Policy Research Working Paper 10789 Intra-national Trade Costs in Low- and Middle-Income Countries Bernardo Díaz de Astarloa Nino Pkhikidze Transport Global Practice June 2024 Policy Research Working Paper 10789 Abstract Casual observation suggests that intra-national trade costs index calculation purposes. It applies the price differential remain high in low- and middle-income countries. Precisely methodology, which aims at estimating trade costs while estimating them is crucial for guiding policies aimed at accounting for the possibility of imperfect competition optimizing economic efficiency within a country’s borders. among intermediaries, controlling for spatial variation in This paper estimates intra-national trade costs for six low- markups. The findings show that the intra-national trade and middle-income countries in Africa and Eastern Europe: costs in the sample of countries are between 2.5 and 14 Kenya, Madagascar, Nigeria, Rwanda, Tanzania, and Geor- times larger than previous estimates for the United States gia. The analysis exploits unit-level price data collected by using the same methodology. countries’ national statistical offices for consumer price This paper is a product of the Transport Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at bernardo. diazastarloa@economicas.uba.ar and npkhikidze@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Intra-national Trade Costs in Low- and Middle-Income Countries ∗ † ‡ ıaz de Astarloa Bernardo D´ Nino Pkhikidze Keywords: Trade costs, Price differential methodology, Travel distance, Kenya, Madagascar, Nigeria, Rwanda, Tanzania, Georgia. JEL classi ication: R12, F1, O1, D02. ∗ This paper was prepared as background research for the World Bank’s Shrinking Economic Distance Flagship Report. Ignacio Caro Sol´ ıs provided superb research assistance. For help and assistance in accessing price data, we thank the World Bank’s country teams and the national statistics offices of the research countries. We have benefited greatly from discussions with Matias Herrera Dappe, Roman Zarate, Theophile Bougna, Alejandro Molnar, Mathilde Lebrand, Aiga Stokenberga, and the participants of World Bank’s Workshop on Shrinking Economic Distance. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. † Inter-disciplinary Institute of Political Economy, University of Buenos Aires. Email: bernardo.diazastarloa@economicas.uba.ar. ‡ World Bank. Email: npkhikidze@worldbank.org. 1 Introduction International trade costs have fallen drastically in the last decades. However, intra-national (i.e. within-country) trade costs still remain high in low- and middle-income countries. Quantifying trade costs and their components is crucial since lower trade costs can increase efficiency, improve the welfare of households and firms, and drive structural transforma- tion through increased trade opportunities (Sotelo, 2019; Fajgelbaum and Redding, 2014; Costinot and Donaldson, 2016). By understanding the costs associated with domestic trade, policy makers can identify barriers hindering market access and formulate targeted policies to reduce these obstacles. Low- and middle-income countries are often characterized by many barriers affecting the free flow of goods within their borders, such as poor road conditions, old transport fleet, and poor logistics. These affect trade flows and limit the gains of reduced interna- tional trade costs and increased regional integration. Remote regions tend to be especially isolated often due to high travel costs, including high travel time. In turn, higher trans- portation costs are often passed on to consumers in the form of increased prices. Therefore, households and firms in these areas face higher spatial price gaps, which limits the variety of goods they can choose from and reduces disposable income (Aggarwal, 2018; Martin, Mayneris and Theophile, 2020). Despite the recent isolated attempts to study the extent of intra-national trade costs,1 there is a gap in the literature on the cross-country comparison of within-country trade costs. In this paper we estimate intra-national trade costs for six low- and middle-income countries in Africa and Eastern Europe: Kenya, Madagascar, Nigeria, Rwanda, Tanza- nia and Georgia. We exploit microdata collected by countries’ national statistical offices (NSO) for consumer price index (CPI) calculation purposes. There are two main estimation-based approaches in the literature to estimate trade sar, 2022). The gravity costs, the gravity approach and the price differential approach (Co¸ approach looks at the impact of varying travel distance or travel time on trade flows, which are inversely proportional to each other.2 Using gravity estimation to estimate the impact of travel distance on trade between Colombian cities, Duranton (2015) estimates that increasing the travel distance between them by 10% decreases trade flows by 7% and weight by 6%. The price differential method combines data on prices from different markets or cities with variables that are thought to affect the cost of shipping goods between locations. Using these data, one can compute price differences across space (spatial price gaps) and assess how different cost shifters affect these differences. Distance between where the 1 See, for example, Atkin and Donaldson (2015) on the estimates for Ethiopia, Nigeria, and the U.S., and Bougna, Ewane, Jones and Kondylis (2020) on Rwanda. 2 Head and Mayer (2014) summarize the literature on gravity equations and provide an extensive overview of best-practice methods and a step-by-step cookbook. 2 product is produced or sourced from (origin locations) and where it is finally sold by retailers (destination locations) typically summarizes the contribution of transport costs to overall trade costs. However, there may be other variables that can be correlated with distance which can also affect trade costs, such as the extent of competition between logistics operators, distributors or retailers (Atkin and Donaldson, 2015) or within-country “border-effects” (Borraz, Cavallo, Rigobon and Zipitria, 2016). In this paper, we follow the price-differential approach and apply the methodology developed by Atkin and Donaldson (2015), which aims at estimating trade costs while taking into account the possibility of imperfect competition among intermediaries, con- trolling for spatial variation in markups. We find that intra-national trade costs in our sample of countries are between 2.5 and 14 times larger than previous estimates for the U.S. by Atkin and Donaldson (2015) using the same methodology. This paper relates to a growing literature that attempts to estimate intra-national trade costs. Given its historically high transport costs, Africa, and in particular Sub- Saharan Africa, has been the focus of several papers studying trade and transport prices. Bougna et al. (2020) estimate intra-national trade costs in Rwanda using data on rural markets and find them to be 10 times higher than in the U.S. They also find that markups are higher in locations with fewer traders. Mayneris, Martin and Theophile (2020) doc- ument spatial differences in the cost of living across Ethiopian cities and find that more remote cities show significantly higher prices, less product availability, and a higher cost of living. Porteous (2019) takes a different methodological approach to Atkin and Donaldson (2015) to estimate intra-national trade costs in Sub-Saharan Africa. The paper focuses on staple cereal grains traded between 230 large regional hubs in 42 countries, and quantifies trade costs by estimating a structural model that features competitive traders (i.e., ab- stracting from markups), grain consumption, storage, and trade. The results in Porteous (2019) suggest trade costs that are qualitatively similar but higher in magnitude to those estimated in Atkin and Donaldson (2015), although the former potentially include non- distance dependent factors. Specifically, the median trade cost is over 5 times higher than international benchmark freight rates. In a counterfactual where trade costs on overland links are reduced to levels similar to the rest of the world, grain prices fall 30% on average and aggregate welfare increases by 1.6% of GDP. Teravaninthorn and Raballand (2009) estimate trade costs using trucker surveys in the U.S. and along several major transport corridors in Sub-Saharan Africa and find 1.88 to 3.28 times higher trade costs compared to the U.S., which are lower than the results of Atkin and Donaldson (2015). In terms of the trade costs in high-income countries, Agnosteva, Anderson and Yotov (2019) use gravity model techniques to understand the effects of intra-regional, inter-regional and international frictions on trade flows in Canadian provinces, and find that further and less-developed regions tend to have relatively low internal frictions and large border ef- fects, while economically more developed and central regions show relatively low border barriers. 3 2 Data Our sample covers 6 low- and middle-income countries in Africa and Eastern Europe: Kenya, Madagascar, Nigeria, Rwanda, Tanzania and Georgia. In order to apply the methodology developed by Atkin and Donaldson (2015), we require two sets of data for each country. First, we need price data on narrowly-defined product varieties, with ob- servations spanning multiple locations within each country at a monthly frequency over several years. Second, we need the locations in which product varieties were produced or from which they were imported. The latter allows for the identification of trading pairs and the corresponding direction of trade from origin to destination locations. In this sec- tion, we briefly describe how we construct these data. Appendix A includes additional details about the data preparation process. 2.1 Price data Our price data come from CPI microdata provided to us by NSOs in each country in our sample. NSOs collect price data from a predefined list of product varieties (including goods and services), designed to capture the consumption basket of a typical household, at specified locations. Apart from the monthly price quote, the information provided to us by NSOs includes a product description, in some cases including the brand and presentation (e.g., “Coca-Cola, canned, 500ml” or “Body lotion, Nivea, 200ml”), the name of the location where the price was collected, and, occasionally, the name of a producer or the country from where the product was imported. We restrict the choice of products to narrowly defined varieties, i.e., those for which we can observe a detailed product description and a brand. For most countries in our sample, prices are at the town or city level, with the exception of Rwanda, where prices are collected at the (more aggregated) district level. The analysis in section 4 uses clean samples of price data, obtained by applying a cleaning algorithm to the raw data. Moreover, as in Atkin and Donaldson (2015), we convert all prices to constant U.S. dollars by deflating them using inflation rates at origin locations and bilateral nominal exchange rates between local currencies and the U.S. dollar. The cleaning and deflating procedures are described in Appendix A. 2.2 Products’ origin locations The estimation strategy proposed by Atkin and Donaldson (2015) depends on precisely identifying locations where a product is produced or imported. To do this, we search over the Internet to determine which manufacturers produce each brand and where their factories are located. For imported goods, we identify the main ports of entry in each 4 country and their locations to assign imported products to them.3 For landlocked coun- tries, we identify ports at neighboring countries through which most imports come (e.g., in some cases, countries are in conflict with some bordering countries and imports do not come through them). Then, we identify the border crossings that are closest to those ports. Finally, we geocode all locations to determine their latitude and longitude. For each product, we define a “trading pair” as an origin and a destination location. Conditional on being narrowly defined, we consider products that are broadly available across locations and periods in our sample. Specifically, we restrict samples to products for which prices are observed in the products’ source and destination locations in more than six months.4 Table 1 describes the main features of the resulting samples we consider for each country. The number of products ranges from 9 in Madagascar to 43 in Rwanda. The full list of products for each country, including their brands, presentations, and manufacturers, is included in the Appendix A.2. The number of destination locations (markets, cities, or districts) ranges from 6 in Georgia to 38 in Nigeria. Figure 1 shows the locations for each country in our sample, as well as the major roads connecting them, and indicates the location where a product is manufactured or imported, on which we elaborate in section 2.2.5 Table 1 also shows the periods covered by each country sample. With the exception of the Nigeria sample, which starts in January 2001, all samples start after January 2010. The longest panel corresponds to Madagascar and covers the period from January 2010 to April 2021. The shortest panel is the one for Kenya, covering from October 2018 to January 2022. Average origin prices, price gaps between origin and destination locations, and dis- tances, are reported in Table 2. Table 1: Main features and coverage of the data Country No. of products Origins Destinations Type of locations Start period End period Kenya 20 10 32 Cities Oct 2018 Jan 2022 Madagascar 9 3 7 Major cities Jan 2010 Apr 2021 Nigeria 25 5 38 Cities Jan 2001 Jul 2010 Rwanda 43 4 13 Districts Jan 2013 Dec 2020 Tanzania 32 5 20 Cities Jan 2012 Apr 2021 Georgia 21 4 6 Cities Jan 2012 Dec 2020 Notes: this table describes the main features of the datasets for each country in our sample. Source: authors’ elaboration based on CPI data from countries’ NSO. 3 Our main sources are Logcluster, a wiki with detailed logistics information about each country, and World Port Source. 4 As discussed in section 3.2, estimation of pass-through rates relies on variation of prices over time for each product-destination pair. 5 The data underlying the maps are taken from the Global Roads Inventory Project (GRIP), available at https://www.globio.info/download-grip-dataset. 5 (a) Kenya (b) Madagascar (c) Nigeria (d) Rwanda (e) Tanzania (f) Georgia Figure 1: Maps of market locations Notes: this figure shows market locations for each country in our sample, as well as major roads connecting locations, indicating origin locations (production locations or ports of entry). Source: authors’ elaboration based on the GRIP dataset. 6 2.3 Distance between locations The baseline measure of distance we employ is geodesic distance, that is, the shortest dis- tance between two locations. Table 2 reports the mean, minimum and maximum geodesic distance between origin and destination locations. Alternatively, following Atkin and Donaldson (2015), we use two additional measures to control for quantity and quality differences in countries’ road networks. The first, which aims at capturing quantity differences, is the distance between two locations along the quickest route, as calculated by Google Maps. Figure 1 shows the road network in each country. The second, which should capture quality differences, is the time it takes to complete a trip between two locations following the quickest route, also from Google Maps. Table 2: Descriptive statistics KE MG NG RW TZ GE Mean origin price 2.306 0.922 0.902 2.328 0.898 2.577 (1.833) (0.451) (1.223) (2.993) (0.840) (2.775) Mean destination price 2.357 0.918 0.960 2.302 0.910 2.584 (1.984) (0.520) (1.386) (2.980) (1.001) (2.729) Mean price gap (trading pairs) 0.052 -0.004 0.058 -0.025 0.013 0.007 (0.454) (0.264) (0.433) (0.480) (0.662) (0.454) Geodesic distance (km) 279 414 597 80 493 170 (208) (224) (292) (35) (276) (79) Log distance to origin location 5.305 5.858 6.214 4.273 5.943 5.001 (0.870) (0.605) (0.691) (0.499) (0.927) (0.551) Min. distance (trading pairs, km) 20.6 114.3 34.2 7.9 2.2 50.9 Max. distance (trading pairs, km) 915.8 874.9 1,254.8 152.9 1,362.6 318.5 Number of trading pairs 134 18 153 30 85 20 Observations 9,280 5,997 48,379 26,215 50,583 6,173 Notes: This table presents descriptive statistics for our sample of countries. The first and second rows report the mean price at the factory location (or location of the port of entry) and the mean price at destinations. The third row reports price gaps using data form trading pairs. All prices are deflated by inflation at the origin location and then converted to U.S. dollars using the exchange rate prevailing during the base period. The fourth row reports mean geodesic distance between trading pairs measured in kilometers. The fifth row reports the average distance to the origin location (in log kilometers). The fifth and sixth rows report the minimum and maximum distances between trading pairs, respectively. The seventh row reports the number of trading pairs. The last row reports the number of observations. Source: Authors’ estimates based on national CPI data and distances calculated using Google. 3 Methodological framework The theoretical framework underlying our estimation strategy is based on the model de- veloped by Atkin and Donaldson (2015). In this section, we outline the main elements 7 and intuition of the theory. Interested readers can find additional details and proofs in that paper. 3.1 Theory The economy is characterized by multiple markets (or locations) indexed by d and a single consumption good. The inverse demand for this product in each market is given by P (Qd , Dd ), where Qd is quantity demanded and Dd characterizes demand conditions in location d. Production of the good can take place at home or abroad at a single factory. What matters for the model is the domestic origin of the product, denoted o. If produced at home, o indicates the factory location; if imported, o indicates the port or border crossing through which the product entered the country. Identical domestic intermediaries buy the product in wholesale markets at the origin location at price Po and sell it to consumers in d, effectively acting as retailers as well. Intermediaries’ trading technology is characterized by their total costs, Cd (qd ), where qd is the quantity traded from o to d. In the extension of the model to many products, we further assume that each product is sourced from one location only, so that o can be omitted from the subscript in qd . Total costs include a fixed cost Fd and a marginal cost cd . The marginal cost includes the cost of buying the good at the origin, given by its price Po , and the cost of trading the food from o to d, given by τ (Xd ), where Xd is a vector of cost shifters, including distance and, potentially, other factors such as road quality and destination-specific retail costs. Total costs are then given by C (qd ) = Fd + [Po + τ (Xod )] . An intermediary choose quantities qd to maximize profits in an imperfectly competitive market, subject to the perceived response of other intermediaries. There is no entry of intermediaries and, under certain assumptions, the first-order conditions of the intermedi- ary’s optimization problem imply that price gaps between the origin and the destination locations are given by: ∆Pod ≡ Pd − Po = τ (Xod ) + µd (τod , Dd , ϕd ) (1) where µd is the mark-up charged by the intermediary at location d, which depends on trade costs, demand conditions at d, and a competitiveness index ϕd , which in turn depends on the number of intermediaries and the extent of strategic interactions among them. From (1), variation in trade costs shifters implies: d∆Pod ∂τ (Xod ) ∂µ ∂Dd ∂µ ∂ϕd = ρd + + , (2) dxod ∂xod ∂Dd ∂xod ∂ϕd ∂xod where ρod ≡ 1 + (∂µ/∂τod ) is the pass-through rate, that is, the effect of trade costs on 8 prices. It can be shown that, in this framework, the pass-through rate depends on the elasticity of the slope of the demand schedule and competitive conditions (ϕd ). In our empirical analysis in Section 4, the main trade cost shifter of interest xod is distance. To simplify the exposition, in what follows we set Xod = xod . Equation (2) illustrates three sources of bias that can arise when aiming at identifying trade costs from spatial price gaps: (1) incomplete pass-through (ρod ̸= 1), (2) variation in demand conditions across locations, and (3) variation in the extent of competition across locations. Under additional assumptions, the pass-through rate can control for this sources of bias. In particular, under the assumption that the pass-through rate is independent of quantities qd , equation (2) simplifies to ∆Pod = ρod τ (xk od ) + (1 − ρod )(ad − Po ), (3) where ad is a local demand shifter. Under this specification, ρod and the local demand shifter ad are sufficient to control for local competitive and demand conditions. The following subsection explains how this theoretical framework can guide the spec- ification of empirical models to estimate the effect of distance on trade costs exploiting origin-destination price gaps, controlling for other factors that correlate with distance and could affect price differentials through variation of markups across locations. Section 4 describes the methods used to estimate these models. 3.2 Estimation strategy The estimation strategy exploits price data on multiple products k sourced from locations o and selling at locations d at different time periods t. With variation across products, destinations, origin locations, and time, equation (3) implies the following relationship between product prices at the origin and the destination: k Pdt = ρk k k k k k od Pot + ρod τ (xodt ) + (1 − ρod )adt , (4) where we have further assumed that the pass-through rate remains fixed over time.6 The latter assumption holds if local demand elasticities and competitive conditions (e.g., entry of new intermediaries) are constant over time. In equation (4), trade costs τ (xk k k odt ), pass-through rates ρod and demand shifters adt are not observable. The strategy aims at identifying τ (xk k odt ) and ρod based on variation in prices over time and space, and distance between locations, while treating ak dt as un- observed heterogeneity and capturing it through a set of fixed effects. For this reason, precisely identifying origin locations is key for the accuracy of the estimates. The estimation procedure involves two steps, as follows: 6 It is infeasible to estimate separate pass-through rates for each period since the number of parameters to estimate would equal the number of observations. 9 1. Pass-through rates. In the first step, we produce estimates of pass-through rates ˆk ρod by regressing destination prices on origin prices, controlling for trade costs and demand shifters. Specifically, we estimate the parameters of the following model: k Pdt = ρk k k k k od Pot + γd + γd t + εodt , (5) k are product-destination fixed effects, γ k t is a product-destination linear where γd d time trend, and εk odt is an unobserved error term that captures shocks to trade costs and local demand shifters. ˆ(xodt ) using 2. Trade costs. In the second step, we recover estimates of trade costs τ ˆk pass-through estimates from the first step. Given (unbiased) estimates ρod , we can manipulate equation (4) to derive and expression for an “adjusted price gap”: k −ρ Pdt ˆk k 1−ρ ˆk ˜k ≡ od Pot ∆Podt = τ (xk odt ) + od ak dt . (6) ˆk ρod ˆk ρod Computing this “adjusted price gap” allows for the estimation of τ (xk odt ) when markups are positive and vary across locations, i.e., ρk od ̸= 1. Local demand con- ditions are controlled for by including fixed effects interacted with the adjustment factor, and trade costs are decomposed as τ (xk k k odt ) = f (xod )+χodt , where χodt captures unobservable time-varying factors that affect trade costs. The estimating model then becomes ˜ k = f (xod ) + αk 1−ρ ˆk od 1−ρ ˆk od k ∆P odt t + αd + ξodt , (7) ˆk ρod ˆk ρod k are product-time fixed effects, α are destination fixed effects, and ξ k where αt d odt captures unobserved shocks to local demand and trade costs. In Section 4.3, we specify the function f (xod ) that we take to the data to recover the effect of distance on trade costs. 4 Estimates of intra-national trade costs This section presents estimation results to quantify the relationship between distance and intra-national trade costs in our sample of low- and middle-income countries. Using the notation introduced in the previous section, we attempt to estimate ∂τ (xod )/∂xod , where xod is one of the different measures of distance described in section 2.3. Ultimately, we apply the strategy described in section 3.2, which aims at estimating this relationship using data about price gaps across locations, controlling for variation in markups across locations. Before turning to that exercise, for benchmarking purposes, we first present estimates that abstract from variation in markups. 10 4.1 Spatial price gaps Under the theoretical framework described in section 3.1, if markups did not vary across locations the pass-through rate would be constant across trading pairs and equal to one, i.e. ρk od = 1. From equation (3), this restriction implies ∆Podt = τ (xodt ). (8) We estimate equation (8) through OLS imposing a linear relationship between distance k + χk , where γ k are product-time fixed effects and trade costs, i.e. τ (xodt ) = βxod + γt odt t that capture economy-wide product-specific shocks to trade costs and χk odt captures other components of trade costs that are unrelated to distance. With the assumption of perfect competition, differences in prices across locations should reflect trade costs and, therefore, we would expect β ˆ to be positive. A negative β ˆ could arise in the more general case of imperfect competition with variable markups. Yet another potential explanation for observing price gaps that decrease with distance is that local distribution or logistic costs in the destination are higher at destination locations that are closer to the origin. Results are presented in Table 3. Except for Rwanda, the results show a positive relationship between origin-destination price gaps and distance. Table 3: Estimates of the relationship between distance and trade costs using origin- destination price gaps KE MG NG RW TZ GE ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Distance to origin (log km) 0.070 0.071 0.030 −0.082 0.026 0.028∗∗ (0.006) (0.005) (0.003) (0.005) (0.002) (0.009) Observations 9, 280 5, 997 48, 379 26, 215 50, 583 6, 173 Adj. R2 0.258 0.408 0.249 0.344 0.238 0.585 Notes : This table presents the relationship between distance and price gaps across regions, under the assumption that markups do not vary across locations. All regressions use the (unad- justed) price gap between trading pairs and include product-time fixed effects. Standard errors in parentheses. ∗∗∗ significant at 1%, ∗∗ at 5%, and ∗ at 10%. Source: Authors’ estimates based on national CPI data and distances calculated using Google. 4.2 Pass-through rates The first step of the procedure for estimating intra-national trade costs outlined in section 3.2 involves estimating pass-through rates. Pass-through rates are informative of how (exogenous) price changes at origin locations translate into (endogenous) price changes in destination locations, in the way described by equation (4), which we reproduce here: k Pdt = ρk k k k k k od Pot + ρod τ (xodt ) + (1 − ρod )adt , (9) Recall that, in equation (9), ak dt are non-observable demand shifters. Following Atkin 11 and Donaldson (2015), we specify equation 9 including a set of fixed effects such that varia- tion in trade costs and local demand shifters that are product-specific within a destination over time is orthogonal to price changes at the origin: k Pdt = ρk k k k k od Pot + γd + γd t + εodt , (10) k are product-destination fixed effects, γ k t is a product-destination linear time where γd d trend, and εk odt is an unobserved error term that captures shocks to trade costs and local demand shifters. In order to obtain unbiased OLS estimates of ρk k od , prices at the origin Pot should not be correlated with destination-specific and time-varying shocks to trade costs or demand captured by εk odt . As noted by Atkin and Donaldson (2015), this is likely to hold for imported goods, whose prices are mostly determined abroad, or if prices at the origin are either pinned down by production costs at the factory gate or set on the basis of demand shocks at the origin. This condition would not hold, however, in the presence of economywide or spatially correlated demand shocks that are product-specific. In this case, estimates of ρk od would be biased upwards. Figure 2 displays our results. For each country in our sample, the figure plots estimates of ρk od for all products k and locations d against the corresponding origin-to-destination distance xod , together with a non-parametric estimate of the relationship between the pass-through rate and distance. The figure reveals substantial variation in the magnitude of pass-through rates, although there are some common patterns. First, few estimates violate the theoretical restriction that the pass-through rate be positive. As detailed in Table 4, between 75% and 85% of our estimates in any country are positive. Second, there is substantial evidence of incomplete pass-through, which suggests imperfect competition. The average pass-through rate is below one in all countries and the fraction of estimates below one ranges from 77% in Tanzania to 94% in Georgia and Nigeria. Third, in three out of six countries the pass-through rate shows a statistically significant negative rela- tionship with distance, as the fourth row of Table 4 indicates. In these countries (Georgia and Nigeria), the pass-through rate is lower in locations that are farther away from the product’s origin location. In those countries where the negative relationship holds, it tends to be fairly monotonic, as Figure 2 shows. 12 (a) Kenya (b) Madagascar (c) Nigeria (d) Rwanda (e) Tanzania (f) Georgia Figure 2: Estimates of pass-through rates and distance to origin location Notes: in these figures, each dot represents an estimate of the pass-through rate for a particular product- location pair. Source: authors’ elaboration based on CPI data. 13 Table 4: Summary statistics of pass-through estimates KE MG NG RW TZ GE Average pass-through 0.543 0.295 0.369 0.396 0.552 0.377 (0.751) (0.646) (2.568) (0.882) (1.247) (0.426) % of estimates below 1 83.2 82.4 94.2 84.0 76.7 94.3 % of estimates below 0 15.1 25.5 20.5 21.2 24.0 18.2 Distance to origin location 0.082 0.115 -0.012 -0,075 0.029 -0.099 (0.057) (0.149) (0.128) (0.091) (0.057) (0.081) Observations 232 51 832 368 562 88 Notes : This table presents summary statistics of pass-through estimates. The third row includes estimates of a linear relationship between the pass-through rate and (log) distance to origin. For average pass-through rates, standard deviations indicated in parenthesis. For the coefficient on distance to origin coefficient, standard errors indicated in parentheses. ∗∗∗ significant at 1%, ∗∗ at 5%, and ∗ at 10%. Source: Authors’ estimates based on national CPI data and distances calculated using Google. 4.3 Estimates of trade costs using pass-through-adjusted price gaps and distance We now turn to the estimation of trade costs under the more general case of imperfect competition. With imperfect competition, price gaps can vary across trading pairs not only because trade costs vary with distance, but also because intermediaries can charge different markups in different locations, i.e. there is spatial variation in markups. Atkin and Donaldson (2015) emphasize three mechanisms by which markups can vary with distance: 1. More remote locations face higher trade costs, which, all else equal (i.e. demand conditions), induces traders to choose lower markups since they face higher marginal costs. 2. More remote locations are characterized by fewer traders and less competition, which induces traders to choose higher markups. 3. Traders in more remote locations may face different demand conditions, which in- duces correlation between distance and markups. In their results, Atkin and Donaldson (2015) find that ignoring variation in markups can considerably underestimate intra-national trade costs. Their methodology to account for spatial variation in markups in the estimation of trade costs is based on using the ˆk pass-through rates estimated above, ρod , to adjust price gaps across trading pairs, as explained in section 3.2. Assuming a linear relationship between distance and trade costs, 14 the resulting model that we take to the data is: ˜ k = βxod + αk 1−ρ ˆkod 1−ρ ˆk od k ∆P odt t + αd + ξodt . (11) ˆk ρ od ˆk ρod k are product-time In this equation, β captures the effect of distance on trade costs, αt k captures unobserved shocks to local fixed effects, αd are destination fixed effects, and ξodt demand and trade costs.7 Panel A in Table 5 presents estimates of β for all countries in our sample, the main results of the paper. All specifications include product-time and destination fixed effects, and standard errors are clustered at the product-time level. The table includes estimates using different measures of distance, but we start discussing the baseline estimates that consider geodesic distance. Except for Georgia and Madagascar, all coefficients are sta- tistically significant. Once we take into account the possibility of imperfect competition, moreover, all coefficients are positive (with the exception of Georgia, where the coefficient is not statistically different from zero), indicating that trade costs increase with distance. Kenya shows the highest coefficient, Tanzania the lowest. Compared to the estimates in Table 3, adjusting for markups does not produce a common pattern across countries. For Kenya, and Nigeria, the coefficients are higher than those in the unadjusted regressions. For Rwanda, the coefficient turns positive. For Madagascar and Tanzania, the coefficients are now lower. Our baseline estimates imply significant variation in the magnitude of trade costs across countries. To illustrate this, we compute the ad valorem trade cost equivalent implied by ˆ, which is defined as β ˆ × xmax − xmin β od od τx ≡ , (12) Pd where xmin max od and xod are the (log) distances between the least and most remote locations away from the source location, respectively, and P d is the average destination price in the sample. This measure quantifies the additional trade cost implied by trading the good in the most remote location compared to the least remote location, as a fraction of the average final consumer price. In our sample, the average minimum distance between origin and destination locations is 38 kilometers and the average maximum distance is 813 kilometers. Using these values, Figure 3 shows the ad valorem equivalent trade cost of trading a good between two locations 775 kilometers apart implied by the coefficients in Table 5.8 Excluding Georgia, for which we obtain a coefficient equal to zero, these ad valorem equivalents range from 3.2% in Rwanda to 19.2% in Kenya. 7 As in Atkin and Donaldson (2015), for our baseline estimates we winsorize pass-through estimates that are below 0.2. 8 We use the geodesic distance coefficient for the results presented in the figure. 15 Figure 3: Implied ad-valorem equivalent trade costs Notes: this figure shows implied ad valorem equivalent trade costs of trading goods between two locations as a percentage of the average destination price, computed from estimated trade cost coefficients βˆ and assuming a geodesic distance of 775 km between locations. Source: authors’ elaboration. 16 Table 5: Estimates of the effect of distance on trade costs KE MG NG RW TZ GE Panel A: All goods Geodesic distance to origin (log km) 0.148∗∗∗ 0.024 0.052∗∗∗ 0.024∗ 0.016∗∗∗ 0.000 (0.014) (0.028) (0.007) (0.011) (0.004) (0.029) Adjusted R2 0.980 0.931 0.849 0.971 0.831 0.990 Driving distance to origin (log km) 0.127∗∗∗ 0.028 0.052∗∗∗ −0.0010 0.023∗∗∗ −0.047 (0.013) (0.026) (0.007) (0.009) (0.004) (0.035) Adjusted R2 0.98 0.931 0.849 0.971 0.832 0.99 Time along best route (log hrs) 0.134∗∗∗ 0.048 0.060∗∗∗ 0.004 0.029∗∗∗ −0.04 (0.013) (0.028) (0.007) (0.010) (0.005) (0.039) Adjusted R2 0.98 0.931 0.851 0.971 0.831 0.99 Observations 9, 280 5, 997 48, 378 26, 215 50, 583 6, 173 Panel B: Locally-produced goods only Geodesic distance to origin (log km) 0.148∗∗∗ 0.068∗ 0.041∗∗∗ 0.120∗∗∗ 0.028∗∗∗ 0.084∗∗∗ (0.014) (0.031) (0.010) (0.015) (0.004) (0.021) Adjusted R2 0.980 0.915 0.846 0.979 0.825 0.978 Driving distance to origin (log km) 0.127∗∗∗ 0.05 0.043∗∗∗ 0.100∗∗∗ 0.031∗∗∗ 0.133∗∗∗ (0.013) (0.029) (0.009) (0.012) (0.004) (0.028) Adjusted R2 0.980 0.915 0.846 0.979 0.825 0.978 Time along best route (log hrs) 0.134∗∗∗ 0.056 0.050∗∗∗ 0.074∗∗∗ 0.037∗∗∗ 0.185∗∗∗ (0.013) (0.032) (0.010) (0.012) (0.005) (0.036) Adjusted R2 0.980 0.915 0.848 0.979 0.825 0.978 Observations 9, 280 5, 271 43, 541 9, 999 36, 541 3, 406 Destination fixed effects Yes Yes Yes Yes Yes Yes Product-time fixed effects Yes Yes Yes Yes Yes Yes Notes : This table presents the effects of distance on intra-national trade costs. All regressions use the adjusted price gap between trading pairs and include product-time and destination fixed effects ˆk multiplied by (1 − ρ ˆk od )/ρ od . Panel A includes all goods, including imported goods. Panel B considers locally produced goods only. Product-time clustered standard errors in parentheses. ∗∗∗ significant at 1%, ∗∗ at 5%, and ∗ at 10%. Source: Authors’ estimates based on national CPI data and distances calculated using Google. 5 Discussion of results and robustness checks The baseline estimation presented above uses geodesic distance as the main exogenous variable shifting trade costs. However, this measure may mask cross-country differences in the quantity and quality of road infrastructure. For example, if country A has more roads connecting two locations than country B, then the cost-advantage of having more roads would be reflected in a lower coefficient in country A. Figure 1 suggests road infrastructure certainly varies across the countries in our sample. In the same vein, if there are cross- country differences in road quality, adjusting for the quality of roads should also reduce cross-country differences in our baseline estimates. Motivated by this, Panel A of Table 5 also includes coefficients when we replace geodesic 17 distance with two alternative measures, the driving distance along the best route (to adjust for quantity) and the time traveled along the best route (to adjust for quality), both computed using Google Maps. Using these alternative measures of distance changes the coefficients differently for different countries, without a clear pattern. The methodology for recovering trade costs that we employ relies heavily on precisely identifying the source location of goods. One concern is that the source locations of imported goods included in our sample are not precisely identified for every period. That is, we may be incorrectly assigning the border or the port of entry. Alternatively, it may be that some of the distribution or logistic costs are already included in the origin price that we observe, because we may not be measuring these goods exactly at the border, so that trade costs for these goods may be underestimated. For these reasons, we rerun steps 1 and 2 of our estimation procedure considering locally-produced goods only. Results are included in Panel B of Table 5, for all distance measures. It turns out that focusing on locally produced goods significantly changes the results. Coefficients estimated using geodesic distance increase markedly for Georgia, Rwanda, and Tanzania, where the share of imported products in the sample is above 30%, and for Madagascar, where only one product is imported, but for which we have only 9 products. In Georgia, the coefficient turns from zero to 0.084 and in Madagascar, Rwanda and Tanzania the coefficients increase by a factor of 2.8, 4.5 and 3, respectively. The coefficient for Kenya remains unchanged, since there are no imported goods, and the coefficients for Nigeria decrease 20%. The implied ad valorem equivalents change in the direction of the change in the coefficients, and are included in Figure 3.9 For example, in Georgia the ad valorem transport cost is now 15.2%, and in Madagascar and Rwanda it is above 20%. 5.1 Robustness checks In this section, we perform a number of robustness checks to study the sensitivity of the results in section 4.3 to the assumptions regarding the estimation of pass-through rates and intra-national trade costs. The estimates of the effect of distance on trade costs obtained under these alternative specifications are presented in Table 6, which reproduced the baseline estimates in the first row. In all cases, we use our preferred sample of locally- produced goods only. We explain each of the robustness checks in what follows. We begin by exploring alternatives to the estimation of pass-through rates, which are important to control for the effect of imperfect competition, in rows 2-4. The baseline ˆk specification bottom-codes estimates of pass-through rates by setting ρ ˆk od = 0.2 if ρ od < 0.2, since the theoretical framework imposes that pass-through rates be positive. Results are sensitive to using the raw estimates, as can be seen from row 2 of Table 6. The effect 9 To compute the ad valorem equivalents, we now use the average destination price of locally-produced goods. 18 of distance on trade costs is lower in all countries but in Georgia and Nigeria, where it becomes significantly higher. For Madagascar and Rwanda, the coefficient becomes negative. Row 3 shows estimates under a specification that relaxes the assumption that the pass-through rate is constant over time. Instead, under this specification we split each sample period in halves and estimate a different pass-through rate for each period.10 Allowing the pass-through rate to vary over time changes the results for all countries except for Tanzania. For Georgia, the coefficient is higher than in the baseline specification while it turns negative for the rest of the countries. In row 4, pass-through rates are estimated under a specification that adds two lagged origin price terms to equation (10), in order to capture long term pass-through effects. The pass-through rate is then the sum of the contemporaneous effect and the coefficients associated with the two lags. Using these estimates lowers the effect of distance (and the precision of the estimates) in all countries, turning it negative in all of them except in Nigeria and Tanzania. In row 5, we control for potential correlation between spatial variation in location- specific trade costs and distance between locations, which could bias our baseline esti- mates. To do this, we include fixed effects for each destination location d, which were ˆk included in specification (11) but interacted with the adjustment term (1 − ρ ˆk od )/ρod . For some countries, coefficient estimates change significantly after including destination fixed effects. For Georgia and Nigeria the coefficients increase markedly (by a factor of 2.9 and 15.3, respectively). For Kenya and Rwanda they also increase, but more moderately. For Madagascar and Tanzania, they turn negative. These results suggest that, in these coun- tries, local components of trade costs, such as fulfilment, distribution and retail costs, may not be independent of distance to the source location. However, the results are ambiguous about the sign of the bias. The last set of robustness checks explores changes in the estimation of both pass- through rates and trade costs altogether. First, specifications in rows 6 and 7 include additional fixed effects to control for more general demand and trade costs structures. Specifically, instead of including destination fixed effects, rows 6 and 7 show estimates from specifications with year-location and month-location fixed effects, respectively, in equations (10) and (11). These alternative specifications change coefficients significantly but without a clear pattern across countries. For some, the coefficient on distance increases and, for others, it decreases and even turns negative (for Nigeria, estimating the model with month-destination fixed effects is computationally infeasible). Finally, the specifications in rows 8 and 9 aim at evaluating the possibility that demand shocks in the destination and origin locations are spatially correlated, which would bias the baseline estimates. To this end, models (10) and (11) are re-estimated with samples restricted to locations that are at least 100km (row 8) and 200km (row 9) apart (the latter is not feasible for Rwanda, for which there are no locations more than 200km apart). Again, coefficients 10 This is approximately 5 years for all countries except for Kenya, for which we only have 3.3 years of data. 19 change significantly for all countries, except possibly for Kenya, without a clear pattern, which suggests that the identifying assumption of spatial independence of demand shocks at origin and destination locations is not plausible for our sample of countries. All in all, the conclusion from the robustness exercises is that our baseline estimates are not stable as we implement alternative specifications or samples, although changes in the estimates do not follow a clear pattern across countries. This suggests that there may be idiosyncratic institutional features in each country that should be taken into account more precisely in order to better understand the nature of the determinants of trade costs. Table 6: Estimates of the effect of distance on trade costs: robustness checks KE MG NG RW TZ GE ∗∗∗ ∗ ∗∗∗ ∗∗∗ ∗∗∗ Baseline 0.148 0.068 0.039 0.122 0.028 0.084∗∗∗ (0.014) (0.031) (0.009) (0.015) (0.004) (0.021) Not winsorizing 0.097∗∗∗ −0.258∗∗∗ 0.208∗∗∗ −0.060∗ 0.006 0.157∗∗∗ (0.013) (0.016) (0.054) (0.025) (0.004) (0.029) ρ por different time periods −0.178∗∗∗ −0.063∗∗ −0.005 −0.063∗ 0.029∗∗∗ 0.141∗∗∗ (0.046) (0.020) (0.016) (0.027) (0.006) (0.030) ρ with 2 lags −0.033∗ −0.253∗∗∗ 0.010 −0.027 0.005∗ −0.024 (0.015) (0.016) (0.008) (0.021) (0.002) (0.023) Destination fixed-effects 0.233∗∗∗ −0.498∗∗∗ 0.596∗∗ 0.165∗∗∗ −0.015∗∗ 0.243∗∗∗ (0.016) (0.048) (0.222) (0.038) (0.005) (0.047) Year-destination fixed effects 0.092∗∗∗ −0.235∗∗∗ −0.242∗∗∗ −0.036 −0.009 0.183∗∗∗ (0.013) (0.016) (0.072) (0.025) (0.004) (0.036) Time-destination fixed effects 0.097∗∗∗ 0.158∗∗∗ – −0.039 −0.014∗∗∗ 0.336∗∗∗ (0.014) (0.032) – (0.027) (0.003) (0.054) Locations >100km apart 0.125∗∗∗ −0.498∗∗∗ 1.152∗ 3.805∗∗∗ −0.076∗∗∗ 4.051∗∗∗ (0.017) (0.048) (0.469) (0.150) (0.011) (0.502) Locations >200km apart 0.349∗∗∗ −0.420∗∗∗ −0.303∗ – −0.082∗∗∗ 6.960∗∗∗ (0.038) (0.106) (0.118) – (0.015) (1.992) Notes : Each cell reports the main coefficient from a regression of price gaps on (log) distance under alternative specifications, which are described in section 5.1. All regressions use the adjusted price gap between trading pairs and include time-product fixed effects multiplied by (1 − ρˆk ˆk od )/ρ od . Standard errors in parentheses. ∗∗∗ significant at 1%, ∗∗ at 5%, and ∗ at 10%. Source: Authors’ estimates based on national CPI data and distances calculated using Google. 6 Comparison to previous literature To put our results in perspective, we start by comparing them to those in Atkin and Donaldson (2015), from whom we borrow the methodology to recover trade costs. Their ˆ is equal to 0.037 for Ethiopia, 0.056 for Nigeria, and 0.011 for the U.S. Our baseline β estimates are slightly lower. In terms of ad valorem equivalents, their estimates imply transport costs of 25%, 15%, and 5% for Ethiopia, Nigeria, and the U.S., respectively.11 11 To make their results comparable, we use the distance between the most and least remote locations that we use to compute the results in Figure 3. 20 Our results for Nigeria are very similar (16.5%) to their estimates. Moreover, with the exception of Kenya (19.2%), the recovered trade costs in the rest of the countries are close to those in the U.S. When we consider the results based on the locally-produced goods samples, however, our results are more sensible and in line with Atkin and Donaldson (2015) for most coun- tries. Coefficients are between 2.6 (Tanzania) and 14 (Kenya) times larger than in the U.S. To try understand the sources of the differences in our estimates with those in Atkin and Donaldson (2015), we make a more precise comparison focusing on Nigeria, for which we have a similar sample covering the same time period, and re-estimate pass-through rates and trade costs restricting the sample to the goods that are included in Atkin and Donaldson (2015) (“AD-sample” from now on). The sample that we have access to in- cludes 15 products out of the original 18 products in their paper and the total number of observations is lower, at 25,189, compared to 26,025 in Atkin and Donaldson (2015). Our results produce a mean pass-through rate of 0.484 (s.e. 3.2) compared to 0.390 (s.e. 0.66) in Atkin and Donaldson (2015).12 Moreover, while Atkin and Donaldson (2015) estimate a significant negative relationship between the pass-through rate and distance (-0.101), we get positive relationship (0.01) which is not estimated with precision. These differences notwithstanding, the estimation of trade costs under the adjusted price gap specification produces identical results to Atkin and Donaldson (2015). That is, we get ˆ = 0.056 (0.007) compared to β β ˆ = 0.056 (0.008). We can also compare our results to previous attempts to recover transport costs in Rwanda using this methodology, albeit with different data, by Bougna et al. (2020). In- stead of using CPI data, Bougna et al. (2020) use market price data from November 2016 to July 2018 collected by the Government of Rwanda in collaboration with the World Bank in the context of an impact evaluation of a large feeder road rehabilitation program. They estimate a pass-through rate equal to 0.34, close to our estimate of 0.39, and β ˆ = 0.106, which is higher than our estimate using the full sample, but similar to our estimate using the sample of locally-produced goods only (0.120). 7 Conclusion We apply the methodology developed by Atkin and Donaldson (2015) to estimate intra- national transport costs, accounting for spatial variation in markups across a sample of Eastern European and African countries. Our results suggest significant heterogeneity in the magnitude of intra-national transport costs across the countries in our sample. More- over, we find that our estimates are sensitive to the choice of products, with most countries showing higher transport costs when we restrict the sample to locally produced goods, ex- 12 The percentage of pass-through rates below one is slightly higher, 93.1%, and we obtain a lower fraction of estimates below zero, 12.6%, compared to 17%. 21 cluding imported goods. We attribute these differences to a more accurate identification of production locations for locally produced goods, which is key for the methodology to produce reliable estimates of transport costs. Our results that rely on the samples of locally produced goods are in line with those in the previous literature. Compared to estimates for the U.S. in Atkin and Donaldson (2015), we find that intra-national trade costs are between 2.5 (Tanzania) and 14 (Kenya) times larger in our sample of countries. 22 A Appendix A.1 Data selection and processing Product selection The set of products included in the CPI in each country is intended to capture a typical consumption basket. To avoid concerns associated with unobserved spatial variation in quality, we focus on manufactured goods that can be “narrowly de- fined”, in the sense that a specific brand and presentation can be determined from the product’s description. Therefore, apart from services, such as housing rent or education services, we discard products that are sold in bulk or unbranded (e.g., “sorghum flour” or “bread, loaf”), products whose prices refer to consumption on premises (e.g., “Coca Cola, 500ml, served in a bar”), and products whose differentiation may go beyond the informa- tion provided in the description, such as clothing, electronics, and other durable goods. Typos and misspelling of brands are prevalent in the CPI product listings maintained by NSOs in low- and middle-income countries. We therefore normalize product descriptions, checking retailers’ or manufacturers’ websites. The types of products we end up with typically include packaged food and beverages, medicine or drugs available to the public at pharmacies, personal care products, and household essentials (e.g. laundry or cleaning supplies). In some cases, different presentations were included for the same product in different locations. For instance, “Coca Cola, 500ml” was included in half of the locations and “Coca Cola, 650ml” was included in the other half. In some countries with relatively few products satisfying our selection criteria, we adjusted prices to reflect size- or unit- equivalent prices applying the product’s unit price (e.g., price per ml). Finally, we left out products which were not broadly available across locations and periods, or products whose prices did not change at all over the sample period. Tables 7 to 12 describe the final set of products we consider for each country in our sample. Identification of origin locations After normalizing product descriptions and brands, manufacturers were identified. For local manufacturers, manufacturing plants were identi- fied through internet searches, verifying that establishments did not correspond to corpo- rate offices only. In some cases (mostly for beverages manufacturers), multiple manufac- turing plants were identified. In these cases, for each destination, the product was assumed to have been distributed from the closest plant. Imported products were identified either because product listings indicated so (in some cases specifying the product’s country of origin), or because the manufacturer does not have a production facility in the country. For imported products, origin locations are determined by identifying locations through which the product is more likely to have entered the country. For landlocked countries, origins were border crossings near the main ports that serve those countries. For countries with ports, origins of imported products were the main sea ports. These were verified using Logcluster.org, World Ports Source, Volza Exportgenius. 23 Additional cleaning Our main analysis is based on a cleaned sample of the data to correct for outlier and abnormal price observations. Specifically, we follow Atkin and Donaldson (2015) and remove price quotes that lie more than 10 standard deviations away from the log mean price of a product. Moreover, whenever a price quote looks abnormally high or low compared to the behavior of the price series for a particular product, we check whether nearby prices for that same product in that same period behaved similarly, and remove them if they did not. If a product has a missing price quote at its origin location in a single month, then we impute that month’s price with the average of the previous and consecutive months. In order to arrive at trade costs estimates in real terms and a comparable currency, we deflate all prices using inflation measured at the origin location and convert them to constant U.S. dollars, as follows. First, for every origin location-product-month price observation used in our analysis, we compute monthly inflation. Second, we construct a measure of monthly inflation at the origin by taking the unweighted mean of product- level inflation, and deflate all price quotes in our sample using this measure of inflation. Whenever a product price is not observed at the origin location in a given month, we use the pro-rated multi-month change between available monthly observations. Constant prices are converted into U.S. dollars using the prevailing exchange rate in each country during the first month of the sample. 24 References Aggarwal, Shilpa, “Do rural roads create pathways out of poverty? Evidence from India,” Journal of Development Economics, July 2018, 133, 375–395. Agnosteva, Delina E., James E. Anderson, and Yoto V. Yotov, “Intra-national trade costs: Assaying regional frictions,” European Economic Review, February 2019, 112, 32–50. Atkin, David and Dave Donaldson, “Who’s Getting Globalized? The Size and Im- plications of Intra-national Trade Costs,” 2015. Unpublished manuscript. Borraz, Fernando, Alberto Cavallo, Roberto Rigobon, and Leandro Zipitria, “Distance and Political Boundaries: Estimating Border Effects under Inequality Con- straints,” International Journal of Finance & Economics, 2016, 21 (1), 3–35. eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1002/ijfe.1517. eophile, Th´ Bougna, Th´ eophile Ewane, Maria Jones, and Florence Kondylis, “Trade Costs, Prices, and Connectivity in Rwanda,” 2020. Unpublished manuscript. Costinot, Arnaud and Dave Donaldson, “How Large Are the Gains from Economic Integration? Theory and Evidence from U.S. Agriculture, 1880-1997,” December 2016. sar, A. Kerem, Overland Transport Costs: A Review Policy Research Working Pa- Co¸ pers, The World Bank, August 2022. Duranton, Gilles, “Roads and trade in Colombia,” Economics of Transportation, March 2015, 4 (1), 16–36. Fajgelbaum, Pablo and Stephen Redding, “External Integration, Structural Trans- formation and Economic Development: Evidence from Argentina 1870-1914,” Technical Report, National Bureau of Economic Research, Cambridge, MA June 2014. Head, Keith and Thierry Mayer, “Chapter 3 - Gravity Equations: Workhorse,Toolkit, and Cookbook,” in Gita Gopinath, Elhanan Helpman, and Kenneth Rogoff, eds., Hand- book of International Economics, Vol. 4 of Handbook of International Economics, Else- vier, January 2014, pp. 131–195. Martin, Julien, Florian Mayneris, and Ewane Theophile, “The Price of Remote- ness: Product Availability and Local Cost of Living in Ethiopia,” March 2020. Mayneris, Florian, Julien Martin, and Ewane Theophile, “The Price of Remote- ness: Product Availability and Local Cost of Living in Ethiopia,” CEPR Discussion Paper Series, March 2020, (DP14515). 25 Porteous, Obie, “High Trade Costs and Their Consequences: An Estimated Dynamic Model of African Agricultural Storage and Trade,” American Economic Journal: Ap- plied Economics, 2019, 11 (4), 327–366. Sotelo, Sebastian, “Domestic Trade Frictions and Agriculture,” Journal of Political Economy, October 2019, 128 (7), 2690–2738. Publisher: The University of Chicago Press. Teravaninthorn, Supee and Gael Raballand, Le prix et le coˆ ut du transport en ´ Afrique: Etude des principaux corridors Directions in Development - Infrastructure, The World Bank, June 2009. 26 A.2 Product lists 27 Table 7: Product list, Kenya Product Brand and presentation Manufacturer Beer Tusker Lager Beer (Can), 500 ml/cc Kenya Breweries Ltd Breakfast Cereals Weetabix Breakfast Cereals, 225 gr Weetabix East Africa Ltd Cement Bamburi Cement, 50000 gr Bamburi Cement Ltd Cement Simba Cement, 50000 gr Simba Cement Company Cigarettes Sportsman Cigarettes, 20 units British American Tobacco (BAT) Kenya Limited Coffee Dormans Coffee, 100 gr Dormans Coffee Ltd Coffee Nescaf´e Coffee, 50 gr Nestl´e Kenya Limited Cooking Oil Fresh Fri Cooking Oil, 1000 ml/cc Pwani Oil Products Ltd Cooking Oil Golden Fry Cooking Oil, 1000 ml/cc Bidco Africa Cooking Oil Rina Cooking Oil, 1000 ml/cc Kapa Oil Refineries Ltd 28 Milk - Powdered Miksi Powder Milk, 250 gr Promasidor Kenya Ltd Soap/Detergent - Laundry Omo Laundry Powder Soap, 500 gr Unilever Kenya Ltd Soap/Detergent - Laundry Sunlight Laundry Soap, 900 gr Unilever Kenya Ltd Soft Drinks Coca-Cola Soft Drink, 500 ml/cc Coca-Cola Beverages Africa - Kenya Sugar Kabras Sugar, 1000 gr West Kenya Sugar Co. Ltd (Kabras Sugar) Tea Baraka Chai Tea, 100 gr Gold Crown Beverages Ltd Tea Fahari Ya Tea, 100 gr Kenya Tea Packers (Ketepa) Water Dasani Mineral Water, 500 ml/cc Coca-Cola Beverages Africa - Kenya Yoghurt Delamere Yoghurt, 450 ml/cc Brookside Dairy Limited Yoghurt Ilara Yoghurt, 500 ml/cc Ilara Dairy Products Notes : This table presents the list of products used in the sample, by country. Source: Authors’ based on national CPI data. Table 8: Product list, Madagascar Product Brand and presentation Manufacturer Baby Cereal Baby Cereal Farilac, 200 gr Socolait Beer Beer THB, 660 ml/cc Brasseries Star (Groupe Castel) Cheese Cheese Le Vache Qui Rit, 140 gr Imported Chocolate - Cocoa Powder Cocoa Powder Rob’Quick, 250 gr Chocolaterie Robert Madagascar 29 Cigarettes Cigarettes Good Look, 1 units Imperial Tobacco Madagascar Coffee Coffee Tsy Lefy, 90 gr Taf Madagascar Margarine Margarine Marna Gold, 250 gr Tiko Oil Refinery & Margarine Factory Medicines - Aspirin Aspirine Farmad, 1 units Laboratoires Pharmaceuitiques Malgaches Farmad SA Pasta Pasta Matsiro, 150 gr Salone Notes : This table presents the list of products used in the sample, by country. Source: Authors’ based on national CPI data. Table 9: Product list, Nigeria Product Brand and presentation Manufacturer Baby Food Cerelac Baby Food, 400 gr Nestl´e Nigeria PLC Beer Guinness Stout Beer, 300 ml/cc Guinness Nigeria (Diageo) Canned Sardines Titus Sardines, 125 gr Imported (Niger border crossing) Cement Elephant Cement, 50000 gr Lafarge Africa PLC Chocolate - Cocoa Powder Milo Drink, 450 gr Nestl´e Nigeria PLC Chocolate - Drink Bournvita Chocolate Drink, 450 gr Cadbury Nigeria PLC Coffee e Coffee, 50 gr Nescaf´ Nestl´e Nigeria PLC Cookies / Biscuits Cabin Biscuit, 800 gr Niger Biscuit Co Ltd Margarine Blue Band Margarine, 400 gr Unilever Nigeria PLC Medicines - Liver Salt (laxative/antacid) Andrew’s Liver Salt, 5 gr GlaxoSmithKline Consumer Nigeria PLC Medicines - Paracetamol Panadol, 1 units GlaxoSmithKline Consumer Nigeria PLC Medicines - Vitamins Multivite Vitamins (Tablets), 1 units GlaxoSmithKline Consumer Nigeria PLC 30 Milk - Evaporated Peak Evaporated Milk, 170 gr FrieslandCampina WAMCO Nigeria PLC Milk - Powdered Peak Skimmed Milk, 400 gr FrieslandCampina WAMCO Nigeria PLC Paint Nigerlux Paint, 4000 ml/cc IPWA PLC Pens Bic Pen, 1 units Nigerian Ball Point Pen (NIPEN) Industries Limited Semolina Golden Penny Semovita Semolina, 2000 gr Golden Penny Foods (Flour Mills of Nigeria PLC) Soap - Toilet/Bath Lux Toilet Soap, 1 units Unilever Nigeria PLC Soap/Detergent - Laundry Key Soap, 1 units Unilever Nigeria PLC Soap/Detergent - Laundry Omo Laundry Detergent, 200 gr Unilever Nigeria PLC Soft Drinks Coca-Cola Soft Drink (Bottle), 350 ml/cc Coca-Cola Nigeria (multiple bottlers) Soft Drinks Coca-Cola Soft Drink (Can), 350 ml/cc Coca-Cola Nigeria (multiple bottlers) Sugar St Louis Sugar, 1 units Imported (sea port) Tea Lipton Tea Bags, 25 units Unilever Nigeria PLC Tomato Paste De Rica Tomato Paste, 70 gr Olam Nigeria Limited Notes : This table presents the list of products used in the sample, by country.Source: Authors’ based on national CPI data. Table 10: Product list, Rwanda Product Brand and presentation Manufacturer Baby Food Cerelac Baby Food, 400 gr Imported (through Uganda) Baby Food Quaker Baby Food, 500 gr Imported (through Uganda) Beer Amstel Malt Beer, 330 ml/cc BraLiRwa Brewery (Coca Cola license) Beer Mutzig Beer, NA BraLiRwa Brewery (Coca Cola license) Beer Primus Beer, NA BraLiRwa Brewery (Coca Cola license) Cement Cimerwa Cement, 50000 gr Cimerwa Cement Limited Cheese Gishwati Cheese, 1000 gr Ingabo Dairy Company Ltd Cigarettes Intore Cigarettes, 20 units TabaRwanda (British American Tobacco) Cigarettes Sweet Menthol Cigarettes, 20 units TabaRwanda (British American Tobacco) Cleaner - Multipurpouse Jik Cleaner, 750 ml/cc Imported (through Uganda) Cleaner - Multipurpouse Vim Powder Cleaner, 500 gr Imported (through Uganda) 31 Cleaner - Toilet Harpic Cleaner, 500 ml/cc Imported (through Uganda) Coffee CafeRwa Coffee, 500 gr CafeRwa Ltd Cookies / Biscuits Golden Biscuits, 33 gr ADMA International Ltd Cookies / Biscuits Riham Biscuits, 50 gr ADMA International Ltd Flour - Corn Maganjo Corn Flour, 5000 gr Imported (through Uganda) Flour - Wheat Azam Wheat Flour, 2000 gr Bakhresa Group Flour - Wheat Pembe Wheat Flour, 2000 gr Pembe Flour Mills Ltd Margarine Blue Band Margarine, 250 gr Imported (through Uganda) Milk Inyange Milk, 500 ml/cc Inyange Industries Milk - Powdered Guigoz Powdered Milk, 400 gr Imported (through Uganda) Milk - Powdered Linda Powdered Milk, 400 gr Imported (through Uganda) Milk - Powdered Nido Powdered Milk, 400 gr Imported (through Uganda) Notes : This table presents the list of products used in the sample, by country. Source: Authors’ based on national CPI data. Table 10: Product list, Rwanda (cont.) Product Brand and presentation Manufacturer Oil - Cooking - Vegetable Golden Vegetable Oil, 1000 ml/cc Imported (through Uganda) Pasta Santa Lucia Spaghetti, 500 gr Imported (through Uganda) Pens Bic Pen, 1 unit Imported (through Uganda) Salt Nezo Salt, 600 gr Imported (through Uganda) Shampoo Movit Shampoo, 1000 ml/cc Imported (through Uganda) Soap - Cleaning Sulfo Cleaning Soap, 5000 ml/cc Sulfo Industries Soap - Toilet/Bath Dawa Bath Soap, 100 gr Sulfo Industries Soap - Toilet/Bath Dettol Bath Soap, 100 gr Imported (through Uganda) Soap - Toilet/Bath Fa Bath Soap, 100 gr Imported (through Uganda) Soap - Toilet/Bath Giv Bath Soap, 80 gr Imported (through Uganda) 32 Soap - Toilet/Bath Imperial Leather Bath Soap, 100 gr Imported (through Uganda) Soap - Toilet/Bath Protex Bath Soap, 100 gr Imported (through Uganda) Soft Drinks Fanta Soft Drink, 330 ml/cc BraLiRwa Brewery (Coca Cola license) Tea Sorwathe Tea, 250 gr Sorwathe Tea Factory Toilet Paper Clear Toilet Paper, 1 unit Trust Industries Toilet Paper Supa Toilet Paper, 1 unit Imported (through Uganda) Tomato Paste Sorwatom Tomato Paste, 70 gr Sorwatom Water Akandi Mineral Water, 500 ml/cc Sina Gerard/Ese Urwibutso Water Inyange Mineral Water, 500 ml/cc Inyange Industries Yoghurt Inyange Yogurt, 250 ml/cc Inyange Industries Notes : This table presents the list of products used in the sample, by country. Source: Authors’ based on national CPI data. Table 11: Product list, Tanzania Product Brand and presentation Manufacturer Baking Powder Simba Baking Powder, 500 gr Chemicotex Beer Castle Beer, 500 ml/cc Tanzania Breweries Limited (ABInBev) Beer Safari Beer, 500 ml/cc Tanzania Breweries Limited (ABInBev) Beer Serengeti Beer, 500 ml/cc Serengeti Breweries Limited Chocolate - Cocoa Powder Primo Cocoa Powder, 500 gr Imported from Kenya Cigarettes Nyota Cigarettes, 20 units Tanzania Cigarette Company Ltd. (JTI) Cigarettes Portsman Cigarettes, 20 units Tanzania Cigarette Company Ltd. (JTI) Cigarettes Sweet Menthol Cigarettes, 20 units Tanzania Cigarette Company Ltd. (JTI) Coffee Africafe Instant Coffee, 100 gr Afri Tea and Coffee Blenders 1963 Ltd Cookies / Biscuits Nice Biscuits, 200 gr Imported from Rwanda Cooking Fat Kimbo Cooking Fat, 1000 ml/cc Imported from Kenya Fruit Jams Zesta Fruit Jam, 500 gr Imported from Kenya Fruit Juices Maaza Fruit Juice, 1000 ml/cc The Coca Cola Company Ice cream Azam Ice Cream, 100 gr Bakhresa Group Konyagi (Spirit beverage) Konyagi Spirit Beverage, 500 ml/cc TDL Tanzania Distilleries Ltd, Konyagi Margarine Blue Band Margarine, 500 gr Imported from Kenya 33 Matches Kangaroo Matches, 1 units Imported (sea port) Medicines - Cough Syrup Koflyn Cough Syrup, 100 ml/cc Shelys Pharmaceuticals Limited Medicines - Paracetamol Panadol Paracetamol, 2 tablets GSK GlaxoSmithKline Oil - Cooking - Sunflower Sundrop Sunflower Oil, 1000 ml/cc Murzah Oil Mills Ltd Pens Bic Pen, 1 units Imported (sea port) Shoe polish Kiwi Shoe Polish, 100 ml/cc Imported from Kenya Soap - Toilet/Bath Giv Toilet Soap, 80 gr Imported (sea port) Soft Drinks Coca-Cola Soft Drink, 350 ml/cc The Coca Cola Company Soft Drinks Fanta Soft Drink, 350 ml/cc The Coca Cola Company Soft Drinks Mirinda Soft Drink, 350 ml/cc SBC Tanzania Limited (Pepsi Kipawa) Soft Drinks Pepsi Soft Drink, 350 ml/cc SBC Tanzania Limited (Pepsi Kipawa) Tea Chai Bora Tea Leaves, 100 gr Chai Bora Limited Toothpaste Colgate Toothpaste, 75 gr Colgate Palmolive Tanzania Ltd Water Dasani Spring Water, 500 ml/cc The Coca Cola Company Water Kilimanjaro Spring Water, 500 ml/cc Bonite Bottlers Limited Wine Drostdy-hof Red Wine, 750 ml/cc Imported (sea port) Notes : This table presents the list of products used in the sample, by country. Source: Authors’ based on national CPI data. Table 12: Product list, Georgia Product Brand and presentation Manufacturer Baby Food Hipp Baby Food (Box), 400 gr Imported (seaport) Baby Food Humana Baby Food (Box), 400 gr Imported (seaport) Beer Natakhtari Beer, 500 ml/cc Natakhtari Brewery Beer Zedazeni Beer, 500 ml/cc Zedazeni Beer Company Butter Sante Butter 60Cheese Sante Cottage Cheese 6Chocolate Barambo Dark Chocolate, 100 gr Barambo Ltd Coffee Jacobs Monarch Instant Coffee, 100 gr Imported (seaport) Margarine Ona Margarine, 1000 gr Imported from/through Turkey Milk Sante 2.5Oil - Cooking - Sunflower Oleina Sunflower Oil, 1000 ml/cc Imported from/through Russia 34 Pasta Divella Macaroni, 1000 gr Imported (seaport) Rice Supremo Rice, 1000 gr Imported (seaport) Soft Drinks Coca-Cola Soft Drink, 1500 ml/cc Coca-Cola Bottlers Georgia Soft Drinks Pepsi Soft Drink, 1500 ml/cc Pepsi (Iberia Refreshment) Tea Gurieli Tea, 50 gr Gurieli Tea Toilet Paper Selpak Toilet Paper, 1 units Imported from/through Turkey Tomato Paste Kula Tomato Paste, 1000 gr Gori Feeding Cannery Kula Tomato Paste Marneuli Tomato Paste, 1000 gr Marneuli Food Factory Water Bakhmaro Still Mineral Water, 500 ml/cc Water Margebeli (JSC Healthy Water) Wine Teliani Valley Saperavi Red Wine, 750 ml/cc Teliani Valley Winery Notes : This table presents the list of products used in the sample, by country. Source: Authors’ based on national CPI data.