NBER WORKING PAPER SERIES AIRPLANES AND COMPARATIVE ADVANTAGE James Harrigan Working Paper 11688 http://www.nber.org/papers/w11688 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 Massachusetts Avenue Cambridge, MA 02138 October 2005 This paper has benefited from audience comments at Ljubljana, Illinois, Michigan, Columbia, the World Bank, NBER, CEPR, Hitotsubashi, Tokyo, City University of Hong Kong, and Hong Kong University of Science and Technology. I thank Christina Marsh and Geoffrey Barrows for excellent research assistance. The views expressed herein are those of the author(s) and do not necessarily reflect the views of the National Bureau of Economic Research. © 2005 by James Harrigan. All rights reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission provided that full credit, including © notice, is given to the source. Airplanes and Comparative Advantage James Harrigan NBER Working Paper No. 11688 October 2005, Revised July 2010 JEL No. F1 ABSTRACT Airplanes are a fast but expensive means of shipping goods, a fact which has implications for comparative advantage. The paper develops a Ricardian model with a continuum of goods which vary by weight and hence transport cost. Comparative advantage depends on relative air and surface transport costs across countries and goods, as well as stochastic productivity. A key testable implication is that the U.S. should import heavier goods from nearby countries, and lighter goods from faraway counties. This implications is tested using detailed data on U.S. imports from 1990 to 2003. Looking across goods the U.S. imports, nearby exporters have lower market share in goods that the rest of the world ships by air. Looking across exporters for individual goods, distance from the US is associated with much higher import unit values. These effects are large, which establishes that the model identifies an important influence on specialization and trade. James Harrigan Department of Economics University of Virginia P.O. Box 400182 Charlottesville, VA 22904-4182 and NBER [email protected] 1 Introduction Countries vary in their distances from each other, and traded goods have differing physical characteristics. As a consequence, the cost of shipping goods varies dramatically by type of good and how far they are shipped. A moment’s reflection suggests that these facts are probably important for understanding international trade, yet they have been widely ignored by trade economists. In this paper I focus on one aspect of this set of facts, which is that airplanes are a fast but expensive means of shipping goods. The fact that airplanes are fast and expensive means that they will be used for shipping only when timely delivery is valuable enough to outweigh the premium that must be paid for air shipment. They will also be used disproportionately for goods that are produced far from where they are sold, since the speed advantage of airplanes over surface transport is increasing in distance. In this paper I show how these considerations can be incorporated into the influential Eaton and Kortum (2002) model of comparative advantage. In this general model, differences across goods in transport costs (both air and surface) and the value that consumers place on timely delivery interact with relative distance to affect global trade patterns. The model of the paper delivers two empirical implications that I test using highly disaggregated data on all U.S. imports from 1990 to 2003. The first implication is that nearby trading partners (Canada and Mexico) should have lower market shares in goods that more distant trading partners ship primarily by air. The second implication is that goods imported from more distant locations will have higher unit values. Both of these implications are resoundingly confirmed, and the size of the effects is economically important. In short, I find that the relative distance and relative transport cost effects emphasized in the model are an important influence on U.S. trade. Finally, I show that air shipment is much more likely for goods that have a high value/weight ratio. There is a small, recent literature that looks at some of the issues that I analyze in this paper. The most direct antecedents of my paper are Limao and Venables (2002) and Hummels and Skiba (2004). Limao and Venables (2002) is a theory paper that models the interaction between specialization and trade costs, illustrating how the equilibrium pattern of specialization involves a tradeoff between comparative production costs and comparative transport costs. The geographical structure has a central location that exports a numeraire good and imports two other goods from more remote locations. These more remote locations have a standard 2  2 1 production structure, and when endowments are the same at all locations and transport costs are the same for both goods the model reduces to one where greater distance from the center has simple effects on aggregate welfare: more distant countries are poorer because they face higher transport costs. When endowments and transport costs differ the analysis becomes more nuanced, with relative transport costs interacting with relative endowments to determine welfare and comparative advantage (for example, a relatively centrally located country that is abundant in the factor used intensively in the low trade cost good will have high trade volumes and high real GDP, while countries that are more distant, and/or that are abundant in the factor used intensively in the high transport cost good, will have lower trade volumes and real GDP). This rich theoretical framework is not evaluated empirically in the paper, nor to my knowledge has it been taken to the data in subsequent work. In contrast to Limao and Venables (2002), the paper by Hummels and Skiba (2004) is mainly empirical. Like Limao-Venables the focus is on the implications of differences in transport costs across goods on trade patterns, but unlike Limao-Venables (and virtually all of trade theory) they challenge the convenient assumption that transport costs take the iceberg form. Hummels-Skiba show that actual transport costs are much closer to being per-unit than iceberg, and they use simple price theory to show the implications for trade: imports from more distant locations will have disproportionately higher f.o.b. prices. This implication is strongly confirmed using a large dataset on bilateral product-level trade. As the model of the paper is partial equilibrium, Hummels-Skiba do not address the equilibrium location of production. A key theoretical motivation to my analysis below is Deardorff (2004). Deardorff works with a series of simple models to make a profound point about trade theory in a world of transport costs: “local comparative advantage” (defined as autarky prices in comparison to nearby countries rather than the world as a whole) is what matters in determining trade in a world with trade costs. I embed this insight into the Eaton and Kortum (2002) model of Ricardian comparative advantage in what follows. A related literature is the “new economic geography”, which is well-summarized in Fujita, Krugman, and Venables (1999) and Baldwin, Forslid, Martin, Ottaviano, and Robert- Nicoud (2003). In new economic geography models, the interaction between increasing returns and transport costs are a force for agglomeration, and through this channel trade costs influence trade patterns. The mechanism in these models is quite different from the comparative advantage 2 mechanism in Limao and Venables (2002) and Eaton and Kortum (2002). David Hummels has written a series of important empirical papers that directly motivated this paper. Hummels (1999) shows that ocean freight rates have not fallen on average since World War 2, and have often risen for substantial periods. By contrast, the cost of air shipment has fallen dramatically. Chart 1 shows that these trends have continued since 1990, with the relative price of air shipping falling 40% between 1990 and 2004. Hummels (2001a) shows that shippers are willing to pay a large premium for faster delivery, a premium that has little to do with the interest cost of goods in transit1. Hummels (2001b) analyzes the geographical determinants of trade costs, and decomposes the negative effect of distance on trade into measured and unmeasured costs. The following section illustrates some key features of U.S. imports by product, trading partner, and transport mode from 1990 to 2003. Section 3 presents the theory, which is then formally tested in section 4. 2 Airplanes and U.S. imports: a first look The import data used in this paper are collected by the U.S. Customs Service and reported on CD-ROM. For each year from 1990 to 2003, the raw data include information on the value, quantity (usually number or kilograms), and weight (usually in kilograms) of U.S. imports from all sources. The data also include information on tariffs, transport mode and transport fees, including total transport charges broken down by air, vessel and (implicitly) other, plus the quantity of imports that come in by air, sea, and (implicitly) land.2 The import data are reported at the 10-digit Harmonized System (HS) level, which is extremely detailed, with over 14,000 codes in 2003. I aggregate the 10-digit import data for analysis in various ways. For most of the descriptive charts and tables, I work with a broad aggregation scheme that updates Leamer’s (1984) classification, which is reported in Table 1. Countries are aggregated by distance and by region, as described in Appendix Table A1. Distance from the United States is measured in 1 By “the interest cost of goods in transit”, I mean the financial cost of having goods in transit before they can be sold. This opportunity cost equals the value of the good  daily interest rate  days in transit. 2 “other” transport modes include truck and rail, and are used exclusively on imports from Mexico and Canada. 3 kilometers from Chicago to the capital city of each country.3 Table 1 illustrates the great heterogeneity in the prevalence of air freight for U.S. imports, as well as some important changes over the sample. Many products come entirely or nearly entirely by surface transport (oil, iron and steel, road vehicles) while others come primarily by air (computers, telecommunications equipment, cameras, medicine). Scanning the list of products and their associated air shipment shares hints at the importance of value to weight and the demand for timely delivery in determining shipment mode. Charts 2 and 3 illustrate the variation in air freight across regions and goods (the regional aggregates are defined in Appendix Table A1).4 Chart 2 shows that about a quarter of US (non-oil) imports arrived by air in 2003, up from 20% in 1990 (for brevity, in what follows I’ll call the proportion of imports that arrive by air “air share”). Excluding NAFTA, the non-oil air share was 35% in 2003. Chart 3 shows that this average conceals great regional variation, which is related to distance: essentially no imports come by air from Mexico and Canada, while Europe’s air share is almost half by 2003, up from under 40% in 1990. East Asia’s air share increased by about half from over the sample, from 20 to 30%. 3 Airplanes and trade: theory The data reviewed in the previous section clearly suggest the influence of distance and transport costs on the pattern of trade. In this section I develop a model than can be used to analyze the effects of transport costs on comparative advantage. My basic framework comes from Eaton and Kortum (2002), simplified in some dimensions and made more complex in others. On the demand side, consumers value timely delivery, and this valuation can differ across goods. On the supply side, timely delivery can be assured in two ways: by surface transport from nearby suppliers, or by air transport from faraway suppliers. Since air transport is expensive, it will only be used by distant suppliers, and on goods which have both a high demand for timely delivery and a high value/weight ratio (and thus a relatively small cost premium for air shipment). I derive two testable empirical implications from the model. The first implication is about the cross section of imported goods: nearby exporters will have a smaller market share in 3 A convenient source for the distance data is http://www.macalester.edu/~robertson/index.html 4 goods that faraway exporters send by air. The second implication concerns the distribution of unit values for a particular good: faraway exporters will sell goods which have on average have a higher unit value and thus lower transport costs as a share of value. 3.1 Demand For many transactions, timely delivery is available for a substantial premium over regular delivery. Why would anybody pay such a premium? Possible answers to this question are analyzed in a few recent papers. Evans and Harrigan (2005) derive the demand for timely delivery by retailers, who benefit from ordering from their suppliers after fickle consumer demand is revealed. Evans and Harrigan show the empirical relevance of this channel using data on the variance of demand and the location of apparel suppliers: for goods where timely delivery is important, apparel suppliers to U.S. retailers are more likely to be located in Mexico, where timely delivery to the U.S. market is cheap, while goods where timeliness is less important are more likely to be located in more distant, lower-wage countries such as China. Harrigan and Venables (2006) focus on the importance of the demand for timeliness as a force for agglomeration. They analyze this question from a number of angles, including a model of the demand for “just in time” delivery. The logic is that more complex production processes are more vulnerable to disruption from faulty or delayed parts, with the result that the demand for timely delivery of intermediate goods is increasing in complexity of final production. While the details of demand and supply differ across models, the message of Evans and Harrigan (2005) and Harrigan and Venables (2006) is that it is uncertainty that generates a willingness to pay a premium for timely delivery. For the purposes of the present paper I will model this result with a simple shortcut, and suppose that utility is higher for goods that are delivered quickly. Looking ahead, timely delivery can be assured in one of two ways: by proximity between final consumers and production, or by air shipment when producers are located far from consumers. The determination of the equilibrium location of producers is a central concern of the model. There is a unit continuum of goods indexed by z, with consumption denoted by x(z). Utility is Cobb Douglas in consumption, and the extra utility derived from timely or “fast” 4 The online appendix includes additional tables charts which show variation by product group. The product aggregates correspond to the headings in Table 1. 5 delivery is f(z) >1.5 Letting F denote the set of goods that are delivered in a timely matter (for brevity I will call these “fast goods”), utility is given by lnU   lnf zxzdz  lnxzdz (1)   zF zF Order goods so that F 0,z0,1. Then the utility function can be informatively re-written as z 1 lnU ln f zdzlnxzdz 0 0 For nominal income Y, the resulting demand functions are Y xz pz That is, all goods have the same expenditure share, regardless of whether or not they are fast. Denoting the prices of fast goods with a superscript f, the indirect utility function is z z 1 lnVp,Yln f zdzlnY ln pf zdzln pzdz 0 0 z Changing the set of fast goods at the margin has the following effect on utility, lnVp,Y ln f zln pf zln pz z which is positive iff pf z f z pz This inequality implies that consumers will prefer fast delivery of a good if and only if the marginal utility of timeliness exceeds the relative price of fast delivery. 3.2 Supply: shipping mode and geography Atomistic producers are assumed to be perfectly competitive, which ensures that FOB price equals unit cost, but there is a choice of shipping mode (air or surface) and consequent CIF price 5 Eaton and Kortum (2002) assume CES preferences. Since the elasticity is substitution plays no role in the solution of their model or mine, I use Cobb-Douglas preferences for simplicity. 6 paid.6 Shipping costs are of the iceberg form, so that for one unit to arrive t(z) ≥ 1 units must be shipped. I now introduce distance into the model. Denote air and surface iceberg shipping costs from origin country o to destination country d respectively as a (z) and s (z), and assume that od od a (z) > s (z) ≥ 1 z: air shipping is never cheaper than surface shipping. If producers are od od located near consumers, then (by assumption) they can achieve timely delivery using surface shipment. If producers are located far away from consumers, then they must decide if the extra expense of air shipment is worthwhile. The answer is yes if consumer preference for fast delivery f(z) is higher than the relative cost of air shipment, a z s z. Given the structure of costs od od and demand, the equilibrium shipping mode for producers located far from their customers is a z t zs z if 1 od od od s z f z od a z t za z if 1 od od od s z f z od a z Now order z so that od is monotonically weakly increasing in z, and define z as the s z f z od od implicit solution to7 a z  od od 1 (2) s z  f z  od od od By the ordering of z, z  z the optimal shipping mode is air and for all other goods the od optimal mode is surface. For every bilateral trade route from origin country o to destination country d, there will be a cutoff z . This cutoff doesn’t depend on wages or technology, only on bilateral transport od costs. As is traditional in trade models, I will assume that preferences, including the demand for timeliness schedule f(z), do not vary across countries, so that bilateral variation in transport costs determine which goods are shipped by surface and which by air. 6 FOB stands for “free on board”, and refers to the price of the good before transport costs are added. CIF stands for “cost, insurance, and freight”, and refers to the price after transport costs have been added. 7 3.3 Supply - competition The supply side of the model is based on Eaton and Kortum (2002), henceforth EK. Labor is the only factor of production, and is paid a wage w. Labor productivity in good z in country o, b z, is a random variable drawn from a Fréchet distribution with parameters T > 0 and  > 1. o o As in EK, competition depends on the CIF price, but here the relevant price is timeliness- adjusted: a country may win the market in a good by virtue of timely delivery rather than the lowest nominal CIF price. For each good and each bilateral route we know the optimal shipping decision from the discussion above, so it will be easy to specify the probability that o will win the competition in d. Let C denote the set of close country pairs, such that timely delivery is possible without air shipment. Define the timeliness-adjusted iceberg t z as od  a z t zmins z, od  o,dC (3a) od od f z   s z t z od o,dC (3b) od f z w Perfect competition implies that the FOB price is unit cost, which is o Then the timeliness- b z o adjusted CIF supply price p z is od w t z p z o od od b z o Country o will win the competition to sell good z in market d if it has the lowest timeliness- adjusted CIF price among all N countries, that is, if p zminp z,,p z od  1d Nd  As with EK, the probability that this happens is the probability that all the other prices on offer are greater than p z. The cdf of b is od o F b;zPrB zb expT b o  o  o 7 For ease of exposition, I make the innocuous assumption that z is unique. 8