack of available objective data.Previous studies are based on surveys which may be unreliable,especially when carriers are asked about their cost structures. Inthis study, we analyze shipper data that include bids fromnumerous carriers and the subsequent contracted tariffs that wereeffective in the market. Our data set includes over 10,000 datapoints from over 75 carriers, corresponding to about 70 percent ofthe transporters registered at the Ethiopian Road Transport Au-thority at the time of data collection. These transporters provideservice on numerous origin-destination lanes within Ethiopia aswell as international corridors from nearby seaports. To ourknowledge, this study is the first to analyze carrier bids andeffective transportation tariffs in order to understand the trans-portation market in an African country.

Dry goods account for the majority of the cargo moved inEthiopia, with the primary commodities being grains and pulses(with a significant portion in the form of food aid), fertilizers,cement, sugar, coffee and oilseeds. In humanitarian supply chains,high transportation costs reduce the number of beneficiaries thataid organizations can reach with the funding they receive. Data forour study were provided by the World Food Programme (WFP), theprimary food aid transporter in Ethiopia, which moves nearly all ofits cargo in the country with third-party transport carriers.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 199

Between 1988 and 2011, the WFP delivered approximately 896,000tonnes of food aid per year on average, which translates into 22,400truckloads (TL) per year or 61 TL per day with 40-tonne trucks.Because of the large volumes and regularity of food assistance, theWFP transportation procurement and truckload operations aresimilar to those of commercial shippers.

We use multiple linear regression to determine which factorsinfluence the linehaul truck transportation tariffs, our dependentvariable. The tariffs are specified in cost per tonne for trans-portation service by lane, which is the route between the originand destination points for the cargo. Our model considers anumber of independent variables grouped into three broad sets:cost driven factors, market structure measures, and socio-economic variables. The cost driven factors include distance,road conditions and risk perception. Exploiting the richness of ourdata set, we have used information from the full set of carrier bidsto generate proxies for the market structure. Socio-economicfactors are based on regional statistics, such as population andeconomic activities.

Our results show that market structure variables, particularlycompetition intensity, are good predictors of the tariffs paid by theWFP in Ethiopia. Contrary to similar studies of transportationmarkets in developed countries, which generally find very littlevariation in prices per kilometer across lanes, our results show thatroad conditions and competition levels explain a significantamount of the variability observed in transportation tariffs forEthiopia. At a macro-level, we are able to quantify the causes for thegenerally poor development of transportation in Ethiopia. Simplereduced form counterfactual analysis shows that better road con-ditions should reduce shipping costs by 18 percent and increasedcompetition by 44 percent on the international corridors. Similarlyfor the domestic routes, better road conditions would reduceshipping costs by 12 percent and increased competition shouldreduce them by 39 percent.

The paper is organized as follows. Section 2 reviews the litera-ture on transportation procurement and market analysis, focusingon truckload and less-than-truckload sectors in both Africa andmore mature markets. Section 3 describes the WFP operations inEthiopia and outlines its transportation procurement process.Section 4 describes the data collected for this study and the vari-ables used in the regression models. Section 5 presents the modelbuilding approach and interprets the results. This is followed byconclusions in Section 6.

2. Literature review

This study concentrates on the contractual aspects, especiallypricing, between shipping organizations and commercial carriersfor transporting in-kind donations (food aid) from their points ofentry in Africa to their final distribution points in Ethiopia. Trans-portation procurement processes in this context are similar to thoseestablished in the commercial sector due to the extent and conti-nuity of such aid operations. The interested reader is referred toRefs. [2,28] for further information on food aid and food security. Inthe following literature review, we thus focus on studies related totransportation market analyses and transportation procurementsystems.

Very few empirical studies have attempted to understandtransportation costs or prices in Africa. One reason is likely due tothe scarcity of digital data archives. Empirical studies based onsurveys of the trucking industry carried out in the last 20 yearsshow that transportation prices were higher in Africa than indeveloped countries andmost developing countries. Rizet and Hine[38] have found freight tariffs to be four to six times higher in threeAfrican countries (Cameroon, Ivory Coast andMali) than in Pakistan

during the late 1980s. These authors also present an analysis of themain factors behind these differences in tariffs, such as trans-portation costs, taxes, profitability and vehicle productivity. Ter-avaninthorn and Raballand [46] studied four main internationalcorridors in four African sub-regions (West, Central, East andSouthern Africa) using carrier surveys. In order to analyze the dif-ferences in transportation cost and price structures, the potentialexplaining factors were disaggregated into three categories: thevehicle operating costs, the overall commercial providers' costs andthe transportation prices paid by end users. The authors observethat market organization and strict regulations lead to low trans-portation service quality and high profit margins in West andCentral Africa, while the trucking market is more mature andcompetitive in East Africa. Southern African corridors are found tobe the most efficient in terms of price competitiveness and servicequality.

Lall et al. [22] aim to explain high transportation prices inMalawi, based on trucking surveys designed according to themethodology developed by Teravaninthorn and Raballand [46].They perform two sets of empirical analyses: farmers' costs fortransporting tobacco to markets, and carriers' prices for trans-porting commodities on routes both within the country and onspecific international corridors. Their analyses reveal that boththe infrastructure quality and the market structure of thetrucking industry are important contributors to transport prices.The quality of the trunk road network is not a major constraint,but differences in the quality of the feeder roads that connectvillages to the main roads have a significant bearing on transportprices. These findings are consistent with those of Pedersen [34]who shows that transportation costs depend on the availabilityof infrastructure and on the density of demand on specific links,which implies that Africa is especially disadvantaged with lowand dispersed demand and many landlocked countries that aredifficult to serve. On a macroeconomic level, some studies basedon bilateral trade flow data have shown that the cost of trans-porting goods and getting them across borders is a majorobstacle to African trade performance. Lim~ao and Venables [24]found that the relatively low level of African trade flows islargely due to poor infrastructure, which also affect trans-portation costs. Portugal-Perez and Wilson [35] provide esti-mates suggesting that improvements in logistics are moreimportant in terms of trade expansion than are reductions ofbarrier tariffs.

In contrast to most African markets, the United States (US)transportation market is mature and well-studied, especiallysince the deregulation of the trucking industry in 1980 thatincreased competition. Caplice [5] and Nandiraju and Regan [30]describe the American truckload (TL) transportation market indetail and explain how the procurement of transportation ser-vices is managed in the context of electronic marketplace formats.Combinatorial auction mechanisms, by which carriers can bid onbundles of items for more economically efficient allocations, arecommonly used for TL transportation procurement in practice(see Refs. [6,16,43]). In transportation procurement, the savingsthat can be generated by the use of combinatorial auctions arebased on the principle of economies of scope. Shippers commonlyuse sophisticated transportation management systems to managetheir TL operations, but determining the auction's winners is anNP-hard problem for which optimization algorithms must beimplemented (see Refs. [40,27]; for examples). Powell et al. [36],Lee et al. [23] and Ergun et al. [17] have studied TL operations andpricing strategies from the carrier's perspective. These optimiza-tion problems are rooted in operations research analysis mostlyconducted on randomly generated instances rather than empiricaldata.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219200

The American TL market is well suited for using combinatorialauctions since the North American TL network is dense and infor-mation technology is widely deployed. However, this is not the casein developing countries like Ethiopiawhere backhaul opportunitiesare rare and operations are managed manually. As mentioned byKunaka [21]; one fundamental factor preventing good quality oflogistics infrastructure and services in lagging regions is the lowdemand spread both spatially and temporally, which preventseconomies of scale. Long distances and dispersed economic activ-ities inhibit access to information on market conditions and lead toerratic freight flows. In such a context, tools similar to the ones thathave been developed for less-than-truckload (LTL) services in theUS are more appropriate to better understand the transportationmarket in Ethiopia than those based on combinatorial auctions.Smith et al. [45] and Ozkaya et al. [33] presented a regression-basedmethodology to estimate the LTL market tariffs in the US. Thesestudies aimed to identify the significant drivers of LTL pricing andproduce tariff estimates that can be useful for negotiation andcontracting practices. Smith et al. [45] relied on data from a singleUS motor carrier to construct statistical models, which focus onquantitative factors such as distance, weight and number of ship-ments to explain observed market tariffs. To produce better tariffpredictions, Ozkaya et al. [33] improved this methodology byquantifying intangible qualitative factors through a scorecard pro-cess. They also used an extensive dataset in terms of lane coverageand transaction diversity for freight classes as well as carriere-shipper combinations.

To our knowledge, our study is the first to use observed markettariffs rather than survey data for a developing country. It is also thefirst contribution to provide empirical results on the Ethiopiantransportation market. The methodology used to support ourfindings is similar to that used by Refs. [45,33] for the LTL USmarkets, but it is applied in the context of an emerging country aswas done by Lall et al. [22] in Malawi.

3. The World Food Programme's transportation managementin Ethiopia

The WFP is the United Nations' core agency for food security. Itprovides a multilateral channel for food aid [12] and this humani-tarianorganizationhasdevelopeda strongexpertise in logistics overthe years [42]. The WFP has maintained a continuous presence inEthiopia since 1974, where food aid is used as an instrument foraddressing persistent acute and chronic food insecurity [41]. Ranked174th out of 187 in the 2011 Human Development Index of theUnitedNationsDevelopment Programme,Ethiopia is oneof the leastdeveloped countries in theworld and has sufferedmajor famines inthe past decades [13]. A case study on Ethiopia [47] notes that anestimated 50 percent of the population is food-insecure, some areconsidered chronically food-insecure while others are increasinglyvulnerable at times of droughts. Considerable amounts of food aidhas been delivered in Ethiopia: an average of almost 10% of the do-mestic cereal production from 1984 to 1998, reaching 20% duringbad production years [19]. For 2013, the WFP has assisted about 7.2million people and distributed 846,000 tonnes of food [48]. Asmentioned above, over the past twenty years theWFP has averagedaround22,400TLper yearof food aid shipments, assuming40-tonnetrucks. The transportation and logistical costs are often very highwhen delivering food aid, it usually accounts for most of the reliefoperation costs [29,37]. The constancy of moving large amounts offood aid in Ethiopia combined with the pressure put on humani-tarian organizations for reducing their operational costs have led totransportationmanagement practices that aremore similar to thoseof the business sector rather than those of disaster response, such as

transporting relief to people affected by a tsunami or an earthquakeduring a short time-period following the disaster.

Ethiopia is a landlocked country located between Djibouti,Kenya, Somalia, Sudan and Eritrea in East Africa (see Fig. 1). Thereare four classes of administrative divisions in the country: kebele,woreda, zone and region (ethnically-based regions and governingadministrative cities). The kebeles are the smallest administrativedivisions while the regional divisions are the largest. As in otheremerging nations, the infrastructure in Ethiopia is underdeveloped.Railways are inoperable and less than 20 percent of roadways arepaved. Because of the landlocked geographical location, limitedrailway capacity and costly air transportation in this country,shippers largely depend on ground transportation to move com-modities in Ethiopia. The Ethiopian Government and theWFP shareresponsibility for food aid transportation within the country.Ground transportation has two phases: international and domestic.The WFP Ethiopia's logistics division is responsible for the inter-national transportation movements, which are primarily carried bycommercial providers. The Ethiopian government manages most ofthe domestic transportation movements as well as the food aiddistribution (hand-out to beneficiaries), except in the Somali re-gion, where the WFP has taken more responsibility due to securityconcerns.

3.1. International and domestic transportation operations

The first phase of transportation consists in transporting foodaid from ports to main warehouses, which are known as ExtendedDelivery Points (EDPs). Because Ethiopia is landlocked, trans-portation from seaports necessarily involves internationalmovements. Djibouti is the main port of operation for Ethiopia,and 90 percent of WFP's food aid is routed through it. Because ofdelays at Djibouti, food aid is occasionally received at Berbera andPort Sudan since 2009. In this study, we concentrate our analysison the ports of Djibouti and Berbera for international trans-portation on main corridors. The EDPs are located in strategicregions in the country where primary storage infrastructure isavailable, and act as transshipment hubs for the second trans-portation phase. This second phase of transportation withinEthiopia, from the EDPs to the Final Delivery Points (FDPs), arereferred to as domestic lanes throughout this paper. The FDPs arelocations such as shelters, schools and community centers, wherethe food is temporarily stored before direct distribution to thebeneficiaries at scheduled times, usually once per month. FDPlocations are based on the needs of the beneficiaries and can be inremote areas of the country. Shipment from the EDPs to the FDPsis considered the last-mile transportation in the WFP supplychain. However, the real last mile delivery is borne by the bene-ficiaries who often have to walk several kilometers to collect theirfood and carry it home. For a detailed description of the logisticsactivities associated with the food aid supply chain at the last milelevel and the responsibilities of the main stakeholders involved inthe process, see Rancourt et al. [37].

The international and domestic transportation markets differ inseveral ways. Trucks operating on international corridors, typicallyhaving a 40-tonne capacity, are uncommon on domestic laneswhich tend to use smaller trucks that can handle poor road con-ditions. Larger quantities of food are movedwithin a short period oftime over international corridors, especially when a ship is docked.The loading and offloading operations are usually faster and easierat ports and at EDPs, which are located in major cities with betterinfrastructure as opposed to FDPs. International transportation alsorequires carriers to have special permits to operate in multiplecountries. They can be subject to delays at ports and borders, whichare not common on domestic lanes.

Fig.

1.Map

ofEthiop

ia[26].

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 201

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219202

3.2. Transportation procurement

In Ethiopia, the WFP contracts third-party carriers instead ofrelying on a private fleet. The organization procures transportationservices using a request for quotation (RFQ) mechanism, commoninmany commercial settings, which allows carriers to gain businessthrough competitive bidding. The WFP first sends out question-naires asking about transportation companies' profiles andlaunches tenders only to those carriers whose profiles match theirstandards (core set of carriers). Every six months, the WFP invitesits core set of carriers to submit rate proposals (bids) on lanes thatmay require transportation services. As part of the tender docu-ment in which WFP's requirements are specified, a standardizedquotation template is provided to transporters for them to submittheir rate proposals. The transporters must respond to the RFQbefore the due date or their proposals will not be considered. ThisRFQ mechanism establishes tariffs and specifies the awarded car-riers for each lane. The WFP actively utilizes its contracts on in-ternational corridors. While principally providing backuptransportation capacity for domestic lanes outside of the Somaliregion, the WFP still conducts extensive national RFQs and estab-lishes contracts for the complete domestic transportation networkto supplement government capacity when necessary.

Carriers receive loading orders when the WFP needs to movecargo, and waybills are used to track shipments. Carriers are moni-tored according to the contract and may be withdrawn from WFP'score set of carriers for a certain period of time if they do not performup to the stipulated standards, for example by not meeting deadlinesfor delivery completions or damaging commodities. As in the com-mercial sector, transporters may decline to provide the transportservice even after contracting with the WFP, which will incur sup-plementary costs and lead-times since other available transporterswill have to be found and hired on the spot market. For the benefi-ciaries, this might generate delays and affect the quality of service.

Better understanding the transactional aspects of transportationprocurement is fundamental to assessing the performance of sup-ply chains both in terms of costs and service quality. In order tocoordinate logistics activities between the different stakeholdersinvolved in food distribution, it is essential for an organization likethe WFP to contract with reliable transporters that can ensure on-time dispatches when needed. Thus, our analysis assumes that theWFP contracting process ensures service quality of its core carriers,which comprise a solid basis to study the tariff structure of theEthiopian transportation market.

4. Data sources and variable description

Data from two different sources were gathered to conduct ouranalysis. We have compiled data from a specific RFQ executed by

Table 1Illustrative RFQ data table for one specific origin (orig1).

Lane Tariff Lane characteristics

Origin(port or EDP)

Destination(EDP or woreda)

Tariff(Birr/tonne)

Distances

Total(km)

Paved(km)

Unpa(km)

orig1 dest1 tariff1 dist1 paved1 unpaorig1 dest2 tariff2 dist2 paved2 unpaorig1 dest3 tariff3 dist3 paved3 unpaorig1 dest4 tariff4 dist4 e e

orig1 dest5 tariff5 e e e

… … … … … …

orig1 destn1 tariffn1 distn1 pavedn1 unpa

the WFP in Ethiopia as well as from different reports of the CentralStatistical Agency of Ethiopia (CSAE). In this section, we discussthese data and describe how they form the dependent and inde-pendent variables in the statistical analyses.

4.1. Variables extracted from the data provided by the WFP

Datawere gathered from an RFQ process executed by theWFP inEthiopia. The contracts derived from this specific RFQwere valid forthe planning period from September 2010 to March 2011. The datadescribe carriers who participated in the RFQ, the lanes of the WFPnetwork and the set of bids offered on each lane. Table 1 illustratesthe compiled information of the RFQ for a given origin (orig1).

This data set provides a unique description of the Ethiopiantransportation market. The access to accurate and up-to-date in-formation in Africa is often limited. As mentioned by Teravanin-thorn and Raballand [46]; trucking data collection is largelyinadequate in most countries of Sub-Saharan Africa. The data pro-vided by the WFP were sourced from one shipper that conductsbusiness with several carriers, and this characterizes the supplyside of the transportation market. However, the WFP is one of themost important purchasers of transport services in the country, andwe consider the data to adequately represent the transportationcosts in the market. We make the assumption that the WFP facessimilar costs as other buyers and does not profit from monopsonypower. The variables based on the WFP RFQ data are the effectivetransportation tariffs, the dependent variable of our statisticalmodels, and two categories of independent variables: cost drivenfactors and market structure measures.

4.1.1. Effective transportation tariffsThe dependent variable in our regression models represents the

tariffs in Birr, the unit of currency in Ethiopia, per tonne paid by theWFP from September 2010 to March 2011 for transportation ser-vices on a specific lane. These effective transportation tariffs (tariff)are determined by the WFP through the bidding process (RFQ). Fordomestic transport, the tariff determined by the WFP for a lanecorresponds to the lowest bid made on that lane in the RFQ. Thecarrier who made this lowest offer is awarded the transportationcontract on that lane although there is no commitment to executeshipments when they will occur. For international transportation,since no carrier has the capacity to handle alone all the trans-portation services required on a corridor, a tariff is determined bythe WFP. There is no formal procedure for fixing the effectivetransportation tariffs on the international corridors. The WFP de-termines international tariffs in order to capture enough trans-portation capacity based on their judgement, the received offers inthe RFQ and market prices for transporting commodities such assugar, fertilizers or cement. This tariff is sent as a counter offer to

Carrier rate offers (bids)

Tonnage estimate(Tonne)

Risk(Scale)

Carrier

ved 1 2 … m

(Birr/tonne)

ved1 ton1 risk1 bid11 bid12 … bid1mved2 e risk2 bid21 e … bid2mved3 ton3 risk3 e bid32 … bid3m

ton4 e bid41 bid42 … e

e e bid51 e … bid5m… … … … …

vedn1 tonn1 riskn1 bidn11 bidn12 … bidn1m

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 203

the carriers who participated in the RFQ and will correspond to theeffective transportation tariff during the planning period of theRFQ. The carriers that accept theWFP counter offers and conditionsare awarded of transportation services proportionally to their totaldedicated fleet capacity. The dependent variable in our regressionmodels corresponds to the natural logarithm of these tariffs(ln(tariff)). As explained in Section 5.1, regressing tariff yieldedheteroscedasticity which was eliminated when regressing ln(tariff).

4.1.2. Cost driven factorsThe transportation tariffs should partially reflect the carrier

operating costs. We define as cost driven factors the variablesrecognized to directly affect carrier operating costs. The linehaulcosts are the costs related to freight movements including fuel, tirewear, maintenance, depreciation, driver wages and others [25]. Inparticular, we include lane-specific cost drivers that are unique tothe context of developing countries: measures of road infrastruc-ture quality and perception of operational risks. Using cost drivenfactors to estimate the transportation tariff will thus enable us todetermine the significant variables of the cost-based pricingstructure in Ethiopia.

4.1.2.1. Distance and road conditions. Distance is the primary costfactor in linehaul movements since the main trucking operatingcosts vary directly according to it [14]. However, in developingcountries road quality also matters. At the macro-level, an analysisof bilateral trade data by Lim~ao and Venables [24] shows the in-fluence of road infrastructure on transportation costs. To accountfor road quality, the total length of the lanes was decomposed intotheir paved (paved) and unpaved (unpaved) distances in kilometersusing Geographical Information System (GIS) data. The disaggre-gation into paved and unpaved distances is available for all thelanes of the international corridors, but for only 72 percent of thedomestic transportation lanes. In order to discard as few observa-tions as possible on the domestic transportation market, we havecreated a road condition indicator based on the percentage of un-paved kilometers per lane with the following four categories: good(less than 10 percent unpaved), intermediate (between 10 percentand 70 percent unpaved), poor (70 percent or more unpaved) andunknown. These bounds were chosen in order to ensure that thenumber of observations in each road condition category was almostequally distributed (see Fig. 2 and Tables 7 and 8 of Appendix 1). Tocapture the effect of the total length of the lanes, we also consider acontinuous variable for the total distance (distance) in addition tothe road condition indicator.

02

46

8N

umbe

r of l

anes

0 20 40 60 80 100% of unpaved kilometers

(a) International corridors.

Num

ber o

f lan

es

Fig. 2. Histograms of the corridor and la

4.1.2.2. Risk perception. Chander and Shear [11] suggested thatcarrier rates are likely to reflect the risks involved in providingtransportation services in the Somali region of Ethiopia. A riskperception level was associated to some lanes within the Somaliregion by the WFP (no security issue, low risk or high risk), whilethe other lanes were not graded in terms of risk. The Somaliregion is recognized as the most underdeveloped and conflict-impacted region of Ethiopia. The international corridors with anon-zero specified risk by the WFP were all put in the high levelcategory. It is important to mention that when no risk level hasbeen specified by the WFP, this does not imply that the risk iszero but only that the information is unknown. The distributionof the risk perception variables is shown in Tables 7 and 8 ofAppendix 1.

4.1.2.3. Tonnage estimates. In TL services, transporting largerquantities on a regular basis often consist of valuable businessopportunities for the carriers. A higher estimated tonnage to betransported per month can thus lead to lower unit tariffs throughvolume discounts. In the RFQ, the WFP provides estimates of themonthly tonnages to be transported for 66 percent of the inter-national corridors and 18 percent of the domestic lanes. These es-timates are based on historical data, projected demands andrequirements. However, the lack of an estimate does not imply thatno transportation service will be required on the lane, nor does agiven estimate guarantee business on the lane. Carriers are invitedto bid on all lanes considering the information provided. We haveincluded estimated tonnage indicators in order to test for potentialvolume discounts reflected in transportation tariffs. Because of thelarge number of unspecified estimates for this variable and in orderto be able to use the available information on tonnage in theregression model, we have generated an indicator for the estimatedtonnage to be transported monthly. The indicator variable desig-nate low or high tonnage estimates depending on a threshold t,with t ¼ 20;000 tonnes/month for the international corridors andt ¼ 1;000 tonnes/month for the domestic lanes. The thresholdswere determined to ensure that the number of observations in eachcategory was fairly distributed, see Fig. 3 and Tables 7 and 8 ofAppendix 1.

4.1.3. Market structure measuresThe second group of explanatory variables concerns the struc-

ture of the trucking market. Truckload transportation markets indeveloping countries such as Ethiopia are unlikely to resemble thecompetitive markets found in developed countries. If competition

050

100

150

200

0 20 40 60 80 100% of unpaved kilometers

(b) Domestic lanes.

ne unpaved kilometer percentages.

02

46

8N

umbe

r of l

anes

0 50 100 150Monthly tonnage estimates (1,000 tonne/month)

(a) International corridors.

020

4060

80N

umbe

r of l

anes

0 5 10 15 20Monthly tonnage estimates (1,000 tonne/month)

(b) Domestic lanes.

Fig. 3. Histograms of monthly tonnage estimates by transportation network segment.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219204

is low, monopoly power will allow carriers to include a highermarkup over marginal cost in their pricing structure. Exploiting theunique nature of our data, we use information from the bid dis-tribution as a proxy to measure the impact of market structure ontransportation tariffs. To this end, proxies associated with compe-tition intensity, market dispersion and market concentration wereincluded in the regression models. These variables were computedfrom information extracted from the internal supply chain of theWFP derived from the RFQ.

4.1.3.1. Competition intensity. We hypothesize that the competitionintensity on a network segment is related to the number of bidsmade on the corresponding international corridor or domestic lane.Fig. 4(a) depicts the tendency for the natural logarithm of tariffs perkilometer per tonne to decline as the number of carriers bidding ona lane increases. It also shows this relationship to be non-linear:each additional bidder reduces the tariffs, but this effect de-creases with the number of bidders. Thus, competition appears toconsiderably affect transportation tariffs, but adding players in thebidding process becomes less influential when the competitionintensity has reached a certain level. This non-linear relationship isestimated in the regression models by including the natural

45

67

8ln

(tarif

f) pe

r kilo

met

er (B

irr/M

T/km

)

0 20 40 60Number of bids

International corridor Domestic lane

(a) Effect of the number of bids on ln tariff ratesper kilometer.

Fig. 4. Tariffs per kilometer as functi

logarithm of the number of bids per lane, ln(#bids), as a proxy forcompetition intensity. Fig. 4(b) shows the relationship between thelogarithm of the transportation tariffs and the logarithm of thenumber of bids to be linear.

4.1.3.2. Market dispersion. In a mature and efficient market, infor-mation about prices should be known by all participating actors,and the dispersion in bids for each transportation network segmentshould be small. We hypothesize that a large dispersion in bidsreveals a disorganized market in which information does notcirculate well between the carriers. Therefore, the difference be-tween the third and the first quartiles of the bids on each lane havebeen computed to measure the variation in bids without consid-ering the extreme bids. To standardize this measure, the differencebetween the third and the first quartiles has been divided by themedian. The ensuing variable is called bid dispersion.

4.1.3.3. Market concentration. To reflect the market concentrationat the shipping origins and destinations, we have created two typesof predictors: the number of active carriers and bid-related Her-findahl-Hirschmann indices (HHI). The number of active carriers(carrier) is the number of carriers that are willing to provide

45

67

8ln

(tarif

f) pe

r kilo

met

er (B

irr/M

T/km

)

1 2 3 4ln(number of bids)

International corridor Domestic lane

(b) Effect of ln(number of bids) on ln tariff ratesper kilometer.

on of the competition intensity.

Table 2Descriptive statistics.

Market International market Domestic market

Network 2 origins (ports) 33 origins (EDPs)24 destinations (EDPs) 98 destinations (FDPs)32 corridors 731 lanes46 carriers 59 carriers

Descriptive statistics n Mean Standarddeviation

n Mean Standarddeviation

Distance (km) 32 756.1 289.7 731 589.8 356.3Estimated tonnage per

lane (tonne/month)21 27,426.9 36,969.3 131 2055.4 3028.2

Number of bids per lane 32 17.9 7.3 731 14.9 10.3Number of bids per carrier 46 14.4 5.5 59 212.3 190.5Percent of winning bids

per carrier46 39.0 37.6 59 8.4 12.1

Bids (Birr/tonne) 572 1317.6 1016.1 12,231 1322.1 1286.7Tariff (Birr/tonne) 32 1405.3 902.5 731 1138.8 844.6Tariff per km

(Birr/tonne km)32 2.0 1.2 731 2.4 2.1

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 205

services from an origin or to a destination, that is the number ofdistinct carriers who have bid on a lane originating from a specificorigin or ending at a specific destination in the RFQ. The Herfindahlindex is equal to the sum of the squared market share st of carrier t(t ¼ 1,…,m): hhi ¼ Pm

t¼1s2t . Lacking information on carriers' market

shares, we have used the number of bids by carriers as a proxy. Letbt be the number of bids made by carrier t, and let B be the totalnumber of bids that have been made at the origin or at the desti-nation. Market shares st are estimated by bt/B. The Herfindahl indexmeasures the amount of market power from which firms profitfrom. An industry with few large competitors will have a high levelof concentration (hhi close to one), while a fragmentedmarket withmany small competitors results in low concentration (hhi close tozero).

4.2. Variables extracted from data published by the CentralStatistical Agency of Ethiopia

The last group of explanatory variables consists of regional de-mographic and economic factors. These data are external to theWFP supply chain, so they can be used in conjunction with the costdriven factors to predict transportation tariffs when extensive RFQdata are not available or not representative of the global market.This situation could also occur for predicting the transportationtariff of a new lane without information on the competition envi-ronment. Thus, the regional geographic concentration of popula-tion and economic activities at the origins and at the destinationscould be used as a substitute for market predictors. Note that thereare 11 administrative regions in Ethiopia: Addis Ababa, Affar,Amhara, Benishangul-Gumuz, Dire Dawa, Gambella, Harari, Oro-mia, SNNPR (Southern Nations, Nationalities, and Peoples' Region),Somali and Tigray.

Data on economic activities were compiled from the latestreports published by the Central Statistical Agency of Ethiopia(CSAE) at the time of data collection. The CSAE is a governmentagency designated to supervise and provide all survey and censusreports in Ethiopia in order to monitor economic and social in-dicators. For the 11 regional states of Ethiopia, we have extracteddata on the following three categories: the number of factories ofthe major industrial groups [9], the agricultural production [8],and major livestock types [7]. We have also extracted data on thepopulation from the latest Population and Housing Census ofEthiopia conducted by the Ethiopian Population Census Com-mission [35]. We use the socio-economic data at the regionaladministrative level because this information is not available forthe smallest administrative levels such as woreda or kebele. Notethat there are no regional socio-economic indicators associatedwith the origins of the international corridors since these corre-spond to foreign ports.

The compiled socio-economic statistics for Ethiopia are pre-sented in Table 5 of Appendix 1. Brief descriptions of the consideredelements are given in the first column of this table, while theirvalues together with their relative percentages are presented in thelast two columns. Although we have used data at the regional levelin the regression models, note that we only provide the statisticsapplicable to the whole country. Information at the regional levelcan be found in the CSAE public reports. The variable populationrepresents the total population, and the variable factory corre-sponds to the total number of factories (the sum of all types offactories) in the region of origin or destination. The variables grain,vegetable and coffee are the total quintals of crop grains, vegetableand coffee produced in all regions. The variable agriculture is thetotal output of agriculture production, which corresponds to thesum of the three previous variables. The variables cattle, sheep, goatare the number of registered animals of each type. The variable

livestock is the total number of animals (the sum of the number ofcattle, sheep and goats). Note that for the agricultural productionand the livestock, we have only considered the three more impor-tant and significant elements regarding the Ethiopian actual eco-nomic situation.

Table 6 of Appendix 1 shows the correlation coefficients be-tween socio-economic variables. We note that all the agriculturalproduction variables (grain, vegetable, coffee and agriculture) andall the livestock variables (cattle, sheep and goat) are stronglycorrelated to each other. This implies that the different agricul-tural activities are concentrated in the same regions. Moreover,when comparing the correlation coefficients of agriculture, live-stock and population, we observe that these three variables arepositively correlated. This indicates that the more populated re-gions are also more active in terms of agricultural production,which is not necessarily the case for the manufacturing activitieshaving no strong significant correlation with population noragriculture. Indeed, more than 85 percent of the Ethiopian popu-lation lives in rural areas where agriculture is the main economicactivity.

4.3. Descriptive statistics

Table 2 contains descriptive statistics for the main variables ofinterest for the two markets. This table clearly shows that the in-ternational transportation network is denser than the domesticnetwork, with fewer lanes to cover but more tonnage to transportper lane. The average distance of the international corridors isgreater than that of the domestic lanes, but both domestic andinternational markets are considered to be long-haul trans-portation. The last row of Table 2 shows that the tariffs per kilo-meter are lower on average for the international corridors than forthe domestic lanes. Moreover, the standard deviation of the tariffsper kilometer is smaller for the international corridors than for thedomestic lanes. The international market seems more competitivesince it generates on average more bids per lane than the domesticmarket. Note that in the remainder of this paper the information onthe carriers is not used to explain the effective transportation tar-iffs, but only to analyze the bids. More detailed descriptive statisticsabout the WFP data for all variables in both the international anddomestic markets are reported in Tables 7 and 8 of Appendix 1.These tables contain information describing the distribution ofeach variable including the mean, median, standard deviation, aswell as the minimum and maximum values.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219206

5. Results

We have used multiple linear regression models to determinewhich of the factors have a significant impact on transportationtariffs in Ethiopia. Separate analyses are conducted for the inter-national and domestic Ethiopian transportation markets since, asdiscussed above, these are two distinct markets. We now explainthe process implemented to determine our best regression modelsbefore presenting and describing them. Note that all statisticalanalyses were executed using Stata 11.

5.1. Methodology

For each transportation market, we have constructed twodistinct type of models which both include the cost driven factors(C): onewith themarket structure measures (M) and the other withregional socio-economic variables (S). There are different reasonsfor this choice. The socio-economic variables are derived frompublic information independent of the WFP operations. Becausethe socio-economic variables partly explain the market structure,they can thus be considered as proxies for it when specific data,such as the WFP data in this case, are not available. These twomodels can be compared to better understand the effects of themarket structure on transportation tariffs, measured from the WFPinformation or from socio-economic statistics. For the internationaland the domestic transportation markets we aim to definemultiplelinear regression models explaining ln(tariff), that isln(tariff) ¼ f1(C,M) þ ε1 and ln(tariff) ¼ f2(C,S) þ ε2, where ε1 and ε2are random errors due to unobservable factors. Heteroscedasticityhas been identified with Breusch-Pagan/Cook-Weisberg and Whitetests when using tariff as dependent variable. This problem isremoved by using ln(tariff), see Appendix 2.

We have constructed various model specifications by usingdifferent sets of independent variables. For each market, we havefirst regressed ln(tariff) on the variables related to the road attri-butes and we have then added the competition intensity to thisspecification. We now explain how the variables were selected tobuild our final regression models. For the models with the marketstructure measures, the HHI indices (hhi) were omitted becausethey were highly correlated with other market structure measures(see Tables 9 and 10 of Appendix 1) and were not effective pre-dictors. For the models with the regional socio-economic vari-ables, the variables associated with agricultural activities(agriculture and livestock) at the origin and destination were dis-carded because of their correlation with the population (popula-tion). Only the remaining suitable variables were used into thestatistical models ln(tariff) ¼ f1(C,M) þ ε1 andln(tariff) ¼ f2(C,S) þ ε2. Lastly, backward regressions allowed us todetermine our final statistical models for the (1) international and(2) domestic transportation markets.

5.2. Multiple regression models

Tables 3 and 4 display the results obtained for six differentspecifications for the international corridors and for the domesticlanes. In each table, the first specification corresponds to aregression of the natural logarithm of transportation tariffs(ln(tariff)) on variables associated with the distance and road con-ditions measures only. The second specification adds variablesdescribing competition intensity. The third specification considersall the explanatory variables, except for the regional socio-economic factors, and the fourth specification results from abackward regression which initially considered the same inde-pendent variables. The last two specifications include the regionalpopulation and the manufacturing activity variables instead of the

market structure measures. The fifth specification includes all theexplaining variables, and the sixth model was obtained by applyinga backward regression.

The number of observation considered (n), the R-squared, theadjusted R-squared and the maximum variance inflation factor(VIF) values are given for each specification presented in Tables 3and 4. VIF statistics were estimated for the different models todetect multi-collinearity issues. There is no theoretical thresholdvalue for recognizing multi-collinearity, but a common rule ofthumb suggests that variables with a VIF value greater than five orten should be discarded, five being more restrictive than ten [32].The highest VIF values obtained are 3.36 and 3.07 for our final in-ternational and domestic models, respectively. As suggested byBerry and Feldman [4], we provide the robust standard errorsassociated with the regression coefficients in Tables 3 and 4. Foreach model, in order to validate the coefficients, we ensured thattheir signs and their magnitudes were the same as those obtainedwhen regressing the transportation tariff (ln(tariff)) on each indi-vidual variable separately. We thus consider that there are nomulti-collinearity problems.

From the results presented in Tables 3 and 4, we note thatincluding the market structure measures as explanatory factorsleads to statistical models that explain more variability intransportation tariffs than those including the socio-economicfactors, this being true for both international and domesticmarkets. The adjusted R-squared associated with Models 1.4 and2.4 are indeed 21.7 percent and 8.9 percent higher than those ofModels 1.6 and 2.6. In the remainder of this paper, we willtherefore consider Models 1.4 and 2.4 presented in Tables 3 and 4as our best models to explain the international and domestictransportation markets.

5.3. Insights

In this section, we highlight the significant determinants oftransportation tariffs and quantify their impact. We aim todiscuss the ability of the different variables to predict Ethiopiantransportation tariffs by putting a particular emphasis on factorswhich are specific to emerging markets. We first look at the costdriven factors and then at the measures associated with marketstructure before discussing the results of the models consideringthe socio-economic variables. Note that [20] suggested thefollowing estimator for the percentage impact of a dummy vari-able on the variable being explained in semi-logarithmic equa-tions: 100%½expðbb � 1=2bV ðbbÞÞ � 1� ; with bb being the dummyvariable coefficient and bV ðbbÞ being an estimate of the variance forbb. We will thus use this estimator, with bV ðbbÞ equal to the squaredrobust error of a coefficient b, in order to analyze the impact ofour indicator variables on tariff. For a continuous independentvariable, 100bb can be interpreted as the percentage change intariffs due to a unit increase in this variable, holding all otherindependent variables constant. However, if the continuous in-dependent variable have been log-transformed, bb is referred to asan elasticity.

5.3.1. Analysis of the linehaul cost driversIn linehaul movements, distance is recognized to be an important

contributor to the cost of doing business for freight carriers [39]. Fig. 5illustrates the relationship between the natural logarithm of trans-portation tariffs anddistance. Thisfigure reveals that, despiteacertainlinear tendency, a large amount of variability in tariffs remains. Thetransaction prices are not tightly concentrated along the fitted lineand distance alone only explains 9 percent and 27 percent of thevariability in tariffs for international and domestic transportation.This is a striking fact, especially considering that transport prices are

Table 3Regression models for ln(tariff) (Birr/tonne) for the international transportation market.

Variables Models

(1.1) (1.2) (1.3) (1.4)a (1.5) (1.6)b

Cost drivenRoad conditionspaved (km) 0.000137

(0.000335)0.000580***(0.000180)

0.000392**(0.000184)

0.000497**(0.000194)

0.000874**(0.000352)

0.000859**(0.000338)

unpaved (km) 0.00160***(0.000408)

0.00103**(0.000389)

0.00109***(0.000302)

0.00116***(0.000318)

0.000940***(0.000310)

0.000927***(0.000299)

Risk perceptionc

high 0.552***(0.170)

0.607***(0.161)

0.549**(0.262)

0.590**(0.251)

not specified 0.0131(0.108)

0.0604(0.119)

0.290(0.190)

0.325*(0.175)

Tonnage estimatesd

low �0.102(0.128)

�0.0816(0.168)

�0.0954(0.162)

high �0.227(0.152)

�0.378*(0.191)

�0.379*(0.187)

Market structureCompetition intensitylnð#bidsÞ �0.823***

(0.0863)�0.881***(0.0809)

�0.888***(0.105)

Market dispersionbid dispersion �1.662***

(0.406)�1.510***(0.282)

Market concentrationcarrier destination �0.00335

(0.00307)�0.00525*(0.00272)

Socio-economicpopulation destination �4.46e-05

(7.71e-05)factory destination �0.000847**

(0.000345)�0.000868**(0.000416)

Constant 6.618***(0.250)

8.829***(0.328)

9.661***(0.398)

9.561***(0.453)

6.550***(0.341)

6.508***(0.330)

n 32 32 32 32 32 32R-squared 0.442 0.738 0.890 0.878 0.713 0.710Adj. R-squared 0.404 0.710 0.845 0.842 0.612 0.625Max. VIF value 1.19 1.53 3.71 3.31 3.52 3.36

Robust standard errors in parentheses, ***p < 0.01, **p < 0.05 and * p < 0.1.a This model corresponds to Model 1.3 after backward elimination of non-significant variables.b This model corresponds to Model 1.5 after backward elimination of non-significant variables.c The category of reference is “not specified” for regions other than Somali.d The category of reference is “not specified”.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 207

known to increase almost directly proportionally to distance inmature transportationmarkets. InNorthAmerica for instance, afixedprice and a constant variable price per kilometer alone explain about80e85 percent of the variability in tariffs [10]. Thus, the need to takeinto account factors other than distance to explain the Ethiopiantransportationmarket is obvious. In Fig. 5, the international corridorsand the domestic lanes where the number of bids is larger than theaverage (16.3 for corridors and 11.6 for lanes) are distinguished withdarkgreydots.Wenote that the transportation tariffs tend tobe loweron these more competitive network segments.

The first specifications presented in Tables 3 and 4 emphasizethe importance of incorporating road conditions to better explaintransportation tariffs. When comparing the adjusted R-squared ofModels 1.1 with the simple univariate model displayed in Fig. 5, wenote that 31.5 percent more variability in tariffs can be explainedonly by decomposing the total distance of the lanes into their pavedand unpaved distances for the international corridors. We can alsoobserve in Model 1.4 that the coefficient of unpaved kilometer ishighly significant while the coefficient of paved kilometer is lesssignificant. This indicates that carriers tend to set their tariffs pro-portionally to the unpaved distance. A one-unit increase in unpavedand paved distance increases tariffs by 0.116 percent and 0.0497

percent on average when controlling for significant factorsincluding the market structure, as in Model 1.4. For the domesticmarket, we note that adding road condition to distance in theregression model increases its predicting power by 34.1 percent(comparing the adjusted R-squared of Models 2.1 and the one of thebasic regression shown in Fig. 5). We can also observe from thestatistical models of Table 4 that the transportation cost on thelanes with poor road conditions is much more expensive than onthe lanes with intermediate or good conditions. Indeed, based onthe coefficients of Model 2.4, and compared with lanes withoutavailable information on road conditions, poor road conditionsincrease the tariffs by 22.4 percent, while good conditions decreasethe tariff by 13.9 percent.

Carriers operating in the Somali region are vulnerable to variousrisks such as hijacking or delays caused by escorting procedures orexpected disruptions. It is likely that they will take these factorsinto account in their tariff structures. The coefficients of riskperception in Models 1.4 and 2.4 confirm that the lanes with ahigher or unknown risk have significantly greater tariffs. For theinternational market, the transportation tariffs are 81.1 percenthigher on corridors in which risk is specified. For the domesticlanes, the transportation tariffs are not significantly lower when it

Table 4Regression models for ln(tariff) (Birr/tonne) for the domestic transportation market.

Variables Models

(2.1) (2.2) (2.3) (2.4)a (2.5) (2.6)b

Cost drivendistance (km) 0.00164***

(7.32e-05)0.00123***(6.42e-05)

0.00109***(6.73e-05)

0.00112***(6.14e-05)

0.00138***(7.61e-05)

0.00138***(7.54e-05)

Road conditionsc

poor 1.102***(0.0685)

0.258***(0.0687)

0.219***(0.0616)

0.204***(0.0604)

0.441***(0.0783)

0.441***(0.0781)

intermediate 0.310***(0.0701)

�0.0573(0.0583)

�0.0761(0.0539)

�0.0838(0.0535)

�0.0867(0.0658)

�0.0867(0.0658)

good �0.325***(0.0634)

�0.256***(0.0517)

�0.137**(0.0545)

�0.149***(0.0495)

�0.376***(0.0632)

�0.378***(0.0631)

Risk perceptiond

none �0.112(0.109)

�0.145(0.0999)

�0.0507(0.139)

�0.0478(0.136)

low 0.468***(0.0799)

0.431***(0.0771)

0.575***(0.0969)

0.577***(0.0943)

high 0.647***(0.0725)

0.644***(0.0701)

0.611***(0.0977)

0.613***(0.0957)

not specified 0.216***(0.0574)

0.232***(0.0527)

0.349***(0.0647)

0.351***(0.0626)

Tonnage estimatese

low �0.0213(0.0622)

�0.0967(0.0799)

�0.0961(0.0800)

high �0.111(0.0717)

�0.310***(0.0773)

�0.310***(0.0774)

Market structureCompetition intensitylnð#bidsÞ �0.717***

(0.0359)�0.605***(0.0610)

�0.564***(0.0402)

Market dispersionbid dispersion �0.724***

(0.0924)�0.718***(0.0891)

OriginMarket concentrationcarrier origin 0.00269

(0.00248)carrier destination �0.00530***

(0.00110)�0.00557***(0.00107)

Socio-economicpopulation origin �6.08e-05**

(2.39e-05)�6.10e-05**(2.39e-05)

population destination �7.72e-06(3.32e-05)

factory origin �0.000616***(0.000118)

�0.000617***(0.000118)

factory destination �0.000831***(0.000138)

�0.000848***(0.000144)

Constant 5.471***(0.0716)

7.737***(0.120)

7.855***(0.158)

7.806***(0.140)

5.977***(0.116)

5.971***(0.107)

n 731 731 731 731 731 731R-squared 0.564 0.727 0.780 0.778 0.689 0.689Adj. R-squared 0.561 0.725 0.775 0.775 0.683 0.683Max. VIF value 1.38 2.23 6.11 3.07 2.82 2.77

Robust standard errors in parentheses, ***p < 0.01, **p < 0.05 and * p < 0.1.a This model corresponds to Model 2.3 after backward elimination of non-significant variables.b This model corresponds to Model 2.5 after backward elimination of non-significant variables.c The category of reference is “unknown”.d The category of reference is “not specified” for regions other than Somali.e The category of reference is “not specified”.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219208

is specified that the lanes are not risky. For low and high specifiedrisk, the transportation tariffs are 53.4 percent and 89.9 percenthigher, compared to lanes with non-specified risk for regions otherthan Somali. Transportation in the Somali region also tends to in-crease the transportation tariff significantly. Even if the risk is notspecified, it is 25.9 percent more expensive to transport food aidwithin the Somali region than on the other domestic lanes.

The indicator variables associated with the estimated quantity offood aid to be transported monthly (low and high tonnage withmissing as the reference category) are not significant at the 90

percent level in Models 1.3 and 2.3 and do not appear in Models 1.4and 2.4. Even by regressing on a smaller sample, where only thelanes with a specified tonnage value are taken into account, theestimated tonnage does not significantly explain transportationtariffs, and the regression coefficients remain similar to those ofModels 1.3 and 2.3. When the socio-economic variables are includedin the model instead of the market structure measures, the regres-sion coefficient for high tonnage becomes statistically significant(Models 1.5, 1.6, 2.5 and 2.6). However, when we remove tonnagefromModels 1.6 and 2.6, the adjusted R-squared value decreases only

66.

57

7.5

88.

5ln

(tarif

f) (B

irr/M

T)

0 500 1000 1500 2000Distance (km)

International tariff Fit line (Adj. R−Squared = 0.089) Competitive corridor

(a) International corridors

45

67

8ln

(tarif

f) (B

irr/to

nne)

0 500 1000 1500 2000Distance (km)

Domestic tariff Fit line (Adj. R−Squared = 0.269) Competitive lane

(b) Domestic lanes

Fig. 5. Natural logarithm of transportation tariffs as a function of distance.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 209

by a small amount (from 0.625 to 0.587 for international corridorsand from 0.682 to 0.677 for domestic lanes). This implies that, withrespect to estimated tonnage datawe have, there is no clear evidenceof volume discount for transportation services in Ethiopia whencontrolling for the other variables in the model, namely cost drivenfactors and market structure measures or socio-economic variables.This is surprising since decreasing unit transport costs for grains cangenerally be expected [39] and larger regional volumes of tradeusually correspond to lower freight tariffs [44].

To further test for volume discount at the level of carriers, we alsoconsider the bid as a dependent variable. We add indicator dummiesper carrier to uniquely identify the carriers across lanes, allowingcontrol for any carrier-specific characteristics that might influencetheir bids. The international and domestic specifications obtainedfrom this regression are presented in Table 11 of Appendix 3. Resultsshow that the tonnage indicators have a statistically significantimpact in lowering the bids, regardless of whether the estimate islow or high. Carriers perhaps bid more competitively when tonnageestimates are provided, regardless of the volume level, since even alow estimate indicates some degree of certainty about volumes,although it is not a tonnage guarantee by the WFP. The decrease inbids when the estimated tonnage moves from low to high for bothmarkets indicates that this aspect of bidding behavior could berelated to volume discount. There is therefore some evidenceregarding the influence of the estimated tonnage at the level ofcarriers among these sets of bids, but the link is relatively weak.

0.331

0.206

0.131

0.0820.061

0.037 0.031

0.0

0.1

0.2

0.3

0.4

DA

ove

rall

aver

aged

add

ition

al c

ontri

butio

n

ln(# b

id)

unpa

ved d

istan

ce

high r

isk

bid di

spers

ion

desti

natio

n carr

ier

not s

pecif

ied ris

k

pave

d dist

ance

(a) International market, Model 1.4.

Fig. 6. Dominance analysis overall additional contributio

5.3.2. Analysis of the market structureBeyond carrier cost structures and economies, transportation

tariffs may also be affected by the level of competition. Simple uni-variate regressions reveal that the number of carriers submittingrate offers in theRFQ for a specific lanehas an impact on the effectivemarket tariff. A regression of the transportation tariffs on ln(#bids)and a constant explains about 55 percent of the variability in tariffsfor both the international corridors and the domestic lanes, whichsuggests that competition intensity has a considerable influence ontransportation tariffs. Models 1.4 and 2.4 further indicate that theintensity of competition, measured by ln(#bids), has a significantimpact on transportation tariffs for both international and domestictransportation markets. Indeed, a one percent increase in #bids, allother variables being held constant, generates a 0.888 percent and0.564 percent decreases (elasticities) in the transportation tariff forthe international corridors and domestic lanes. The importance andconsequences of the competition intensity in the transportationmarket will be further analyzed in the following sub-section.

An alternate way to test the importance of competition is toregress not only the effective transportation tariffs but also thecomplete distribution of bids. Since we are able to uniquely identifythe carriers across lanes, we can add indicator dummies per carrierto control for systematic differences in the average bid submittedby each carrier. Results from this regression show that the numberof bids per lane also has a statistically significant and negativeimpact on the bids. The intensity of competition not only has an

0.249

0.202

0.101

0.069

0.0520.036 0.029 0.027

0.007 0.006

0.0

0.1

0.1

0.1

0.2

0.3

DA

ove

rall

aver

aged

add

ition

al c

ontri

butio

n

ln(# b

id)

total

distan

ce

not s

pecif

ied ris

k

desti

natio

n carr

ier

poor

road

bid di

spers

ion

good

road

high r

isk

interm

ediat

e roa

d

low ris

k

(b) Domestic market, Model 2.4.

ns for the models considering the market measures.

0.229

0.160

0.128

0.096

0.045 0.040

0.012

0.0

0.1

0.1

0.1

0.2

0.3

DA

ove

rall

aver

aged

add

ition

al c

ontri

butio

n

unpa

ved d

istan

ce

high t

onna

ge

high r

isk

desti

natio

n man

ufactu

re

pave

d dist

ance

not s

pecif

ied ris

k

low to

nnag

e

(a) International market, Model 1.6.

0.242

0.109 0.103

0.0620.047 0.041

0.026 0.026 0.0230.008

0.0

0.1

0.1

0.1

0.2

0.3

DA

ove

rall

aver

aged

add

ition

al c

ontri

butio

n

total

distan

ce

not s

pecif

ied ris

k

desti

natio

n man

ufactu

re

poor

road

good

road

origin

man

ufactu

re

high t

onna

ge

high r

isk

origin

popu

lation

low ris

k

(b) Domestic market, Model 2.6.

Fig. 7. Dominance analysis overall additional contributions for the models considering the socio-economic factors.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219210

impact on the effective transportation tariffs but also on the carrierbids in general, see Table 11 of Appendix 3.

From Models 1.4 and 2.4, we also note that transportation tariffstend to decrease when there are more active carriers at destination(variable carrier destination for the corridor and lane destinations), forboth international and domestic markets. The number of active car-riers at a location is an indicator of business attractiveness, andtransportation tariffs tend tobe lower fora laneendingatanattractivedestination. Although the international data do not have an indicatorvariable for the number of carriers at the origin, the coefficient of thenumber of carriers at the lane origin (carrier) is not significant for thedomestic market. Behrens and Picard [3] show that the opportunitycosts of returning empty increase freight rates, and that transportcosts should be higher from destinations that are net exporters thanfrom regions that are net importers. Our findings can be reconciledwith this theory bymaking the hypothesis that carriers tend to locatein net exporting regions rather than in net importing regions. How-ever, given the limited data, this hypothesis cannot be tested. If true, arelatively higher number of carriers at origin indicates a net exportinglocation, for which there will exist relatively fewer backhaul oppor-tunities. In addition, a destination locationwithmore carriers is a netexporter, with more business activities and lower tariffs, which iswhat we observe.

010

0020

0030

0040

00

Tarif

f (B

irr/M

T)

0 500 1000 1500 2000Distance (km)

Observed Fitted with all roads paved

(a) Predicted tariffs if the network segments were

all paved.

Fig. 8. Illustration of the predicted tariffs in “what if” scenario

We also observe inModels 1.4 and 2.4 that the coefficients of thestandardized dispersion in bids (bid dispersion) are negative andsignificant, which implies that the disparity between bids leads tolower transportation tariffs and that prices are not necessarilyhigher in disorganized markets. Having a wide dispersion betweenbids on a lane increases the chance of having a lower bid, resultingin a lower effective transportation tariff. This may mean that somecarriers are bidding below the cost drivers to be competitive or thatthey do not understand their cost structure well. Informal discus-sions with shippers in Ethiopia indicated that they feel that somecarriers are bidding at levels that do not support their operations,which may provide short term cost savings but not sustainablecapacity. Another explanation is that this could constitute an evi-dence for collusion. If carriers collude to set prices on certain lanesas in a cartel, there will be less variability in bids on those lanes, andthe effective transportation tariff will be higher. However, furtheranalyses would need to be conducted in order to identify cartelsand analyze their impact.

In order to compare the relative importance of the cost driven andthemarket structure predictors of Model 1.4 andModel 2.4, we haveimplemented the dominance analysis (DA) proposed by Azen andBudescu [1]. These authors have introduced several quantitativemeasures of dominance for predictors in multiple regression, based

010

0020

0030

0040

00

Tarif

f (B

irr/M

T)

0 500 1000 1500 2000Distance (km)

Observed Fitted with more competition

(b) Predicted tariffs if roads the network seg-

ments more competitive.

s for the international corridors and the domestic lanes.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 211

on an examination of the R-squared values for all possible subsetmodels of the independent variables. Fig. 6 illustrates the overallaveraged additional contribution to the proportion of variance inln(tariff) accounted for by each predictor in the subset models. Whenthe DA overall averaged additional contribution of a predictor isgreater than another, this predictor is said to generally dominate itand, therefore, we can state that this predictor is relatively moreimportant than the other. These values thus provide an idea on themagnitude of importance of the different explaining factors consid-ered in our models. For the international corridors (Fig. 6(a)), thecompetition intensity and the unpaved distance appear to be themajor determinants of transportation tariffs, followed by a high riskperception to go in Somali. For the domestic lanes (Fig. 6(b)),competition intensity appears to be the dominant determinant oftariffs, followed by distance. Providing transportation services in theregion of Somali and the number of active carriers at destination alsohave an impact on the tariffs, but their importance drops significantlycompared with distance. The results of Models 1.4 and 2.4, as well asthe bar charts of Fig. 6, show the importance of the competition in-tensity on the transportation tariffs in Ethiopia. Note that for the in-ternational corridors, the competition intensity completelydominates all the other predictors [1].

5.3.3. Analysis of the socio-economic factorsComparing the adjusted R-squared values of Models 1.4 and 2.4

with the ones of Models 1.6 and 2.6, we observe that the socio-economic variables do not explain the variability in tariffs as wellas the market structure measures. However, they still provide verygood results while making use of variables derived from publicinformation to measure economic activities, independent of theWFP operations, which justifies the relevance of such models. Asabove, in order to assess the relative importance of the combinedcost and socio-economic factors in Models 1.6 and 2.6, we haveconsidered the DA overall averaged additional contribution of eachpredictor shown in Fig. 7. Unpaved distance is the most importantexplaining factor for the transportation tariffs, followed by hightonnage and risk for the international corridors (Fig. 7(a)). Whenusing the socio-economic variables instead of the market structuremeasures, the high estimated tonnage indicator becomes a signif-icant variable, but as mentioned above, its contribution toexplaining tariff is small. The paved distance remains one of theleast important predictor of transportation tariffs on the interna-tional corridors. For the domestic lanes (Fig. 7(b)), distance has amajor impact on the transportation tariffs, followed by non-

010

0020

0030

0040

00

Tarif

f (B

irr/to

nne)

0 500 1000 1500Distance (km)

International tariff Potential outlier

(a) International corridors

Tarif

f (B

irr/to

nne)

Fig. 9. Illustration of the WFP potentia

specified risk in the region of Somali and the number of factoriesat destination. For both markets, international and domestic, re-gionswithmoremanufacturing activities appear to be attractive forthe carriers, which is reflected by a negative impact on trans-portation tariffs, indicating that the manufacturing activities couldlead to more competitive transportation markets. As when con-trolling for the market structure, transporting food aid within theSomali region of Ethiopia also has an important impact on thetariffs when controlling for the socio-economic factors.

5.4. Discussions and implications

We now show how our models can be used to compute coun-terfactual costs and determine potential outliers. Assessing coun-terfactual costswill enable us to evaluate “what if” scenarios such asthe consequences on transportation tariffs of improving road con-ditions or increasing competition in Ethiopia. These counterfactualsare done using the simple reduced-form estimates and do notconsider the dependence with some of the other independent var-iables. Theyare therefore not predictors for long-termdevelopment,but can assist short-term scenario analyses. We also identify po-tential outliers to support the WFP by pointing to transportationnetwork segments where tariffs might need to be reviewed.

5.4.1. Counterfactuals and development policiesUsing the regression models 1.4 and 2.4, we can predict what

would be the transportation service tariffs when improving theroad conditions or when increasing the competition intensity. Wehave therefore estimated the total transportation costs incurred byunpaved roads and lack of competition. To account for better roadconditions, we have computed the predicted transportation tariffsas if all the international corridors were paved and as if all thedomestic lanes were in good condition. To account for morecompetition intensity, we have computed the predicted trans-portation tariffs if the competition intensity variable values werereplaced by themaximumvalue between the actual number of bidsand the third quartile of the number of bids for each networksegment (ln(#bids) equal to 3.23 for the international corridors, andln(#bids) equal to 3.04 to for the domestic lanes).

In Fig. 8, thegreycrosses correspondto theeffective transportationtariffs and the black diamonds are the predicted values obtainedwhen all roads are paved on the left chart, and when increasingcompetition intensity on the right chart. The results depicted in thesefigures indicate that increasing the level of competition has a greater

010

0020

0030

0040

00

0 500 1000 1500 2000Distance (km)

Domestic tariff Potential outlier

(b) Domestic lanes

l outliers of transportation tariffs.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219212

impact on reducing transportation tariffs than does paving all theroads of the network. By computing the counterfactual costs, we seethat paving the roadswould reduce shipping costs by 18 percent, andimproving competition would reduce them by 44 percent on the in-ternational corridors. Similarly for the domestic roads, better roadconditions would reduce shipping costs by 12 percent and morecompetition by 39 percent. This nevertheless neglects the possibilityfor better roads to stimulate competition.

These findings are consistent with the results of [46]; high-lighting the low level of competition as one of the blocks towardachieving lower transportation tariffs in Africa. Large investmentsin infrastructure are not sufficient to improve transportation mar-kets, since the lack of competition can result in high markups fortransportation services. Improving road infrastructure in morecompetitive environment is more likely to yield lower trans-portation tariffs, as we found in results for the international corri-dors compared to the domestic lanes. The development of thephysical infrastructure is recognized to be fundamental for eco-nomic growth and poverty reduction. A study conducted by Diaoet al. [15] revealed that development in agriculture can highlycontribute to decreasing poverty and to increasing growth inEthiopia, but this would also require reduced market transactioncosts and more investment in transportation infrastructure. Ourresults suggest that government authorities responsible for eco-nomic development in Ethiopia should consider that policiesfacilitating competition in transportation markets are as importantas investments in the transportation infrastructure. Creatingmarket-supportive institutions which ensure competition andencourage market flexibility is essential to make Ethiopia interna-tionally competitive. Measures should also be implemented in or-der to lower the barriers to entry for trucking companies and toincrease information transparency for favoring coordination. For alarge humanitarian agency such as the WFP that has developed astrong expertise in logistics, this means that sharing their knowl-edge with local transporters and supporting new players in thesector could be beneficial for them as well as for the developmentof the Ethiopian economy.

5.4.2. Managerial toolModels 1.4 and 2.4 highlight the importance of the trans-

portation tariff structures for the international corridors and do-mestic lanes in Ethiopia. These models can thus be used to identifythe tariffs that deviate from the pricing tendencies in the WFPtransportation network. In order to test for outliers, we haveidentified the outlying ln (tariff) observations. The potential outliersare the cases with large deleted standardized residuals in absolutevalue identified by means of the Bonferroni test procedure witha ¼ 10 percent (see Neter et al. [31]). Fig. 9 illustrates the lanes forwhich there appear to be systematic differences in pricing. In thisfigure the light grey crosses correspond to the effective trans-portation tariffs and the black diamonds are the transportationnetwork segment suspected to be outliers. These potential outliersare the lanes for which the actual transportation tariffs deviateconsiderably from the estimates provided by Model 1.4 for the in-ternational corridors and by Model 2.4 for the domestic lanes.

The tariffs of the corridors and lanes corresponding to the blackdiamonds in Fig. 9 should be examined carefully. The deviations intariffs for these lanesmay be explained by factors that have not beentaken into account in our models and they may even be legitimate.However, if this is not the case, this indicates a need for managerialintervention. The lanes with a significant positive deviation arelikely to be especially lucrative for carriers andmaybe characterizedby large markups. The lanes with a significant negative deviationmay indicate tariffs that are not effectively covering carrier costs,which would prevent carriers to offer reliable transport capacity at

the bid price. TheWFP should really consider renegotiating its tariffson these network segments or ensure that the actual tariffs will notbe challenged by the contracted carriers.

6. Conclusions

This study constitutes a first attempt to analyze markets forfreight transportation in African countries using shipper data suchas carrier bids and the subsequent contracted tariffs. Using objec-tive data provided by the WFP, one of the major shippers in theEthiopia, the regression models reveal the impact of varioustransportation price drivers in the country and explain around 80percent of the variability in the resulting tariffs. These results canserve several important purposes. First, shippers, like the WFP, canimprove procurement processes and lower costs. By understandingthe cost drivers and estimating fair pricing for transportation ser-vices, shippers can identify opportunities for costs savings andfacilitate better contracting processes. For example, the lanes forwhich there is an objective need for negotiations can easily beidentified based on the difference between predicted and actualtariffs. Second, accurate cost estimates improve broad supply chaindecisions for shippers, such as selecting the best corridors andstocking locations, and evaluating expansion into newmarkets. Forcarriers, better information about the cost drivers and competitivelandscape help them identify market opportunities and refine theirbusiness model. Finally, government authorities can use modelinsights to guide investment strategies in infrastructure and policydecisions. For example, transportation market efficiency should bea priority for Ethiopia development strategies in the same way as itis currently for road infrastructure investments.

Improved understanding of transportation markets in Africawill result in more competitive logistics capacity that fosters eco-nomic development and, in the case of the WFP, enables aid bud-gets to help more in need. This study demonstrates the importanceof analyzing shipper data to better understand transportationmarkets in Africa and provides a foundation for further researchefforts in this direction. While access to data in developing coun-tries is challenging, we hope this work inspires similar studies inother markets.

Acknowledgments

This work was partly funded through a scholarship from theInteruniversity Research Centre on Enterprise Networks Logisticsand Transportation (CIRRELT) and the Fonds de Recherche duQu�ebec e Nature et Technologies (FRQNT). This support is grate-fully acknowledged.We thank theWorld Food Programme for theirkind cooperation and the Contribute Project for funding our trip inEthiopia. Thanks are due to Justin Caron and Gilbert Laporte fortheir help and valuable comments.

Appendix 1. Variables and cross-correlation tables

In this appendix, we first present descriptive statistics of thesocio-economic variables (Table 5). We then present the variablesextracted from the data sources for the 32 international corridorsand the 731 domestic lanes (Tables 7 and 8). We also provide thecross-correlation tables for the socio-economics variables and forthe continuous variables of the two transportation markets. Table 6presents the correlation coefficients of the socio-economics vari-ables extracted from the CSAE. Table 9 presents the correlationcoefficients of the variables associated with the international cor-ridors and Table 10 presents the correlation coefficients of thevariables associated with the domestic lanes for which the distanceis known.

Table 6Cross-correlation table for the regional socio-economic factors (n ¼ 11)

Variables population grain vegetable coffee agriculture cattle sheep goat livestock factory

Demographicpopulation 1.000Agriculturegrain 0.949

(0.000)1.000

vegetable 0.820(0.002)

0.633(0.037)

1.000

coffee 0.845(0.001)

0.788(0.004)

0.814(0.002)

1.000

agriculture 0.958(0.000)

0.999(0.000)

0.661(0.027)

0.807(0.003)

1.000

Livestockcattle 0.986

(0.000)0.978(0.000)

0.777(0.005)

0.865(0.001)

0.985(0.000)

1.000

sheep 0.952(0.000)

0.965(0.000)

0.661(0.027)

0.678(0.022)

0.964(0.000)

0.952(0.000)

1.000

goat 0.937(0.000)

0.960(0.000)

0.624(0.040)

0.705(0.015)

0.959(0.000)

0.948(0.000)

0.964(0.000)

1.000

livestock 0.982(0.000)

0.985(0.000)

0.728(0.011)

0.797(0.003)

0.989(0.000)

0.992(0.000)

0.981(0.000)

0.976(0.000)

1.000

Manufacturingfactory 0.333

(0.317)0.261(0.439)

0.263(0.434)

0.246(0.467)

0.265(0.431)

0.275(0.413)

0.242(0.473)

0.227(0.502)

0.261(0.438)

1.000

Table 5Data on Ethiopia economic activities and population

Distribution of number of factories per major industrial group [9] n %

Food products and beverages 562 25.51Tobacco products 1 0.05Textiles 47 2.13Wearing apparel, except fur apparel 41 1.86Tanning and dressing of leather; footwear, luggage and handbags 89 4.04Wood and products of wood and cork, except furniture 48 2.18Paper and products and printing 127 5.76Chemicals and chemical products 75 3.4Rubber and plastic products 87 3.95Other non-metallic and mineral products 608 27.6Basic iron and steel 18 0.82Fabricated metal products, except machinery and equipment 120 5.45Machinery and equipment 5 0.23Motor vehicles, trailers and semi-trailers 12 0.54Furniture and manufacturing 363 16.48Distribution of agricultural production [8] Quintal %

Grain Crops 104,713,967 95.07Vegetables 3,763,778 3.42Coffee 1,665,791 1.51Distribution of major livestocks [7] n %

Cattle 49,297,898 51.25Sheep 25,017,217 26.01Goat 21,884,223 22.75Population [18] n

Population 72,840,864

Table 7Summary of the variables for the 32 international corridors

Variable Description Type Mean Std deviation Median Min. Max.

Transportation tarifftariff Effective market tariff Continuous 1405.27 902.59 1058.51 532 3999ln(tariff) ln of the effective market tariff Continuous 7.08 0.56 6.96 6.28 8.29Linehaul cost driversDistancedistance Total distance (km) Continuous 756.14 289.68 831.21 297.06 1477.29Road conditionspaved Paved distance (km) Continuous 507.8 285.31 435.99 19.37 1136.18unpaved Unpaved distance (km) Continuous 248.33 239.95 227.25 0 1094Somali risk perceptionhigh somali High WFP risk perception Categorical 4/32 e e e e

not specified somali No risk specified by the WFP for corridors ending in Somali Categorical 12/32 e e e e

(continued on next page)

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 213

Table 8Summary of the variables for the 731 domestic lanes

Variable Description Type Mean Std deviation Median Min. Max.

Transportation tarifftariff Effective market tariff Continuous 1138.84 844.63 900 40 3950ln(tariff) ln of the effective market tariff Continuous 6.69 0.95 6.80 3.69 8.28Linehaul cost driversDistancedistance Total distance (km) Continuous 589.82 356.33 564.97 0 1712Road conditionspoor Poor road condition base on % unpaved Categorical 171/731 e e e e

intermediate Intermediate road condition base on % unpaved Categorical 170/731 e e e e

good Good road condition base on % unpaved Categorical 184/731 e e e e

unknown Unknown road condition Categorical 206/731 e e e e

Somali risk perceptionhigh High WFP risk perception in Somali region Categorical 42/731 e e e e

low Low WFP risk perception in Somali region Categorical 25/731 e e e e

none No WFP risk perception in Somali region Categorical 35/731 e e e e

not specified somali Risk not specified by the WFP in Somali region Categorical 367/731 e e e e

not specified other Risk not specified by the WFP for other regions than Somali Categorical 262/731 e e e e

Estimated tonnagehigh High WFP estimation for tonnage to transport Categorical 65/731 e e e e

low Low WFP estimation for tonnage to transport Categorical 66/731 e e e e

not specified No estimated tonnage specified by the WFP in the RFQ Categorical 600/731 e e e e

Market structureCompetition intensitylnð#bidsÞ ln of the number of bids Continuous 2.45 0.74 2.56 1.1 4.03Market dispersionbid dispersion Standardized interquartile range Continuous 0.4 0.22 0.34 0.02 1.34Market concentrationcarrier origin Number of active carrier at origin Discrete 25.85 13.7 26 6 46carrier destination Number of active carrier at destination Discrete 47.59 19.09 46 23 82hhi origin Bid based Herfindahl-Hirschmann index at origin Continuous 0.08 0.06 0.05 0.03 0.22hhi destination Bid based Herfindahl-Hirschmann index at destination Continuous 0.06 0.03 0.05 0.03 0.13Socio-economicsDemographicpopulation origin Population in the region of origin (1 unit ¼ 10,000 people) Continuous 1097.67 1,1019.39 443.92 34.28 2715.85population destination Population in the region of destination (1 unit ¼ 10,000 people) Continuous 846.92 783.72 443.92 18.33 2715.85Economic activitiesManufacturing activitiesfactory origin Number of factories in the region of origin Continuous 181.09 210.66 44 12 887factory destination Number of factories in the region of destination Continuous 121.87 172.64 12 8 887Agricultural activitiesagriculture origin Agricultural production at origin region (1 unit ¼ 10,000 quintals) Continuous 1691.791 2250.78 75.51 8.77 5326.88agriculture destination Agricultural production at destination region (1 unit ¼ 10,000 quintals) Continuous 1044.45 1747.69 75.51 6.06 5326.88livestocks origin Livestocks at origin region (1 unit ¼ 10,000 animals) Continuous 1419.50 1576.29 315.79 0 3899.13livestocks destination Livestocks at destination region (1 unit ¼ 10,000 animals) Continuous 1007.43 1241.36 315.79 0 3899.13

Table 7 (continued )

Variable Description Type Mean Std deviation Median Min. Max.

not specified other No risk specified by the WFP for corridors not ending in Somali Categorical 16/32 e e e e

Estimated tonnagehigh High WFP estimation for tonnage to transport Categorical 9/32 e e e e

low Low WFP estimation for tonnage to transport Categorical 12/32 e e e e

not specified No estimated tonnage specified by the WFP in the RFQ Categorical 11/32 e e e e

Market structureCompetition intensitylnð#bidsÞ ln of the number of bids Continuous 2.79 0.46 2.91 1.95 3.26Market dispersionbid dispersion Standardized interquartile range Continuous 0.22 0.15 0.17 0.09 0.63Market concentrationcarrier destination Number of active carrier at destination Discrete 57.78 21.17 62 12 82hhi destination Bid based Herfindahl-Hirschmann index at destination Continuous 0.04 0.02 0.04 0.03 0,1Socio-economicsDemographicpopulation destination Population in the region of destination (1 unit ¼ 10,000 people) Discrete 979.59 893.56 443.91 34.28 2715.85Economic activitiesfactory destination Number of factories in the region of destination Discrete 164.28 200.20 43 12 887agriculture destination Agricultural production at destination region (1 unit ¼ 10,000 quintals) Continuous 1289.402 1966.337 75.51 10.71 5326.89livestocks destination Livestocks at destination region (1 unit ¼ 10,000 animals) Continuous 1185.00 1380.28 315.79 0 3899.13

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219214

Table 10Cross-correlation table for the domestic lanes (n ¼ 731).

Variables Tariff Cost driven Market structure Socio-economic

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.

Tariff1. ln(tariff) 1.000Cost driven2. distance 0.520

(0.000)1.000

Market structure3. lnð#bidsÞ �0.737

(0.000)�0.170(0.000)

1.000

4. bid dispersion �0.217(0.000)

�0.193(0.000)

�0.002(0.954)

1.000

5. carriers origin �0.431(0.000)

0.109(0.003)

0.750(0.000)

0.054(0.147)

1.000

6. hhi origin 0.452(0.000)

�0.175(0.000)

�0.781(0.000)

�0.039(0.287)

�0.878(0.000)

1.000

7. carrier destination �0.444(0.000)

0.028(0.453)

0.532(0.000)

�0.038(0.307)

0.348(0.000)

�0.425(0.000)

1.000

8. hhi destination 0.623(0.000)

0.117(0.002)

�0.745(0.000)

0.097(0.008)

�0.388(0.000)

0.476(0.000)

�0.774(0.000)

1.000

Socioeeconomic9. population origin �0.264

(0.000)0.113(0.002)

0.411(0.000)

�0.044(0.233)

0.551(0.000)

�0.447(0.000)

0.304(0.000)

�0.298(0.000)

1.000

10. agriculture origin �0.275(0.000)

0.126(0.001)

0.435(0.000)

�0.029(0.437)

0.562(0.000)

�0.492(0.000)

0.307(0.000)

�0.306(0.000)

0.983(0.000)

1.000

11. livestock origin �0.270(0.000)

0.133(0.000)

0.423(0.000)

�0.036(0.335)

0.546(0.000)

�0.477(0.000)

0.313(0.000)

�0.312(0.000)

0.992(0.000)

0.996(0.000)

1.000

12. factory origin �0.334(0.000)

0.107(0.004)

0.547(0.000)

�0.047(0.202)

0.641(0.000)

�0.553(0.000)

0.368(0.000)

�0.377(0.000)

0.560(0.000)

0.564(0.000)

0.551(0.000)

1.000

13. population destination �0.439(0.000)

�0.150(0.000)

0.503(0.000)

�0.038(0.306)

0.259(0.000)

�0.296(0.000)

0.433(0.000)

�0.516(0.000)

0.231(0.000)

0.226(0.000)

0.234(0.000)

0.264(0.000)

1.000

14. agriculture destination �0.456(0.000)

�0.136(0.000)

0.524(0.000)

�0.032(0.383)

0.274(0.000)

�0.318(0.000)

0.394(0.000)

�0.527(0.000)

0.240(0.000)

0.236(0.000)

0.245(0.000)

0.281(0.000)

0.969(0.000)

1.000

15. livestock destination �0.134(0.000)

0.530(0.000)

�0.039(0.000)

0.277(0.296)

�0.320(0.000)

0.402(0.000)

�0.536(0.000)

0.242(0.000)

0.238(0.000)

0.246(0.000)

0.283(0.000)

0.984(0.000)

0.994(0.000)

1.000(0.000)

16. factory destination �0.527(0.000)

�0.148(0.000)

0.605(0.000)

�0.054(0.143)

0.322(0.000)

�0.377(0.000)

0.359(0.000)

�0.603(0.000)

0.250(0.000)

0.255(0.000)

0.260(0.000)

0.311(0.000)

0.664(0.000)

0.653(0.000)

0.660(0.000)

1.000

Table 9Cross-correlation table for the international corridors (n ¼ 32)

Variables Tariff Cost driven Market structure Socio-economic

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12.

Tariff1. ln(tariff) 1.000Cost driven2. distance 0.344

(0.054)1.000

3. paved �0.207(0.255)

0.652(0.000)

1.000

4. unpaved 0.662(0.000)

0.432(0.013)

�0.402(0.023)

1.000

Market structure5. lnð#bidsÞ �0.751

(0.000)0.063(0.732)

0.484(0.005)

�0.500(0.004)

1.000

6. bid dispersion 0.499(0.004)

0.094(0.608)

�0.497(0.004)

0.705(0.000)

�0.648(0.000)

1.000

7. carrier destination �0.344(0.054)

�0.153(0.402)

�0.006(0.973)

�0.178(0.331)

0.224(0.218)

�0.236(0.194)

1.000

8. hhi destination(0.000)

0.618(0.950)

0.012(0.006)

�0.473(0.001)

0.577(0.000)

�0.697(0.000)

0.680(0.000)

�0.735 1.000

Socioeeconomic9. population destination �0.327

(0.067)0.305(0.089)

0.521(0.002)

�0.251(0.167)

0.425(0.015)

�0.319(0.075)

0.240(0.185)

�0.411(0.019)

1.000

10. agriculture destination �0.340(0.057)

0.195(0.285)

0.456(0.009)

�0.307(0.087)

0.409(0.020)

�0.321(0.073)

0.187(0.305)

�0.398(0.024)

0.967(0.000)

1.000

11. livestock destination �0.336(0.060)

0.256(0.157)

0.506(0.003)

�0.292(0.105)

0.435(0.013)

�0.330(0.065)

0.188(0.303)

�0.408(0.020)

0.986(0.000)

0.992(0.000)

1.000

12. factory destination �0.440(0.012)

0.399(0.024)

0.741(0.000)

�0.400(0.023)

0.599(0.000)

�0.448(0.010)

�0.058(0.755)

�0.421(0.016)

0.564(0.001)

0.553(0.001)

0.554(0.001)

1.000

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 215

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219216

Appendix 2. Heteroscedasticity

Heteroscedasticity occurs when the error terms from a regres-sionmodel do not have constant variance, which can be detected byplotting the standardized residuals on the fitted values [31]. Fig. 10shows the scatter plots of the standardized residuals obtainedwhen regressing tariff and ln(tariff) on the cost driven and marketstructure variables. Comparing Fig. 10(a and b), we observe thatregressing tariff as opposed to ln(tariff) generates non-constanterror terms (the absolute values of the standardized residuals in-

−5−4

−3−2

−10

12

34

Sta

ndar

dize

d re

sidu

als

0 1000 2000 3000 4000Fitted values

International corridor Domestic lane

(a) tariff as dependent variable.

−5−4

−3−2

−10

12

34

Sta

ndar

dize

d re

sidu

als

5 6 7 8 9Fitted values

International corridor Domestic lane

(b) ln(tariff ) as dependent variable.

Fig. 10. Scatter plots of standardized residuals on fitted values for models using the cost driven and market structure variables.

010

2030

Per

cent

age

(%)

−3 −2 −1 0 1 2Standardized residuals

(a) International corridors (Model 1.4).

05

1015

20

Per

cent

age

(%)

−6 −4 −2 0 2Standardized residuals

(b) Domestic lanes (Model 2.4).

Fig. 11. Histograms of standardized residuals for models of ln(tariff) as the dependent variable using the cost driven and market structure variables.

creasewhit the fitted value). Using ln(tariff) as a dependent variableinto the regressionmodels does not generate significant problem ofheteroscedasticity as is the case for tariff. The same conclusion canbe drawn for the models using the socio-economic variables asshown in Fig. 12. To complete the analysis of the OLS assumptions,Figs. 11 and 13 show the histograms of the standardized residualsfor Models 1.4 and 2.4 as well as for Models 1.6 and 2.6. We observein these figures that the standardized residuals are close to beingnormally distributed.

−4−3

−2−1

01

23

4

Sta

ndar

dize

d re

sidu

als

0 1000 2000 3000Fitted values

International corridor Domestic lane

(a) tariff as dependent variable.

−4−3

−2−1

01

23

4S

tand

ardi

zed

resi

dual

s

5 6 7 8 9Fitted values

International corridor Domestic lane

(b) ln(tariff ) as dependent variable.

Fig. 12. Scatter plots of standardized residuals on fitted values for models using the cost driven and soci-economic variables.

010

2030

Per

cent

age

(%)

−2 −1 0 1 2Standardized residuals

(a) International corridors (Model 1.6).

05

1015

Per

cent

age

(%)

−6 −4 −2 0 2Standardized residuals

(b) Domestic lanes (Model 2.6).

Fig. 13. Histograms of standardized residuals for models of ln(tariff) as the dependent variable using the cost driven socio-economic variables.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 217

Appendix 3. Regressions to explain bid values

Table 11 shows the specifications obtained when regressing thecarrier bids on the cost driven factors, the market structure mea-sures and by controlling for carrier specificities through indicatorvariables for the carriers.

Table 11Bid regression models

Variablesa (1) International (2) Domestic

lnðbidÞ (Birr/tonne) lnðbidÞ (Birr/tonne)Cost drivenDistancespaved (km) 0.000404***

(5.13e-05)unpaved (km) 0.000939***

(7.26e-05)distance (km) 0.00139***

(1.62e-05)Road conditionsb

poor 0.295***(0.0193)

intermediate 0.0436***(0.0131)

good �0.146***(0.0106)

Risk perceptionc

none �0.0154(0.0241)

(continued on next page)

Table 11 (continued )

Variablesa (1) International (2) Domestic

lnðbidÞ (Birr/tonne) lnðbidÞ (Birr/tonne)low 0.378***

(0.0319)high 0.610***

(0.0480)0.501***(0.0262)

notspecified 0.0551(0.0344)

0.144***(0.0131)

Tonnage estimatesd

low �0.113***(0.0302)

�0.0929***(0.0132)

high �0.241***(0.0339)

�0.210***(0.0155)

Market structureCompetition intensitylnð#bidsÞ �0.744***

(0.0429)�0.388***(0.0158)

Market dispersionbid dispersion �0.682***

(0.147)�0.443***(0.0225)

Market concentrationcarrier origin 0.00113*

(0.000611)carrier destination �0.00341***

(0.000616)�0.00412***(0.000295)

Constant 9.075***(0.162)

7.537***(0.0463)

n 562 10,377R-squared 0.852 0.772Adj. R-squared 0.845 0.771Max. VIF value 2.39 5.20

Robust standard errors in parentheses.***p < 0.01, **p < 0.05, *p < 0.1.

a Carrier indicator variables are also included in the models but not shown in Table 11.b The category of reference is “unknown”.c The category of reference is “not specified” for regions other than Somali.d The category of reference is “not specified”.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219218

References

[1] Azen R, Budescu DV. The dominance analysis approach for comparing pre-dictors in multiple regression. Psychol Methods 2003;8:129e48.

[2] Barrett CB, Maxwell D. Food aid after fifty years: Recasting its role. New York:Routeledge; 2005.

[3] Behrens K, Picard PM. Transportation, freight rates, and economic geography.J Int Econ 2011;85:280e91.

[4] Berry WD, Feldman S. Multiple regression in practice. Newbury Park, CA: SagePublications; 1985.

[5] Caplice C. Electronic markets for truckload transportation. Prod Operat Manag2007;16(4):423e36.

[6] Caplice C, Sheffi Y. Theory and practice of optimization-based bidding formotor carrier' transportation services. J Bus Logist 2003;24(2):109e28.

[7] Central Statistical Agency of Ethiopia. Report on livestock and livestockcharacteristics; 2008.

[8] Central Statistical Agency of Ethiopia. Agricultural sample survey; 2009.[9] Central Statistical Agency of Ethiopia. Report on large and medium scale

manufacturing and electricity industry survey; 2010.[10] Chainalytics. State of truckload transportation. Shipper semi-annual release.

Freight Market Intelligence Consortium; January 2014.[11] Chander V, Shear L. Humanitarian aid in less secure regions: an analysis of

World Food Programme operations in the Somali region of Ethiopia. Master'sthesis. Massachusetts Institute of Technology; 2009.

[12] Clapp J. Hunger in the balance: The new politics of international food aid.Ithaca, New York: Cornell University Press; 2012.

[13] Comenetz J, Caviedes C. Climate variability, political crises, and historicalpopulation displacements in Ethiopia. Glob Environ Change Part B EnvironHazards 2002;4(4). ISSN: 1464-2867:113e27.

[14] Daganzo CF. Logistics systems analysis. 4th ed. Berlin: Springer-Verlag; 2005.[15] Diao X, Nin Pratt A, Gautam M, Keough J, Chamberlin J, You L, et al. Growth

options and poverty reduction in ethiopia, a spatial, economy wide modelanalysis for 2004e15. Technical Report No. 20. Washington: InternationalFood Policy Research Institute, Development Strategy and Governance Divi-sion; 2005.

[16] Elmaghraby W, Keskinocak P. Combinatorial auctions in procurement. In:Lee H, Billington C, Harrison T, Neale J, editors. The practice of supply chainmanagement. Norwell, MA: Kluwer; 2003.

[17] Ergun €O, Kuyzu G, Savelsbergh MWP. Bid price optimization for simultaneoustruckload transportation procurement auctions. Technical report. H. MiltonStewart School of Industrial and Systems Engineering, Georgia Institute ofTechnology; 2007.

[18] Federal Democratic Republic of Ethiopia Population Census Commission.Summary and statistical report of the 2007 population and housing census;2008.

[19] Jayne TS, Strauss J, Yamano T, Molla D. Targeting of food aid in rural Ethiopia:Chronic need or inertia? J Dev Econ 2002;68(2). ISSN: 0304-3878:247e88.

[20] Kennedy PE. Estimation with correctly interpreted dummy variables insemilogarithmic equations. Am Econ Rev 1981;71:801.

[21] Kunaka C. Logistics in lagging regions: Overcoming local barriers to globalconnectivity. Washington D.C: The World Bank; 2011.

[22] Lall SV, Wang HG, Munthali T. Explaining high transport costs within Malawi -bad roads or lack of trucking competition?; 2009. World Bank Policy ResearchWorking Paper No. 5133.

[23] Lee C-G, Kwon RH, Ma Z. A carrier's optimal bid generation problem incombinatorial auctions for transportation procurement. Transp Res Part LogistTransp Rev 2007;43(2):173e91.

[24] Lim~ao N, Venables AJ. Infrastructure, geographical disadvantage, transportcosts and trade. World Bank Econ Rev 2001;15:451e79.

[25] Liu SC. Production cost analysis of road freight enterprise. J Highw Transp ResDev 2006;23(4):143e6.

[26] Logistics Cluster.Mission statement;November2012. http://www.logcluster.org.[27] Ma Z, Kwon RH, Lee C-G. A stochastic programming winner determination

model for truckload procurement under shipment uncertainty. Transp ResPart Logist Transp Rev 2010;46:49e60.

[28] McDonald BL. Food security. Malden, United States: Polity Press; 2010.[29] Mousseau F. Food aid or food sovereignty? Ending world hunger in our time.

Oakland, California: The Oakland Institute; October 2005.[30] Nandiraju S, Regan A. Freight transportation electronic marketplaces: A sur-

vey of market clearing mechanisms exploration of important research issues.In: Proceedings of the 84th annual meeting of the transportation researchboard; 2008.

[31] Neter J, Kutner MH, Wasserman W, Nachtsheim C. Applied linear statisticalmodels. 4th ed. Boston: McGraw Hill; 1996.

[32] O'Brien RM. A caution regarding rules of thumb for variance inflation factors.Qual Quant 2007;41(5):673e90.

M.-�E. Rancourt et al. / Socio-Economic Planning Sciences 48 (2014) 198e219 219

[33] €Ozkaya E, Keskinocak P, Weight R, Joseph VR. Estimating and benchmarkingless-than-truckload market rates. Transp Res Part E: Logist Transp Rev2010;46:667e82.

[34] Pedersen PO. Freight transport under globalisation and its impact on Africa.J Transp Geogr 2001;9:85e99.

[35] Portugal-Perez A, Wilson JS. Trade costs in Africa: Barriers and opportunitiesfor reform; 2008. World Bank Policy Research Working Paper No. 4619.

[36] Powell WB, Sheffi Y, Nickerson KS, Butterbaugh K, Atherton S. Maximizingprofits for North American van lines’ truckload division: A new framework forpricing and operations. Interfaces 1988;18:21e41.

[37] Rancourt M-�E, Cordeau J-F, Laporte G, Watkins B. Tactical network planningfor food aid distribution in Kenya. Technical Report 9. Montreal, Canada:CIRRELT; February 2013.

[38] Rizet C, Hine JL. A comparison of the costs and productivity of road freighttransport in Africa and Pakistan. Transp Rev 1993;13:151e65.

[39] Rodrigue J-P, Comtois C, Slack B. The geography of transport systems. 2nd ed.New York: Routledge; 2009.

[40] Sandholm T, Suri S, Gilpin A, Levine D. CABOB: A fast optimal algorithm forwinner determination in combinatorial auctions. Manag Sci 2005;51(3):374e90.

[41] Shaw DJ. Global food and agricultural institutions. New York: Routledge;2009.

[42] Shaw DJ. The world's largest humanitarian agency. New York: PalgraveMacmillan; 2011.

[43] Sheffi Y. Combinatorial auctions in the procurement of transportation ser-vices. Interfaces 2004;34:245e52.

[44] Skiba A. Regional economies of scale in transportation and regional welfare.Technical Report 200705. Department of Economics: University of Kansas;2007.

[45] Smith LD, Campbell JF, Mundy R. Modeling net rates for expedited freightservices. Transp Res Part E Logist Transp Rev 2007;43:2007.

[46] Teravaninthorn S, Raballand G. Transport prices and costs in Africa: A reviewof the main international corridors. Washington: The World Bank; 2010.

[47] The WFP Ethiopia Country Team. Joint evaluation of the effectiveness andimpact of the enabling development policy of the World Food Programme e

Ethiopia Case Study, vol. 1; December 2004.[48] The World Food Programme. WFP activities in Ethiopia; February 2013.

http://www.wfp.org/countries/ethiopia/operations.

[49] Thoburn J. Finding the right track for industry in Africa e some policy issuesand options. Technical report. Vienna, Austria: United Nations IndustrialDevelopment Organization; 2002.

[50] UNIDO. The globalization of industry: Implications for developing countriesbeyond 2000. Technical report. Vienna, Austria: United Nations IndustrialDevelopment Organization; 1996.

Marie-�Eve Rancourt is an assistant professor of operations management at theBusiness School of the University of Qubec in Montral. She is also affiliated with theCIRRELT, the MIT, and a member of OCCAH. Her research interests are in the areas oftransportation system modeling and humanitarian logistics using techniques based onoperations research and econometrics. Her recent research projects focus on logisticsissues related with food security in Africa. Dr. Rancourt works in collaboration withdifferent organizations to develop analytical and optimization tools for planning dis-tribution. She received her Ph.D. in Administration (Management Science) from HECMontral in 2013.

François Bellavance is Professor of Statistics in the Department of Decision Sciences atHEC Montral and member of CIRRELT and GERAD research centers. He received hisPh.D. in Statistics from the Universit de Montral in 1994. From 1990 to 1993, hedirected the activities of the Statistical Consulting Service at Simon Fraser University.From 1994 to 1998, he was Biostatistician at St-Mary’s Hospital Center and part-timeAssistant Professor in the Department of Epidemiology and Biostatistics at McGill Uni-versity. His research interests and activities are in data mining applied to large admin-istrative databases in the areas of transportation, road safety and business.

Jarrod Goentzel is founder and director of the MIT Humanitarian Response Lab. Hisresearch focuses on meeting human needs in resource-constrained settings throughbetter supply chain management, information systems and decision support technol-ogy. Dr. Goentzel works with humanitarian and international development organiza-tions on several continents and has over 20 years of experience working withstartup ventures and multinational corporations. Dr. Goentzel has developedgraduate-level courses in humanitarian logistics, international operations and supplychain finance. He received a Ph.D. from the School of Industrial and Systems Engineer-ing at the Georgia Institute of Technology.