Bilateral transport cost, infrastructure, common bilateral ties and political stability.

• Author: Molina, Danielken
• Position: Report
• Pages: 221(27)

1. Introduction

   Transport Costs are one of the most important sources of barriers to trade. As Obstfeld and Rogoff (2001) explain, several puzzles can be explained by their existence. But as Hummels (2006) points out, the literature has not yet devoted enough attention to try to model the determinants of transport costs. Previous approaches have modeled them, but the approaches are completely ad-hoc and they are characterized by heuristic functional specifications. Therefore, we believe that future work can be developed in this area.

   Given this concern, we use an ad-hoc empirical model to examine the possible determinants of transport costs, inspired specially by the fact that they have been decreasing since the early 1980's. (1) In particular, we estimate the effect of infrastructure and distance on bilateral transport costs. Using a country aggregate bilateral transport costs database for 1990, Limao and Venables (2001) (henceforth LV) estimate that the deterioration of the infrastructure index of a country from the median level to the 75th percentile raises transport costs by 12 percentage points. This effect implies a reduction in trade volume by around 28 percent.(2)

   Following the same approach, we replicate LV results, and we find that the effect of his infrastructure index on transport costs is not robust. Adopting Micco (2004) and Micco and Serebrinzky (2005) we calculate two alternative measures of infrastructure which we consider are a better proxy of the port infrastructure of a country. In addition, we estimate the effect of common bilateral ties, country political stability and open sky agreements on transport costs.

   We found that in addition to distance, infrastructure, political stability, common bilateral ties and open sky agreements are other important channels through which transport costs can be reduced.

   The structure of this paper is as follows: Section II sets up the model and the empirical estimation strategy that LV used. Section III performs a detailed description of the required data. In particular, the section describes how we calculate the infrastructure measures. Section IV describe the results obtained. First, we replicate the results obtained by LV in tables 2, 3, 5, 6, 7 and 8. Second we expand the data, and we perform panel data estimates of the effects with the LV infrastructure index, and our new infrastructure variables and the new sets of controls. Section V concludes.

2. The Model

   Following LV, we know that the unit costs of shipping a good ([T.sub.ij]) from country i to country j in period t is given by the following equation: (3)

   [T.sub.ij] = T ([x.sub.ij], [X.sub.i], [X.sub.j], [[mu].sub.ij]) (1)

   Where [x.sub.ij] vector of characteristics related to the journey between country i and country j. [X.sub.i] is a vector of characteristics related to country i. [X.sub.j] is a vector of characteristics related to country j, and [[mu].sub.ij] is a vector of non-observables.

   Following the trade literature,(4) LV used the bilateral distance between country i and county j and a border dummy variable as the journey specific variables defined in [x.sub.ij.] As expected, the higher the distance between the two destinations the higher the transport cost are, and second, when two countries share a common border, we would expect that other factors besides distance might reduce transport costs.

   First, neighboring countries might have more integrated transport networks that reduce transport costs. Second, neighboring countries might have similar customs agreements or transit rules that decrease the amount of time expended in transporting goods between destinations. This implies a reduction in the shipping and insurance costs charged per good. Third, the higher the trade volume among neighboring countries, the lower the fixed costs are. (5) The reason is due to the fact that the high volume of trade between two locations reduces the fixed costs shared by the senders within the two destination points because of cargo backhauling. (6)

   As for the country specific characteristics defined in [X.sub.i] or [X.sub.j], LV only focus their attention on geographical and infrastructure measures. For geographical characteristics, LV controls if either the country is an island or not, and they also control whether the country is landlocked or not. (7)

   The motivation to focus on these two measures is given by the fact that landlocked countries might have higher transport costs since they do not have ports; therefore, cargo must first pass through neighboring countries first, which increases the time of transportation and in consequence increases transport costs. In relation to islands, they expect that islands may have better ports infrastructure which implies a reduction in transport costs.

   Since LV use information at country level (8) obtained from the Directional Trade Data (DOTS), the ad valorem transport cost factor between country i and country j is defined as the ratio between C.I.F and F.O.B trade values. (9) Then, our measure of transport costs is determine by the following equation:

   [t.sub.ij] = ci[f.sub.ij]/f[ob.sub.ij] = [[p.sub.ij] + [T.sub.ij]]/[p.sub.ij] = t([x.sub.ij], [X.sub.i], [X.sub.j], [[??].sub.ij]) (2)

   Under the assumption that [t.sub.ij] can be approximated by a log linear function, we can proxy transport cost factors between country i and country j by the following equation: (10)

   [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (3)

   Where we assume that [[omega].sub.j] is uncorrelated with the explanatory variables previously defined in [x.sub.ij], [X.sub.i] and [X.sub.j.] As we discuss in the following section, there are a couple of factors that should be taken into account before advancing to estimate equation (3). First, the results obtained by LV are based on the strong assumption that transport costs are a linear function of the variables defined. This can be tested by introducing some non linear relationships with the infrastructure variable per country. Since we are following LV, we do not test for the non linear relationship of the infrastructure indexes. Second, Yeats (1978) and Hummels (2006) point out transport cost data proxy by the matched partner technique (11) has an important portion of its own variance explained by noise. Third, since DOTS data is obtained from three different sources, it is subject to data differences that are not related to trade costs. Certain statistical offices may value goods differently, i.e. fluctuation of the exchange rate within the time that it takes to transport the good between location may determine different valuation methods for products in C.I.F and FOB values respectively. Fourth, discrepancy in levels of C.I.F and FOB among matched country data can induce to high ad valorem transport costs. Fifth, DOTS data for a single year for a given country might change within different publications. Although the change is small in levels, it's implication on the ad valorem transport cost rate (as defined in equation 2) could be important.

3. Data

   Since this paper replicates tables 2, 3, 5, 6, 7 and 8 from LV, the variables that we need to replicate the results obtained by the authors are: ad valorem transport costs, bilateral distance between countries i and j, common border dummy, island dummy for countries i and j, GDP and GDP per capita for countries i and j, infrastructure index for countries i and j, landlocked dummy, African country dummy for countries i and j and rule of law and control of corruption of countries i and j.

   Ad valorem transport costs ([t.sub.ij]) defined by equation (2) were directly calculated from DOTS trade data for year 2005. As explained by Hummels (2006), this data has the advantage that it has a very good coverage, it has information of bilateral imports and exports at C.I.F and F.O.B trade values for almost all the countries of the world from 1948 up to the present (with a lag of two years). But as we mentioned before, it has some problems as well. The IMF builds up this data from several sources, which implies that the data has variations that are not due to transport costs. So when we proxy [t.sub.ij] by equation (2), part of the variation in [t.sub.ij] is not due to transport costs per se.

   Differences across countries can be due to different valuations of trade; i.e. at F.O.B a country can report value of good in the ship, and another can report the value of the good before entering the ship, the U.S. values imports at the exchange rate due the same day the product enters the first port, but other countries use the average exchange rate within the period that it took the good to arrive from country i to country j. Other differences can be explained by the country quality of the data. This bias is particularly important for developing countries where the quality of data is usually lower than in developed countries.

   Another source of differences are the re-adjustments made by the IMF to clarify errors and mistakes from previous years. So even though we access the raw data to calculate [t.sub.ij], our measure by definition is different from the measure used by LV because IMF has corrected the numbers.

   In addition, IMF has a rule of imputing values for imports and exports. If data is available for F.O.B exports but not for C.I.F. imports, for countries i and j, then it replaces the C.I.F missing values by a factor of 1.1 respect to F.O.B, and if there is data on C.I.F but not on F.O.B, then it replaces the missing F.O.B discounting 9% on C.I.F data.

   Finally, the data reports values for imports and exports near zero, which could lead to over estimation of the transport cost. According to the summary statistics reported by LV, [t.sub.ij] ad valorem transport costs are around 87% and 400% which is a very good example of the noise of the data used by LV.

   To address this problem, first, we do not take into account the bilateral import and export data that has the imputation rule previously described...