Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Capabilities, Wealth and Trade John Sutton London School of Economics and Daniel Trefler∗ Rotman School of Management, University of Toronto; CIFAR; NBER August 11, 2012 ABSTRACT: This paper re-explores the relation between a country’s level of wealth and the mix of products it exports. We argue that both are simultaneously determined by countries’ capabilities i.e. by countries’ productivity and quality levels for each good. Our theoretical setup has two features. (1) Some goods have fewer high-quality producers/countries than others i.e. there is Ricardian comparative advantage. (2) Imperfect competition allows high- and low-quality producers to coexist, which we refer to as ‘product ranges.’ These two features generate a very particular non-monotonic, general equilibrium relationship between a country’s export mix and its GDP per capita. We show that this non-monotonicity permeates the international data on trade and GDP per capita. Key words: quality, economic development, international trade JEL classification: F12, O10 ∗. We thank Q. Zhang, Michelle Liu and especially Leilei Shen for excellent research assistance. We have benefited from seminar presentations at the Canadian Institute for Advanced Research (cifar), the LSE, Princeton, Stanford, Toronto, and the World Bank and are grateful of comments from Daron Acemoglu, Philippe Aghion, Bernardo Blum, Avner Greif, Elhanan Helpman, David Hummels, Peter Morrow and Bob Staiger. Trefler thanks cifar and the Social Sciences and Humanities Research Council of Canada (sshrc) for financial support. E-mails: [email protected] and [email protected] 1. Introduction A country’s capability — meaning the set of goods the country is able to produce and its quality and productivity in producing them — drives its per capita income and the sectoral mix of its exports. Aspects of the relationship between quality, income, and the sectoral mix of exports have been analyzed by a number of researchers. Hummels and Klenow (2005) estimate the impact of a country’s per capita income and size on export quality. Hausmann, Hwang and Rodrik (2007) explore how the process of ‘cost discovery’ affects the sectoral mix of exports, which in turn affects per capita incomes. Flam and Helpman (1987) and Fajgelbaum, Grossman and Helpman (2011) examine the co-determination of quality, income, and the sectoral mix of exports in models where demand-side consumer heterogeneity plays a central role. In contrast, we use a supply-side Ricardian model to show how the general equilibrium logic of comparative advantage provides important theoretical and empirical insights into how quality capabilities simultaneously affect per capita incomes and the sectoral mix of exports (as well as prices, markups and profits at the firm level). To bring out these insights as clearly as possible we focus theoretically and empirically on characterizing the range of countries exporting a specific good, as in Schott (2004), and in characterizing how these countries’ market shares vary with their incomes. A standard intuition for the relationship between market shares and incomes runs through quality. Hummels and Klenow (2005) show that rich countries must have high-quality exports because, at the aggregate level, rich countries have high prices and high world market shares. A related inference appears in Khandelwal (2010), Hallak and Schott (2011), and Baldwin and Harrigan (2011). We show both theoretically and empirically that this aggregate insight does not carry over in general equilibrium to the sectoral level because of Ricardian comparative advantage. For example, even though the United States is a high-quality producer of stainless steel, it has a small world market share: It cannot compete with markedly inferior Chinese stainless steel because U.S. wages have been bid up by high demand for military aircraft, virtualization software, and other hard-to-make goods and services that only a handful of rich countries are capable of producing. Our model has two key elements. (a) We make the Ricardian assumption that products can be ordered by the scarcity of quality capabilities. Specifically, if a country has high quality in good g then it has high quality in all goods ranked below g. This means that low-g goods are ones for which most countries have high quality and are in this sense ‘easy’ to make. In contrast, high-g goods are ones for which few countries have high quality: they are ‘hard’ to make. This assumption captures the notion of relative (in a Ricardian sense) scarcity of quality capabilities. (b) We assume that goods are differentiated only by quality (pure vertical differentiation) and are supplied in markets characterized by Nash equilibrium in quantities (Cournot competition). We use this assumption to ensure that differing levels of quality will co-exist in equilibrium. Elements (a) and (b) generate a correlation between a country’s income and its export mix. A country with high quality in just a few goods will only be able to survive in a few markets and these will be the low-g or ‘easy’ markets. As a result, derived demand for the country’s labour will be low and wages will be low. Thus, low-wage countries will export low-g goods. A country with high quality in many goods will have a high derived demand for its labour and have high wages. High wages will make the country a high-cost producer of low-g goods. Hence, a high-wage country will only survive in high-g markets. This Ricardian sorting generates an inverted-U relationship between income and market shares at the sectoral level. To understand why, consider a country whose capabilities improve at the sectoral level: that is, the country improves its quality in a g-ranked good until it reaches the world quality frontier and then it improves its quality in the next, higher-ranked, good. During this quality-improvement process, demand for the country’s labour rises, as does its wages. As quality rises in good g, the country’s world market share of the good rises initially because quality must rise faster than wage costs. This ‘direct’ or ‘quality’ effect underpins the Hummels and Klenow aggregate correlation. However, as quality then rises in a higher-ranked good, wages continue to rise, thus killing off the country’s competitiveness in good g: even though the country is a high-quality producer of good g, its world market share must decline as capabilities rise in higher-ranked, tougher-to-make sectors. This familiar inter-sectoral, general equilibrium feedback through the labour market is what we call the Ricardian or ‘wage’ effect. It is the reason for the downward-sloping section of the inverted-U relationship between income and market share. The model generates a large number of other theoretical predictions which we describe below, but our empirics are concentrated on this inverted-U relationship. Turning to our empirical work, we investigate the sectoral-level, inverted-U relationship between income and market shares using data for 94 countries in 2005. Data are from COMTRADE (4-digit SITC and 6-digit HS) and, to a lesser extent, the U.S. imports file (10-digit HS). The theory states that the inverted-U relationship is driven by labour-market spillovers across sectors. We thus confine our attention to country-good pairs for which the good is important in the country’s export basket 2 and, by implication, in the country’s labour market. For each good separately, we build a ‘product range,’ i.e., a range of incomes defined by the income levels of the poorest and richest exporters of the good. Product ranges are related to Khandelwal’s (2010) quality ladders: the latter describes a range of qualities while the former describes a range of incomes. We will not be estimating quality and hence will have nothing to say about quality ladders.1 Schott’s (2004) work on ’overlap’ leads us to expect that product ranges will be large, and this is indeed what we find. (The finding is not driven by China.) We then non-parametrically estimate the relationship between income and world market shares and show that it is exactly as predicted by the theory. (1) For those products produced only by the richest countries, the relationship is positive: this is the direct or quality effect. (2) For those products produced only by the poorest countries, the relationship is negative: this is the Ricardian or wage effect. (3) For the remaining ‘middle’ products, the relationship is inverted-U, as first the quality effect and then the wage effect kicks in. Re-stated, Ricardian comparative advantage based on relative scarcity of quality capabilities leads to general equilibrium wage effects that are central for understanding the cross-country, cross-sector relationship between quality, per capita income, and the sectoral mix of exports. Our paper has four key elements: (1) multiple sectors that are ranked based on Ricardian scarcity of quality capabilities, (2) an imperfectly competitive market structure that supports the co-existence of differing levels of quality, (3) endogenous income so that there can be general equilibrium spillovers across sectors via the labour market (wages), and (4) empirical work relating income to market shares at the sectoral level. With this in mind, we relate our paper to the existing literature. Our results are driven entirely by supply-side considerations. Demand considerations play no role in our work. Allowing for demand-side heterogeneity and demand for quality that rises with income has yielded important insights for comparative advantage and per capita incomes e.g., Flam and Helpman (1987), Hallak (2006, 2010), Choi, Hummels and Xiang (2009) and Fajgelbaum et al. (2011). The endogeneity of income allows us to bring issues of economic development to the forefront of our research.The relationship between per capita income and the mix of exports has been the subject of investigation at least since the discussion of ladders of development by Chenery (1960) and more recently by Leamer (1984, 1987), Michaely (1984), and Schott (2003). In these papers, as in ours, sectors are asymmetric and ordered. For example, Schott orders sectors by labour intensity. 1. The fact that we are not estimating quality means that our agenda is very different from that of Khandelwal (2010) and Hallak and Schott (2011). 3 However, these papers do not consider quality, which is how we order sectors. Lall, Weiss and Zhang (2006), Hausmann et al. (2007) and Schott (2008) provide policy-oriented discussions of the thesis that ‘what your export matters.’ We do not discuss policy in this paper. Implicitly, however, our work shifts the policy prescription away from getting the right mix of exports and towards raising the quality of what is exported. This paper builds on a series of papers by Sutton. In Sutton (1991, 1998, 2007a,b), firms optimally invest in building quality capabilities and, once these capabilities are developed, firms engage in Cournot competition. This leads to a world in which the relative scarcity of capabilities is an endogenous equilibrium outcome. We can motivate this idea of relatively scarce capabilities by reference to a key idea in the modern ‘market structure’ literature: if firms must incur fixed and sunk outlays to develop their capabilities, then the number of firms that find it profitable to develop these capabilities will be limited: the greater the elasticity of quality (or productivity) responses to R&D or other fixed outlays, the greater the degree to which firms ‘escalate’ their R&D spending in competing with rivals, and the fewer the number of producers that survive in the market. As a result, capabilities are scarce and scarcer in some markets than in others: relative scarcity emerges endogenously. This argument holds for a broad class of models, of which the Cournot model with quality is the simplest and most tractable example; it does not hold for CES models with atomistic firms. For a concise review, see Sutton (2007a). For a general equilibrium analysis of the mechanism of entry and R&D competition leading to this see Sutton (2007b). In this paper, we simply take as a given that some capabilities are relatively scarce.2 3 Sutton (2007b) provides an international trade model with two countries that produce final goods, and a third country that produces raw materials. He shows that when there are raw materials that are internationally traded, quality and productivity are not isomorphic and, in particular, that there is a minimum level of quality (independent of productivity or wages) that must be attained if a firm or country is to enter world export markets. Thus, there are limits on how much low wages can offset low quality. Hallak and Sivadasan (2009) also break the quality-productivity isomorphism by 2. Cournot competition, unlike models of CES monopolistic competition with atomistic firms, explicitly allows for ‘large’ firms and this is increasingly recognized as a crucial feature of international trade. The idea that firms are not a point in a continuum appears in di Giovanni and Levchenko (2011) and Eaton, Kortum and Sotelo (2012). As the former note, a single New Zealand firm (Fonterra) is responsible for one-third of global dairy exports and Samsung is responsible for 23% of Korean exports. See Melitz and Trefler (2012) for other examples. Cournot competition in international trade models appears in Neary (2003) and Neary and Tharakan (2011). 3. Sutton’s (1991; 1998; 2007a) work is related to the literature on the endogenous choice of quality e.g., Verhoogen (2008), Kugler and Verhoogen (2012), Khandelwal (2010), Baldwin and Harrigan (2011), and Johnson (2012). 4 postulating the existence of a minimum quality threshold needed for exporting. In our paper, by contrast, quality and productivity are isomorphic. The paper is structured as follows. Section 2 sets up the model. In section 3, we consider a limited form of international differences in quality capabilities in order to develop a benchmark model with assortive matching of countries to goods. Section 4 develops the more general model and the key results. Section 5 draws out the empirical implications of the model that are then examined in sections 6–9. 2. Set-Up 2.1. Consumer Choice All countries are of the same size and composed of a unit mass of workers. All individuals have identical Cobb-Douglas utility functions defined over G goods, indexed by g, U= ∏(u g x g )δ g (1) g where ∑ δg = 1, and u g and x g denote the quality and quantity of good g consumed. It follows g from the form of the utility function that each consumer spends fraction δg of income on good g. We assume that all profits accrue to a separate group of individuals, who also have a utility function of the form (1). From this it follows that we can treat all firms in the global market for g as facing a unit-elastic market demand schedule, i.e. the total global expenditure on good g is a constant, which we denote as Sg , independently of equilibrium prices. We note that the Sg are proportional to the δg . We will assume throughout that all the δg are equal (δg = δ), and so all the Sg are equal (Sg = S). 2.2. Equilibrium in the Product Markets We characterize product market competition using the standard ‘Cournot model with quality’ introduced in Sutton (1991). In this model, firms are characterized by a level of capability, consisting of a quality level and a productivity parameter denoting the number of worker hours per unit of output 5 produced,4 together with a (‘local’) wage rate specific to the country in which the firm is located. (Firms are wage takers.) At equilibrium, some subset of firms are active in the production of the good. For each active firm, indexed by i, its output level is related to the inverse of its productivity ci , its quality ui , and its (local) wage rate wi . Solving for a Nash equilibrium in quantities (Sutton, 1998, Appendix 14.1), we obtain the firm’s quality-adjusted equilibrium price, wj cj 1 pi = ∑ ui Ng − 1 j u j (2) and its quality-adjusted output level, Ng − 1 wi ci /ui xi ui = S 1 − ( Ng − 1) wj cj wj cj ∑j ∑j uj (3) uj where Ng (≥ 2) denotes the total number of firms that are active in the global market for good g, S is total expenditure on good g and the sum ∑ j is taken over all active firms. One can see from equation (2) that pi /ui is the same for all active firms. The condition for firm i to be active, i.e. to have strictly positive output at equilibrium, is that wj cj wj cj Ng wi c i 1 < = ( ) ∑ ui Ng − 1 j u j Ng − 1 u j where ( wj cj uj ) denotes the mean of wj cj uj (4) over all active producers. We will refer to wi ci /ui as firm i’s ‘effective cost level.’ It is useful to plug equation (2) into equation (3) to obtain an alternative expression for output: xi = 1 pi 1− wi c i pi S. (30 ) Thus, a firm is active in equilibrium if its price pi exceeds its marginal cost wi ci . Note that the right-hand side of equations (2) and (3) depend on ui and ci only through the ratio ui /ci , which we refer to as the ‘capability’ of firm i. It follows that some key relationships between capabilities and wages developed below will depend only on ui /ci and not on the absolute levels 4. Thus all costs are labour costs, and fixed costs are sunk, and so do not enter the present (short-run) analysis. Materials cost, though of crucial importance in general, are here ignored in order to keep the analysis as clear as possible. This issue is examined in depth in Sutton (2007b) who shows that the key point is this: in the absence of material cost, low-wage countries can become viable in world markets even at low quality once their wage costs are sufficiently low: only the ratio of unit costs (wages times labour input) to quality matters to viability, and shortcomings in quality can be offset by a low value of the wage. But once material inputs as well as labour are required, a fall in the wage can only reduce unit costs to the world-market value of the material input. This places a floor on price, and so establishes a corresponding minimum quality level, independent of local wages, that is required for viability. Deficiencies in productivity can always be compensated for by low wages, but deficiencies in quality cannot. This is an important reason for emphasizing the role of quality in our present discussion. 6 Figure 1: Types of Countries and Groups of Goods 1, . . . ,N1 | {z } 1, . . . ,Nk | {z } 1, . . . ,NK | {z } 1 z }| { 1, . . . ,m k z }| { 1, . . . ,m K z }| { 1, . . . ,m Types of Countries Groups of Goods of ui and ci . Since our empirical focus is on quality ui , without loss of generality we set ci = 1 for the remainder of the paper and periodically remind the reader that our comments about quality are also germane to productivity. 3. Theory I: The Perfect Sorting Baseline Our aim is to establish a baseline equilibrium in which Ricardian comparative advantage leads to assortative matching between countries and goods. We proceed in steps. 3.1. The Scarcity of Quality Capabilities Our key assumption is that countries differ only in terms of their quality capabilities for different goods. Assume for the moment that there are k = 1, ..., K goods. A type-k country is a country whose firms can produce goods 1 to k at a ‘standard’ level of quality u, but cannot produce goods k + 1 to K. The interpretation is that goods with higher indexes require capabilities that are scarcer. They are ‘harder’ to make. Countries differ only in this capability dimension. There are thus K types of countries. We next generalize this slightly by assuming that there are K groups of goods. Within each group there are m identical goods and the above applies to each good: Specifically, a type-k country has firms that can produce each of the goods in groups 1 to k (at quality u ) but cannot produce any of the goods in groups k + 1 to K. See figure 1. There are mK goods. Following up on our introductory comments on the scarcity of capabilities, we assume that each type-k country is endowed with a finite number of firms. For simplicity alone, we assume that each country has exactly one firm per group-k good. It thus has m firms. (Whether a firm is active or not is endogenous.) Let Nk be the number of type-k countries. In an equilibrium with assortative matching, group-k goods are produced by and only by firms from type-k countries. Given the symmetry in the model, 7 it is easy to show that all firms will be active so that in an equilibrium with assortative matching Nk is also the number of firms worldwide producing each group-k good. Thus, for there to be fewer equilibrium producers of hard-to-make goods, Nk must be decreasing in k. In fact, something slightly stronger is needed. Recall that a firm from a type-k country can produce a good either from group k or from any group ranked below k. To ensure that in equilibrium the group-k good is chosen, we need Nk to be decreasing in k sufficiently rapidly: Assumption 1 (Ricardian Relative Scarcity) Nk ≥ Nk+1 + 4. Finally, in this section u does not vary across firms; such variation is introduced in section 4 and is essential to our argument.5 3.2. Equilibrium in a Type-k Country All type-k countries are identical. Let wk be the common wage. Let g index the mK goods. Our assumptions above imply that in equation (2), w j = wk , u j = u, c j = 1 and Ng = Nk . Hence, for any good g in group k that is produced in a type-k country, the equilibrium price in equation (2) becomes p gk = Nk w Nk − 1 k 0 (2 ) and the equilibrium quantity in equation (30 ) becomes: x gk 1 = p gk w 1− k p gk S= Nk − 1 S . Nk2 wk 00 (3 ) Since one unit of output requires one unit of labour, the total demand for labour in a type-k country is the output per good (x gk ) times the number of goods (m): LkD = mx gk . We assume that there are N workers in each country, each of whom supplies one unit of labour so that labour supply is LSk = N. Hence LkD = LSk ⇐⇒ x gk = N . m (5) 00 From equations (3 ) and (5), the labour-market clearing wage is: wk = Nk − 1 mS . Nk2 N (6) 5. Allowing u to vary with k does not alter anything because quality comparisons are always done within groups of goods, not across groups of goods. 8 0 Equilibrium in product markets is already built into equation (2 ). Hence, plugging equation (6) 0 into (2 ) and noting that the result is independent of g, equilibrium product prices are given by pk = 1 mS . Nk N (7) Income per worker yk is wages plus profits (i.e., revenues) summed over the m goods: yk = mpk x gk . N (8) For each good g that a type-k country produces, its net exports are: Net Exports = (1 − δm) x gk > 0 . (9) This follows from the fact that the value of good-g net exports equals the value of production (pk x gk ) minus the value of consumption (δyk N = δmpk x gk ).6 An equilibrium is a set of wages {wk }k and product prices { pk }k such that consumers maximize utility, producers maximize profits, labour markets clear nationally, product markets clear internationally, and trade is balanced. Under our assumption 1 restriction on the Nk s there will be perfect sorting in equilibrium. Lemma 1 Under assumption 1, all equilibria display perfect sorting i.e., group-k goods are produced by type-k countries and only by type-k countries. Wages wk , GDP per worker yk , product prices pk , markups pk /wk , and net exports are unique and given by equations (6)–(9). In addition, wk , yk , pk , and pk /wk are all increasing in k i.e., countries with scarce capabilities have higher wages, higher GDP per worker, charge higher prices, and have higher markups. All proofs appear in Appendix A. In this section we set up a perfect sorting equilibrium. Note that the Cournot model played no essential role in this. Its role is central however in the next section, where we introduce quality differences across different firms (countries) operating in an industry. 4. Theory II: Quality Differences In this section, we relax the assumption characterizing the baseline model, that all producers of a good have the same quality (and productivity.) The central aim is to characterize the band of 6. Note that δ is the share of expenditures spent on each good i.e., δ = 1/(mK ). This implies 1 − δm > 0. 9 countries, in terms either of their real wage rate or GDP per worker, that will be active in a given industry at equilibrium. With this in mind, we examine one of the countries initially of type k − 1, whose capability in the production of all goods of group k advances, in the sense that its quality, denoted vk , rises from zero to the standard quality u. We will refer to this country as the ‘developing’ country. As vk rises, the developing country’s mix of output will gradually shift from the production of goods of group k − 1 to goods of group k. This change will, in general, affect the equilibrium wage rate of all countries of adjacent types, and the total income (and expenditure) in each country and market. We start with some notation. For a typical type-k country, let wk , be its wage, let pk be its price, let x k be its quantity, and let u be its quality. Let wk−1 , pk−1 , x k−1 , and u be the corresponding variables for a typical type-(k − 1) country. For the developing country, which may be producing both group-(k − 1) and group-k goods, let w be its wage, let pk−1 and pk be its prices, let xk−1 and xk be its quantities, let u be its quality in group-(k − 1) goods, and let vk be its quality in group-k goods. Income or GDP per worker for our three types of countries is: yk−1 = mpk−1 x k−1 /N, yk = mpk x k /N and y = m { pk−1 xk−1 + pk xk } /N (10) i.e., revenue per good times the number of goods (m) divided by the workforce (N). Consider the situation in which vk has risen to the point where the home country is producing both groups of goods. All producers of a good charge the same quality-adjusted price so that: p p k −1 = k −1 u u (11) p pk = k . vk u (12) For each group-(k − 1) good there are Nk−1 − 1 typical producers with wage wk−1 and the one developing-country producer with wage w. Hence multiplying equation (2) through by the common u, p k −1 = 1 Nk−1 − 1 [( Nk−1 − 1)wk−1 + w] . (13) The developing country’s presence in both the k − 1 and k industries means that it now has 2m firms, the original m firms producing each of the group-(k − 1) goods and the m new firms 10 producing each of the group-k goods.7 The total number of firms producing each type-k good is therefore Nk + 1: Nk firms have wage wk and quality u while the one developing-country firm has wage w and quality vk . Hence, pk 1 = u Nk w w Nk k + u vk . (14) To determine prices and wages, we need simply look at the labour-market clearing conditions. Recalling that labour supply is given by N and that labour demand is given by total output, labour-market clearing for a typical type-(k − 1) country is N = mx k−1 . Plugging the equation (30 ) expression for x k−1 into N = mx k−1 yields: N=m 1 p k −1 w 1 − k −1 p k −1 S, (15) where we have used the fact that every firm in a typical type-(k − 1) country charges price pk−1 and has wage wk−1 . For a typical type-k country, labour-market clearing is likewise given by: 1 N=m pk w 1− k pk S. (16) For the developing country, labour-market clearing is N = mx k−1 + mx k or N=m 1 p k −1 1− w p k −1 S+m 1 pk 1− w pk S. (17) Equations (11)–(17) are seven equations in the four prices and three wages. It is very easy to solve explicitly for these seven variables in terms of a numeraire and this is done in Appendix A.2, where the numeraire is expenditures per good S. This establishes that for vk in the range where the developing country produces both groups of goods, there is a unique equilibrium and in this equilibrium there are closed-form expressions for all the equilibrium prices. Since our focus is comparative statics with vk , we write all equilibrium outcomes as functions of vk e.g., w(vk ), pk (vk ), xk (vk ) and S(vk ) We now explore the effect of raising the exogenous variable vk from the critical value at which xk (vk ) becomes positive, past the critical value at which xk−1 (vk ) becomes zero, and through to the value vk = u at which the developing country becomes identical to a type-k country. 7. An alternative interpretation is as follows: the group-k good is produced by the same firm that produces the corresponding group-(k − 1) good. Each firm now has two independent businesses, making k and k − 1. Given Cobb-Douglas preferences, constant marginal costs, and the fact that the firm takes S as given, it follows that the firm’s profit function is additively separable in its activities in the k − 1 and k markets so that the firm sets the Cournot output in each of these markets. 11 Lemma 2 Suppose assumption 1 holds. Then there are constants vkL and vkH with 0 < vkL < vkH < u such that: vkL < vk < vkH ⇐⇒ xk−1 (vk ) > 0 and xk (vk ) > 0; vkH < vk ≤ u ⇐⇒ xk−1 (vk ) = 0 and xk (vk ) > 0. We refer to the situation where the developing country is producing both xk−1 (vk ) and xk (vk ) as Phase I. We refer to the situation where the developing country is producing only xk (vk ) as Phase II. A third phase, where the developing country has gained the capability of producing goods in group k + 1, will be described later. These three phases appear at the top of figure 2. Note that when vk < vkL , the developing country’s quality is so low that it specializes in group-(k − 1) goods i.e., we are in the perfect sorting case of the previous section. Phase I: We begin by characterizing the equilibrium properties of Phase I. These are illustrated in Figure 2 and stated in the following proposition. Proposition 1 (Phase I) Let assumption 1 hold and consider vk ∈ (vkL , vkH ) so that xk−1 (vk ) > 0 and xk (vk ) > 0. Then as vk rises from vkL to vkH : 1. xk−1 (vk ) falls to 0 and xk (vk ) rises from 0 i.e., the developing country reduces output of group-(k − 1) goods and increases output of group-k goods; 2. pk−1 (vk )/w(vk ) falls and pk (vk )/w(vk ) rises i.e., markups charged by developing-country firms fall for group-(k − 1) goods and rise for group-k goods; 3. w(vk )/S(vk ) rises i.e., the wage in the developing country rises relative to the numeraire; 4. y(vk )/S(vk ) rises, i.e. GDP per worker in the developing country rises relative to the numeraire. The core Phase I insight is that as the developing country’s quality capabilities improve in group-k goods, the developing country experiences rising demand for its output and labour and hence rising wages. This creates two effects. (1) Improved quality improves the country’s competitiveness in group-k goods. This is the direct or ‘quality’ effect. (2) Higher wages reduce the country’s competitiveness in group-(k − 1) goods. This is the general equilibrium Ricardian or ‘wage’ effect. While not used for our empirics, we note that there are cross-country general equilibrium impacts. For example, as China improves its quality capabilities in autos we would expect that 12 Quality of Goods k, k + 1 Figure 2: Advancing Quality Phase Phase Phase I II III • • • vk • v k +1 • • • Wages or Markup w or y/N • • • • • • • of Good k GDP per worker • • pk /w Output of Goods k − 1, k + 1 • • • x k −1 • x k +1 Good k Output of • • • • • xk • vkL • vkH u vkL+1 vkH+1 u Quality vk v k +1 Notes: The top panel shows the developing country’s quality in group-k goods advancing from vk = 0 to vk = u, and then its quality in group-(k + 1) goods advancing from vk+1 = 0 to vk+1 = u. The critical value of vk at which production of group-k goods becomes viable is labelled vkL , and is marked by the left-most dashed vertical line. The critical value of vk at which production of group-(k − 1) goods becomes unviable is labelled vkH , and is marked by the dashed vertical line separating Phases I and II. The second and third panels show how equilibrium income (wages and GDP per worker) and the markup rise as qualities rise. The bottom two panels show how the output of goods in groups k − 1, k, and k + 1 change. 13 clothing manufacturers in Bangladesh would be better off and auto manufacturers in the United States would be worse off. The next proposition makes this point. Proposition 2 (Phase I) Let assumption 1 hold. Then as vk rises from vkL to vkH : (1) wk−1 (vk )/S(vk ) rises and wk (vk )/S(vk ) falls i.e., relative to the numeraire, the wage rises in type-(k − 1) countries and falls in type-k countries; and (2) w(vk )/wk−1 (vk ) rises i.e., the wage rises faster in the developing country than in type-(k − 1) countries. Phase II: When quality capabilities rise above vkH , our developing country’s wage becomes so high that its firms are no longer viable (profitable) in the markets for group-(k − 1) goods. Equations (12)–(17) continue to hold with only two modifications. The labour-market equilibrium condition for the developing country (equation 17) now becomes: N = mxk = m 1 pk 1− w pk S. (17–II) Also, the price of group-(k − 1) goods (equation 13) must be modified because there are now only Nk−1 − 1 producers, all with wage wk−1 and quality u. Hence, from equation (2) with Ng = Nk−1 − 1, p k −1 = Nk−1 − 1 w . Nk−1 − 2 k−1 (13–II) Equations (12), (13–II), (14), (15), (16) and (17–II) are six equations in the three prices and three wages. It is very easy to solve explicitly for these six variables in terms of the numeraire S and this is done in Appendix A.3. This establishes existence and uniqueness and provides a complete characterization of equilibrium in Phase II. The impact of vk rising to u appears in figure 2 as the early part of Phase II (where vk+1 is still 0), and is stated in the next proposition. Proposition 3 (Early Phase II) Let assumption 1 hold and consider vk ∈ (vkH , u) so that xk−1 (vk ) = 0 and xk (vk ) > 0. Then as vk rises from vkH to u: 1. pk (vk )/w(vk ) rises i.e., the markup charged by developing-country firms rises; 2. w(vk )/S(vk ) rises i.e., the wage in the developing country rises relative to the numeraire; 3. y(vk )/S(vk ) rises i.e., GDP per worker in the developing country rises relative to the numeraire. 14 Phase III: We now allow the developing country’s quality capabilities in group-(k + 1) goods, which we denote by vk+1 , to rise from 0 to u (figure 2, top panel). All of our lemma 2 and proposition 1 results relating changes in vk to changes in variables subscripted by k − 1 and k now apply when relating changes in vk+1 to changes in variables subscripted by k and k + 1, respectively. In particular, there is a critical value vkL+1 at which the developing country begins producing goods in group k + 1 and a critical value vkH+1 at which the developing country stops producing goods in group k. During this process markups rise for group-(k + 1) goods (the direct or ‘quality’ effect). Further, wages and GDP per worker rise relative to the numeraire, and this reduces the markups for group-k goods (the Ricardian or ‘wage’ effect). See figure 2. 5. Towards Empirics: Implications for Exports To examine these predictions empirically, we will use international data and therefore need expressions for exports. We start by noting that, in equilibrium, all producers charge the same quality-adjusted price so that consumers do not care which producer they buy from. We therefore assume that in equilibrium there is no cross-hauling of goods across international borders i.e., a good is either imported or exported, but not both. Let X k−1 (vk ) and X k (vk ) be the value of exports for typical countries of type k − 1 and k, respectively. Note that a lower-case x is the quantity of output and an upper-case X is the value of exports. Also note that these are values (price times quantity) since that is what we observe in the data. The developing country’s value of exports for a good in group k − 1 and a good in group k is denoted by Xk−1 (vk ) and Xk (vk ), respectively.8 The next lemma states that exports behave like production. Lemma 3 Let assumption 1 hold. 1. As vk rises in Phase I, Xk−1 (vk )/S(vk ) and Xk−1 (vk )/X k−1 (vk ) fall to 0. 2. As vk rises in Phases I and II, Xk (vk )/S(vk ) and Xk (vk )/X k (vk ) rise from 0. In our empirics we will examine the value of exports of group-(k − 1) and group-k goods as a share of world exports. For the developing country these shares are given by: θkX−1 (vk ) ≡ X k −1 ( v k ) Xk−1 (vk ) + ( Nk−1 − 1) X k−1 (vk ) and θkX (vk ) ≡ Xk ( v k ) . Xk (vk ) + Nk X k (vk ) (18) 8. Typical type-(k − 1) countries produce and export a single good so that exports are both gross and net. For some values of vk , the developing country produces two goods and one of these is produced in such small amounts that the good is imported. Thus, for the developing country, exports are gross rather than net. 15 θkX+1 is the same as θkX but with the k index incremented by 1. The evolution of these world export shares follow immediately from lemma 3. In Phase I, the developing country is shifting out of group-(k − 1) production and into group-k production. This benefits type-(k − 1) countries and hurts type-k countries so that θkX−1 falls and θkX rises. In Phase II, the developing country is specialized in group-k production and getting better at it, which hurts type-k countries. Thus, θkX rises. We now make the transition from theory to empirics. We do not observe capabilities, but changes in capabilities induce observable changes in GDP per worker and the value of exports. These are illustrated in figure 3, which plots export shares against GDP per worker for the developing country. In Phase I, y(vk ) increases, θkX−1 (vk ) falls to 0, and θkX (vk ) rises from 0. Note that θkX (vk ) does not start rising as soon as Phase I is entered because at this point production of group-k goods is so small that these goods are still imported. The point at which exporting of group-k goods starts is indicated in figure 3 by ymin ,k . This ymin ,k is very important for our empirics. In the first part of Phase II where vk is rising to u, both y(vk ) and θkX (vk ) rise. In the second part of Phase II, where capabilities in k + 1 are rising but have not yet reached a level at which the developing country can enter group-(k + 1) markets, nothing happens. See figure 2. It follows that the system is ‘stuck’ at the point where y = y(u). In Phase III, the developing country enters group-(k + 1) markets and grabs world market share i.e., θkX+1 rises. This drives up wages and makes the developing country less competitive in group-k markets. As a result θkX (vk ) falls. As the process continues, the developing country reaches the point where it consumes the small amount of xk that it is still producing i.e., θkX (vk ) = 0. By definition, Phase III ends when xk = 0, so θkX (vk ) goes to 0 just before the end of the Phase. This is indicated on figure 3 by the point ymax ,k , which is also important for our empirics. It is straightforward to calculate closed-form solutions for ymin ,k and ymax ,k .9 Two key empirical points emerge from figure 3. First, poorer, low-quality countries can produce the same good as richer, higher-quality countries. Low-quality countries compete because in equi9. There are two minor points in the figure that are not used in what follows. (1) At the start of Phase I, all countries are identical so that each of the type-(k − 1) countries has a world market share of θkX−1 = 1/Nk−1 . At the end of Phase II, all countries are again identical so that each type-k country has a world market share of θkX = 1/( Nk + 1). At the end of Phase III it is θkX+1 = 1/( Nk+1 + 1). (2) Production always go to 0 at the end of a phase. Since gross exports go to 0 before production does (i.e., at the point where all domestic production is consumed domestically), gross exports always go to 0 before the end of a phase. Likewise, at the start of a phase, production starts and is consumed domestically so that exporting starts just after the start of a phase. 16 Figure 3: Empirics: Quality and the Income-Export Nexus Phase I Phase II Phase III Export Shares (Direct or ‘Quality’ Effect) 0 (Ricardian or ‘Wage’ Effect) θkX+1 θkX θkX−1 ymin ,k y(vkH ) y(u) ymax ,k GDP per Worker Product Range (k) Notes: The figure illustrates two of the central empirical predictions of the model. First, a single good can be produced both by low-wage, low-quality countries and by high-wage, high-quality countries. ymin ,k and ymax ,k are the GDP per worker of the poorest and richest countries producing a group-k good. Second, world market shares will be an inverted-U shaped function of GDP per worker. librium they have low wages. This implies that there are ‘product ranges,’ that is, ranges of GDP per worker compatible with viability in the market. In figure 3, the product range for group-k goods is (ymin ,k , ymax ,k ). Second, whereas in a single-sector model, world market shares increase with quality, in a multi-sector world there are Ricardian forces which lead to inverted-U shaped world market shares. As quality rises, wages rise and this leads to a loss of competitiveness in low-k goods. As a result, export shares eventually ‘turn down.’ These are the two main empirical facts that we will examine. A third fact, that prices rise with quality and hence with income, is an important implication of the model; however, we explore it empirically only in Appendix C because it has 17 been examined elsewhere e.g., in Schott (2004).10 11 We can restate this in a way that makes one of the key points of our thesis crystal clear. A poor country can advance out of low-ranked goods and still remain poor: this happens when the country enters as a low-quality producer into goods with wide product ranges. Since, as we shall show empirically, most goods have wide product ranges, we might expect this type of no-growth shift in product mix to be common. By the same token, a country may move from being poor to being rich without changing its product mix: this happens when it improves the quality of the wide-product-range goods that it already exports. In short, GDP per worker depends not just on what a country produces (as in Hausmann et al. 2007), but on the quality of what is produced. 6. Data Trade data are from COMTRADE for 2005. We use the 4-digit SITC Revision 2 classification (henceforth SITC4).12 To verify that all of our cross-sectional results hold for more detailed commodity breakdowns we also use the 2005 COMTRADE data at the 6-digit HS level (1996 revision, henceforth HS6) and the 2005 U.S. import data at the 10-digit HS level (henceforth HS10). We exclude countries whose population was less than two million in 2005 and/or whose territorial integrity changed substantially between 1980 and 2005 e.g. the USSR. (The exception is Germany, which we include.) This leaves us with the 94 countries listed in Appendix B. GDP per capita and population data are from the United Nations. We do not use a PPP adjustment because we are interested in nominal price competition in world product markets. 10. Additionally, we note that our model, which has endogenous wages, delivers a clear statement about how prices rise both because of quality improvements and because of rising marginal costs (wages). Many models of international trade and quality treat wages as exogenous and so cannot give such a clear statement. 11. Some readers will have noticed that the output-quality or output-income relationship in figure 3 looks like the Heckscher-Ohlin output-capital or cones-of-diversification relationship (see Leamer, 1984; Schott, 2003). One might wonder, then, why was our theory needed? For one, improvements in quality are very different empirically from capital deepening. More importantly, this prediction is just one of several predictions that arise in our model, and all these predictions flow from the fact that there is imperfect competition. Imperfect competition is needed to ensure that different qualities co-exist, that prices are a non-constant markup over marginal cost, that prices are correlated with quality, and that export shares are correlated with income in ways that reflect quality. In short, our output-quality relationship is just one of several predictions. The output prediction in isolation can be modelled more simply; however, we are interested in a bundle of predictions that require us to append an imperfectly competitive market structure onto a trade model. Thus, a cones-of-diversification Heckscher-Ohlin model delivers at best only a small part of what is needed and a more natural trade model in our setting is the Ricardian model, which emphasizes the role of technological capability for delivering quality. 12. This allows us to go back to 1980 for a large number of countries and check that our results hold for these earlier years. 18 7. Product Ranges A key prediction of our theory is that in general equilibrium countries with different quality capabilities may nevertheless export the same good. See the product range in figure 3. While we do not observe quality, an observable implication is that at least some goods will be produced both by rich and poor countries. To investigate, for each product g we identify the poorest and richest exporters of the product. Denote these by ymin,g and ymax,g , respectively. To avoid ‘noise’ associated with small reported export values, a problem to which trade data are notoriously prone, for each good we look only at the set of countries for whom the good is a ‘significant’ export, in the sense that the value of its exports in that good constitute at least 1% of the value of exports of the country’s principal export good.13 An important theoretical reason for using this 1% cut-off is that it ensures that the good is sufficiently important to the exporter to generate the general equilibrium wage impacts upon which our theory rests. Product ranges are displayed in figure 4. Each point corresponds to a unique SITC4 good (g) and the figure plots (ymin,g ,ymax,g ). A point therefore shows the range of income levels of countries for which g is a significant export. All the points necessarily lie above the 45◦ line. For reference, along the axes we show the log GDP per capita of Nepal, China, Poland and the United States. The striking feature of figure 4 is the preponderance of points in the top left corner, i.e. the preponderance of products for which the income range is very wide. To get a clearer sense of magnitudes, consider the points lying in the top left corner where ln ymin,g < 8.25 and ln ymax,g > 10. If a product is in this region then its richest significant exporter is at least 5.8 times richer then its poorest significant exporter (e10−8.25 = 5.8). For the median product in the region the corresponding difference is 79-fold (e4.4 = 79). These are huge differences. And there are a lot of goods in this region: the region contains 81% of all products displayed in the figure and accounts for 70% of world trade. We will shortly show the reader that this observation about wide product ranges is robust, and holds even in the most detailed trade data (HS10). However, we first draw three economic insights from the wideness of product ranges. The first deals with Hausmann et al. (2007). Their 13. More formally, let l index countries, let g index goods, let Xgl be the value of country l’s exports of good g, and let ∑ g Xgl be country l’s total exports. Identify the good that accounts for the largest share of country l’s exports i.e., the good with the largest Xgl /Xl . Call this good g(l ). Then good g is a significant export of country l if Xgl /Xl > 0.01Xg(l ),l /Xl . Next, let K ( g) be the set of countries for which g is a significant export. Then ymin,g is the poorest country in K ( g) and ymax,g is the richest country in K ( g). 19 Figure 4: Product Ranges Richest (GDP per Capita) Producer of Good g: ln(ymax,g) 12 USA 10 POL 8 CHN 6 NPL NPL POL CHN USA 4 4 6 8 10 12 Poorest (GDP per Capita) Producer of Good g: ln(ymin,g) Notes: Each point represents an SITC4 product. The horizontal axis is ln ymin,g , the poorest country for which the product is a significant export. The vertical axis is ln ymax,g , the richest country for which the product is a significant export. 20 exercise uses all goods in a country’s export basket even though products with wide ranges are ‘uninformative’ about a country’s income in the sense that knowing that a wide-range product is a significant contributor to a country’s export basket tells us little about the country’s income. Figure 4 shows that such ‘uninformativeness’ is the norm rather than the exception. Second, our theory emphasizes that for each product, multiple quality levels can coexist in equilibrium. One can therefore interpret the wide ranges as support for the theory provided that one is willing to accept that product ranges are the result of quality differences. As is well known, quality is difficult to identify without detailed data about product characteristics. Since we do not have this information we refer to the ranges as product ranges rather than as quality ranges and take the weaker position that wide product ranges are implied by the theory but do not imply the theory. Third, there are two distinct groups of points that lie far from the top-left corner in figure 4. These are ‘informative’ products. The first group lies to the bottom left (ln ymax,g < 10) and consists of those goods exported only by relatively low- and middle-income countries (the ‘L group’). The second group lies to the top right (ln ymin,g > 8.25) and consists of those goods exported only by relatively high-income countries (the ‘H group’). On our present interpretation, L-group goods are not produced by high-income countries because these countries’ wage costs are too high, whereas H-group goods are not produced by low-income countries because their quality capabilities are too low. The reader will and should be skeptical about the wide product ranges in figure 4. For the remainder of this section we anticipate five possible objections to the figure. 1. It is all aggregation bias: One would expect that the large product ranges displayed in figure 4 would become much narrower with finer product-level data. This is not the case. In figure 5 we repeat the exercise using HS6 data (world trade data from COMTRADE) and using HS10 data (U.S. import data). The distribution of product ranges in figures 4 and 5 are very similar. In particular, product ranges remain large and in both panels just over 70% of total exports are accounted for by the uninformative products in the top left (ln ymin,g < 8.25 and ln ymax,g > 10).14 2. Finer disaggregation is always better: The fact that nothing changes when moving to finer levels 14. In figure 5 there are thousands of points, many of which lie on top of each other. To make the figure clearer, instead of plotting ln ymax,g on the vertical axis we have plotted ln ymax,g + e where e is a uniformly distributed random variable on (−0.05,0.05). This adds a tiny random vertical shift to the data, which helps the reader see where the bulk of points are located. Likewise, we have added a tiny random horizontal shift to ln ymin,g . 21 Figure 5: Product Ranges: Insensitivity to Aggregation Notes: Each panel in this figure is constructed in the same way as figure 4, but with different data. Figure 4 used the SITC4 classification and COMTRADE (world) data. The left-hand panel of the current figure uses the HS6 classification and COMTRADE data. The right-hand panel uses the HS10 classification and U.S. import data. of product disaggregation may seem puzzling, since if the move to a finer level of aggregation involved the breaking up of technologically disparate sub-industries into individual industries, we might expect the range to narrow as we move to this new level of aggregation. An examination of the way in which industries are broken up in the HS6 and HS10 data throws light on why disaggregation beyond SITC4 does not alter the distribution of ranges. In some cases the SITC4 industry is as disaggregated as the HS6 and even the HS10 industries e.g. ‘New tires for motor cars’ is a single category in both SITC4 and HS6. In other cases, the disaggregation is based only on size or value, without any reference to capabilities e.g. ‘New tires for motor cars’ feeds into seven HS10 codes that distinguish between technology-ambiguous differences in the diameter of the tire. In yet other cases the SITC4 code is disaggregated only by introducing a capability-irrelevant ‘parts of’ HS6 or HS10 code. This is pervasive e.g. see the HS6 categories associated with SITC4 7817 ‘Nuclear reactors.’ Finally, in those cases where a technology-based disaggregation of products is introduced it is often unclear whether this disaggregation conveys any information about differences in required capabilities: for example, SITC4 7252 ‘Machinery for making paper pulp, paper, paperboard; Cutting machines’ is disaggregated in HS10 into a number of industries, including ‘Machines for 22 making paper bags etc.’ and ‘Machines for making paper cartons etc.’ Thus, finer disaggregation is typically not more informative about quality capabilities. Were an ideal disaggregation of industries to be constructed on the basis of the quality capabilities required, this would doubtless lead to some narrowing in the relevant ranges. However, the limitations of the published data are quite serious even at the most disaggregated level.15 3. Estimation error: Another possible objection to our wide product ranges is that we have not sig reported standard errors. Let Ng be the number of countries for which g is a significant export.16 sig It is possible that products with wide ranges are products for which Ng is small i.e. for which there are very few observations and hence large standard errors. This is not the case; indeed, the sig opposite is true. The correlation between Ng and the product range ln ymax,g − ln ymin,g is positive sig (0.57) and, for example, products with Ng ≥ 20 (one quarter of all products) all have large product ranges. However, to deal with this objection in the simplest way possible, in figures 4–5 we have sig only displayed those products for which there are at least three significant exporters (Ng ≥ 3). That is, we displayed only 547 of the possible 746 SITC4 goods. These 547 products account for 98.3% of world trade so that we are only excluding very minor products. We conclude from this that wide product ranges are not an artifact of statistical uncertainty. To be safe though, we will sig continue throughout this paper to restrict attention only to products for which Ng ≥ 3. 4. Wide product ranges are an artifact of using a 1% cut-off for ‘significant exporters’: Again, this is not the case. On-line appendix figure A1 shows that the inference we have drawn from figures 4–5 is not sensitive to the choice of cut-offs. It repeats figure 4 for a low percentage cut-off (0.1%), a high percentage cut-off (10%), and cut-offs based on mixtures of percentages and dollar values (x gk > $5 million and x gk > $50 million). In every single case the pattern displayed in figures 4–5 is repeated.17 5. Wide product ranges are driven by China: Omitting China does not alter the impression that 15. For what we are doing, the relevant market is never equatable with an item in a government commodity classification, be it SITC4, HS6 or HS10. Sometimes the relevant market is more detailed than HS10 (as in many electronic parts) and sometimes the relevant market is less detailed than SITC4 (as in many apparel products). Thus, all of our conclusions must be thought of relative to a definition of the market that is determined by the commodity classification, not the actual product producers. sig 16. Ng is the dimension of K ( g) in footnote 13. 17. There is a minor technical point about figure 5 that should be reviewed. Since the United States is far from most countries and since trade costs increase in distance we expect that countries’ exports to the United States will be more concentrated on a few goods than their exports to the world. This is indeed the case. Therefore, for the HS10 panel of figure 5, which is based on U.S. data, we use a 0.1% cut-off instead of a 1% cut-off. This results in far more points in the figure, but does not alter the distribution of points in the figure. See on-line appendix figure A1 for the HS10 figure using a 1% cut-off. 23 product ranges are wide. Indeed, the reader can omit China from these figures simply by deleting all points for which either ln ymin,g = 7.5 or ln ymax,g = 7.5. (China’s log GDP per capita is 7.5.) Having established the robustness of figure 4, we can now restate our conclusion. Our theory implies that there will be product ranges: the empirical surprise is that product ranges are often so large. 8. Market Share Predictions Figure 3 presented our predictions about a country’s share of world exports. Underlying that figure is a comparative static in which a country that previously specialized in producing good k − 1 first sees its quality in good k rise up from a very low level to that of the world standard and then sees its quality in good k + 1 rise up from a very low level to that of the world standard. This comparative static highlighted two mechanisms affecting world export shares. First, as capabilities rise in good k, the country produces more of k and gains an increasing share of world exports. This is the direct or quality effect. Second, as quality rises for good k + 1 wages are pushed up, which erodes the country’s competitiveness in good k. This is the general equilibrium Ricardian or wage effect. These two mechanisms lead to the world export share predictions in figure 2. For middle-capability goods (k), world export shares display an inverted-U shaped relationship with income as first the quality effect and then the wage effect come into play. For low-capability goods (k − 1), only the wage effect is relevant and world export shares decline in income. For high-capability goods (k + 1) only the quality effect is relevant and world export shares are increasing in income. See figure 3. We operationalize these distinct export-share predictions of goods k − 1, k, and k + 1 as follows. Recall that informative goods were those with ln ymin,g > 8.25 (the H group) and ln ymax,g < 10 (the L group). We associate good k + 1 with goods in the H group, good k with goods in the top half of the L group (6 < ln ymax,g < 10), and good k − 1 with goods in the bottom half of the L group (ln ymax,g < 6). This gives us three groups: the ‘High’ group (k + 1), the ‘Middle’ group (k), and the ‘Low’ group (k − 1). Since we must pool across exporters and products, we will need to consider normalizations of export shares that are designed to control for country and product size. Consider country `’s share of world exports of good g. We normalize the income level of ` by reference to the income levels y g and y g of the poorest and richest exporters, respectively, of g; we represent the position of log GDP 24 per capita within this range as (ln y` − ln y )/(ln y g − ln y ).18 g g We also need to adopt some normalization for the level of exports. This will be affected, as the theory indicates, by product-market size and country size. The global market size for product g is given by Sg (or equivalently δg ) in the theory. See section 2.1. To control for Sg , we scale country `’s exports of good g, Xg` by world exports of g, Xg ≡ Σ` Xg` . Country size is constant in the theory. In pratice, larger countries will produce a larger subset of the m products in any given capability group. To control for this we scale Xg` /Xg by its average Σ g ( Xg` /Xg )/n` where n` is the number of goods exported by country ` i.e. the number of goods for which Xg` > 0. Summarizing, we plot (Normalized GDP per Capita) g` ≡ ln y` − ln y ln y g − ln y g (19) g against (Normalized World Export Share) g` ≡ ( Xg` /Xg ) 1 n` Σ g ( X g` /X g ) (20) where the numerator is country `’s share of world exports of g and the denominator is country `’s average share of world exports. Figure 6 plots normalized world export shares against normalized GDP per capita for our three groups of goods. The first thing to note about the plots is the preponderance of points on or very near the horizontal axis. This reflects the fact that poorer countries have zero exports of many goods. This fact could have been built into the model by having our small country produce only one of m possible products in group k. It is also a feature of the Eaton and Kortum (2002) model where a country can potentially produce all goods but draws high productivity for only a subset of goods. Hence in interpreting these scatters, our focus of interest lies not on means — which tend to be dominated by the many ( g,`) pairs with near-zero exports — but on the upper bound of the scatter. With this in mind, we estimate a quantile regression (the 90th quantile). This appears as the curve shown in each of the panels of figure 6.19 The upper, middle, and lower panels correspond to Low-group goods (k − 1), Middle-group goods (k), and High-group goods (k + 1), respectively. The panels bear out the world-export-share predictions of the model. World export shares are decreasing in income for the Low group, increas18. Note that y differs from ymin,g . The former is the minimum across all countries that export any positive amount of g good g while the latter is the minimum across any country for which g is a significant export. Likewise for the difference between y g and ymax,g . 19. We use the SAS QUANTREG procedure with a 6th order polynomial. 25 Figure 6: Normalized World Market Shares Notes: Each point in the plot corresponds to a product-country ( g,`) pair. The vertical axis is country `’s share of world exports of good g, normalized as in equation (20). The horizontal axis is country `’s income, normalized as in equation (19). The curves are the 90th quantile regressions. 26 ing in income for the High group, and display an inverted-U relationship for the Middle group. This is exactly as predicted in figure 3. We round out this section with a discussion of specification issues. In constructing figure 6 we made five choices. First, we defined the low, middle and high groups by reference to the cut-offs ln ymin,g = 6, ln ymin,g = 8.25 and ln ymax,g = 10. The choice of cut-offs does not matter. As on-line appendix figure A2 shows, figure 6 is not substantially altered by lowering the ln ymin,g = 6 cut-off to 5.25 or raising it to 6.75; by lowering the ln ymin,g = 8.25 cut-off to 7.25 or raising it to 9.25; or by lowering the ln ymax,g = 10 cut-off to 9. Changing the cut-offs beyond these ranges results in groups that include either too few observations or too many wide product-range, uninformative goods. The only sensitivity occurs when ln ymin,g = 6 is raised to 6.75, in which case the interior peak for the Low group shifts from 0.05 to 0.20. This is because the Low group now contains so many Middle group goods. Thus, figure 6 is robust to changes in cut-offs. Second, we used SITC4 data but the figures look virtual identical using HS6 data. See on-line appendix figure A3. The exception is that the Low group again has an interior peak near 0.20. Third, we chose the normalization (ln yk − ln y )/(ln y g − ln y ). This choice of normalization plays almost g g no role empirically because y g − y does not vary much across goods i.e. it is not much different g than normalizing by a constant. Specifically, across all goods its maximum is 6.4, its median is 6.0 and its 5th percentile is 5.2. Fourth, in equation (20) we normalized world export shares Xg` /Xg by 1 average world export shares n− k Σ g ( X g` /X g ). We have experimented extensively with alternative normalizations of Xg` /Xg , including the 50th, 75th, 90th, and 99th percentiles of each country’s world export shares.20 All of these normalizations produce curves with the same shapes as those in figure 6. Fifth, we reported quantile regressions based on the 90th quantile. The curves do not change when using the 85th, 95th, or 99th quantiles.21 9. Development Ladders in the Cross-Section: The Role of Product Ranges Wide product ranges and their correlation with unit values suggest that a country’s wealth depends not just on what goods it produces, but also on the quality of the goods produced. To investigate 20. For each country ` these are percentiles of the set { X1` /X1 , . . . , Xg` /Xg , . . . , xG` /xG }. 21. There are a few extreme ‘vertical’ outliers that would ‘squash’ figure 6 down to the horizontal axis if displayed. Rather than leave them off the figure entirely, we shrink them towards the horizontal axis as follows. In the top panel, if a vertical point y exceeds 9, then it is replaced by 9 + f (∆) where ∆ = y − 9 and f (∆) = ln(1 + ∆)/5 so that f (0) = 0 and f 0 > 0. Likewise for the middle and bottom panels, but with 4 instead of 9. This does not have and cannot have an effect on the position of the quantile regressions. 27 further, return to figure 4 and, as before, divide goods into three groups: (i) the L group to the lower left (ln ymax,g < 10); (ii) the H group to the upper right (ln ymin,g > 8.25); and, (iii) the ‘uninformative’ group, comprising the remaining goods. In what follows we confine attention to the L and H goods because these are the only ones that are informative.22 We show in figure 7 the relationship of a country’s GDP per capita with (a) the share of L goods in its export basket (left panel) and (b) the share of H goods in its export basket (right panel). We see a clear fall in the share of L goods and a rise in the share of H goods as GDP per capita increases. But an important feature of figure 7 lies in the fact that the relation between the product mix and income is not bi-directional: while significant exporters of H goods are necessarily rich, it is not the case that rich countries are necessarily significant exporters of H goods. A very low contribution of H goods is consistent with a relatively high level of GDP per capita. (See Malaysia in the right-hand panel). Similarly, while a high share of L goods necessarily implies that a country is poor, many poor countries have a low share of L goods. (See Zimbabwe or Bangladesh in the left-hand panel.) We can restate this in a way that makes one of the key points of our thesis crystal clear. A poor country can advance out of L goods and still remain poor: this happens when the country enters as a low-quality producer into uninformative goods i.e. goods with wide quality ranges. Since most goods are uninformative, we might expect this type of no-growth shift in product mix to be common. By the same token, a country may move from being poor to being rich without changing its product mix: this happens when it improves the quality of the uninformative, wide-quality-range goods that it already exports. 10. Conclusions The aim of this paper has been to explore the way in which advances in wealth are associated both with changes in the product mix and with changes in quality (and productivity) within a given set of industries. The central point relates to the fact that the range of income levels of significant exporters of most products is very wide. At a theoretical level, one reason for the wide product ranges lies in aggregation of disparate sub-industries (though the result is not sensitive to the level of aggregation); another reason lies in the fact that within any industry, there will, in a general 22. The conclusions of this section are in no way sensitive to the choice of 8.25 and 10 as cut-offs: we have chosen these because they represent break points in figure 4. Also, in figure 4 there is one good that is in both the L and H groups. This is cameras (SITC4 8732). It lies at the point (ln ymin,g , ln ymax,g ) = (8.9,9.8). We place this in the H group, though where it goes makes no difference because it is only one of hundreds of goods. 28 Figure 7: The Share of L and H Goods in Each Country’s Export Basket 0.12 Low (L) Group 0.50 Share of H Goods in a Country's Export Basket Share of L Goods in a Country's Export Basket 0.60 0.40 0.30 0.20 0.10 0.00 ZWE 4 High (H) Group 0.10 0.08 0.06 0.04 0.02 MYS BGD 0.00 6 8 10 12 4 6 log GDP per Capita, ln(yk) 8 10 12 log GDP per Capita, ln(yk) Notes: Each point represents a country (there are 94 points in each panel). The horizontal axis is log GDP per capita in 2005. The vertical axis is a country’s exports of L goods (right panel) or H goods (left panel) as a share of the country’s total exports. Using the product ranges in figure 4, a good is an L good if ln ymax,g < 10 and an H good if ln ymin,g > 8.25. equilibrium multi-country setting, be a viable range of producer income levels. In this viable range, poor low-quality exporters compete with rich high-quality exporters. The central property of product ranges is that, in a multi-market general equilibrium setting, the relation between quality and price on the one hand, and output and global market share on the other, is non-monotonic. There is at equilibrium a range of producer qualities (and so wealth levels) that are viable in a given industry. As quality rises, the country moves into the production of higher-ranked goods, and its equilibrium wage (and GDP per capita) rises. But this means that its output and global market share all exhibit an inverted-U relationship with quality, and so with GDP per capita. As quality rises, market share rises, and wages rise also. As the country advances into the production of higher-ranked products, the rise in wage causes its effective cost level to rise, and its global market share in this industry to fall. It is this inverted-U relation that is the basis of the selection effect that links a country’s wealth to its product mix. 29 References Baldwin, Richard and James Harrigan, “Zeros, Quality, and Space: Trade Theory and Trade Evidence,” American Economic Journal: Microeconomics, May 2011, 3 (2), 60–88. Bernard, Andrew B., Jonathan Eaton, Bradford Jensen, and Samuel Kortum, “Plants and Productivity in International Trade,” American Economic Review, September 2003, 93 (4), 1268–1290. Chenery, Hollis B., “Patterns of Industrial Growth,” American Economic Review, 1960, 50 (4), 624–654. Choi, Yo Chul, David Hummels, and Chong Xiang, “Explaining Import Quality: The Role of Income Distribution,” Journal of International Trade, April 2009, 77 (2), 265–275. di Giovanni, Julian and Andrei A. Levchenko, “Country Size, International Trade, and Aggregate Fluctuations in Granular Economies,” Working Paper No. 17335, National Bureau of Economic Research, August 2011. Eaton, Jonathan and Samuel Kortum, “Technology, Geography, and Trade,” Econometrica, September 2002, 75 (5), 1741–1779. , , and Sebastian Sotelo, “International Trade: Linking Micro and Macro,” Working Paper No. 17864, National Bureau of Economic Research, February 2012. Fajgelbaum, Pablo, Gene M. Grossman, and Elhanan Helpman, “Income Distribution, Product Quality, and International Trade,” Journal of Political Economy, August 2011, 119 (4), 721– 765. Flam, Harry and Elhanan Helpman, “Vertical Product Differentiation and North-South Trade,” American Economic Review, December 1987, 77 (5), 810–822. Hallak, Juan Carlos, “Product Quality and the Direction of Trade,” Journal of International Economics, January 2006, 68 (1), 238–65. , “A Product-Quality View of the Linder Hypothesis,” Review of Economics and Statistics, August 2010, XCII (3), 453–466. and Jagadeesh Sivadasan, “Firms’ Exporting Behavior under Quality Constraints,” Working Paper No. 14928, National Bureau of Economic Research, April 2009. and Peter Schott, “Estimating Cross-Country Differences in Product Quality,” Technical Report 2011. Hausmann, Ricardo, Jason Hwang, and Dani Rodrik, “What You Export Matters,” Journal of Economic Growth, March 2007, 12 (1), 1–25. Hummels, David and Peter Klenow, “The Variety and Quality of a Nation’s Exports,” American Economic Review, June 2005, 95 (3), 704–23. Johnson, Robert C., “Trade and Prices with Heterogeneous Firms,” Journal of International Economics, January 2012, 86 (1), 43–56. Khandelwal, Amit, “The Long and Short (of) Quality Ladders,” Review of Economic Studies, October 2010, 77 (4), 1450–1476. Kugler, Maurice and Eric Verhoogen, “Prices, Plant Size, and Product Quality,” Review of Economic Studies, January 2012, 79 (1), 307–339. 30 Lall, Sanjaya, John Weiss, and Jinkang Zhang, “The “Sophistication” of Exports: A New Trade Measure,” World Development, February 2006, 34 (2), 222–37. Leamer, Edward E., Sources of International Comparative Advantage: Theory and Evidence, Cambridge, MA: MIT Press, 1984. , “Paths of Development in the Three-Factor, n-Good General Equilibrium Model,” Journal of Political Economy, 1987, 95, 961–991. Melitz, Marc J., “The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity,” Econometrica, November 2003, 71 (6), 1695–1725. and Daniel Trefler, “Gains from Trade when Firms Matter,” Journal of Economic Perspectives, Spring 2012, 26 (2), 91–118. Michaely, Michael, Trade, Income Levels, and Dependence, Amsterdam, North-Holland and New York: Elsevier Science Pub. Co, 1984. Neary, J. Peter, “Presidential Address: Globalization and Market Structure,” Journal of the European Economic Association, Papers and Proceedings, April–May 2003, 1 (2/3), pp. 245–271. and Joe Tharakan, “International Trade with Endogenous Mode of Competition in General Equilibrium,” Journal of International Economics, 2011. Schott, Peter, “One Size Fits All? Heckscher-Ohlin Specialization in Global Production,” American Economic Review, June 2003, 93 (2), 686–708. , “Across-Product versus Within-Product Specialization in International Trade,” Quarterly Journal of Economics, May 2004, 119 (2), 647–678. , “The Relative Sophistication of Chinese Exports,” Economic Policy, 2008, 53, 5–49. Sutton, John, Sunk Costs and Market Structure, Cambridge: MIT Press, 1991. , Technology and Market Structure, Cambridge: MIT Press, 1998. , “Market Structure: Theory and Evidence,” in Mark Armstrong and Robert Porter, eds., Handbook of Industrial Organization, Volume III, Amsterdam: North-Holland, 2007, pp. 2301–2368. , “Quality, Trade and the Moving Window: The Globalisation Process,” Economic Journal, November 2007, 117 (524), F469–F498. Verhoogen, Eric, “Trade, Quality Upgrading and Wage Inequality in the Mexican Manufacturing Sector,” Quarterly Journal of Economics, May 2008, 123 (2), 489–530. 31 Appendix A. Proofs Appendix A.1. Proof of Lemma 1 (Perfect Sorting) In the main text we established that the wk and pk of equations (6) and (7) uniquely clear labour and product markets. It is straightforward to show that trade is balanced (the value of total exports equals the value of total imports). This is built into the above equations via the consumer optimization problem i.e., via Walras’ Law. Thus, all of the requirements of equilibrium are satisfied except one. Specifically, in deriving equations (6) and (7) we have assumed that firms from country group k maximize profits by producing within product group k. We must now verify that these firms could not earn higher profits or even positive profits by producing in some other product group. By assumption, a firm from country k0 has quality 0 when producing a good in group k > k0 . Hence, it will never produce a group-k good. Consider a firm from country k0 = k + i (i > 0) that is considering whether to produce a good from group k. The firm earns lower profits doing so if and only if the firm faces wages wk+i (> wk ) that are so high as to render the firm unviable in the production of good type-k goods. Using equation (4) with the inequality reversed and with u j = u for all j (recall that u denotes the common standard of quality shared by all active firms in the market), the firm is not viable in type-k goods if wk+i > ( Nk wk + wk+i )/( Nk − 1 + 1) or Nk w k +i > . wk Nk − 1 (21) Substituting in wages from equation (6) and simplifying, inequality (21) is equivalent to Nk > Nk2+i /( Nk+i − 1) or, upon subtracting Nk+i from both sides, Nk − Nk+i > Nk+i . Nk+i − 1 (22) By assumption 1, the left-hand side is strictly greater than 3. By the assumption that there are at least two producers globally, Nk+i ≥ 2 and the right-hand side is less than or equal to 2. Hence inequality (22) holds. Thus, a firm from country k + i (i > 0) earns negative profits producing a group-k product.23 Finally, equations (5)-(8) establish that wk , yk , pk and pk /wk are decreasing in Nk and hence, by assumption 1, increasing in k. 23. Notice that we only used Nk ≥ Nk+1 + 3, which is weaker than assumption 1. In order to ensure that inequality (22) is still satisfied when we abandon the perfect sorting equilibrium and allow one country to move from type k − 1 to type k, we will need the stronger assumption 1 that Nk ≥ Nk+1 + 4. 32 Appendix A.2. Equilibrium Conditions and Outcomes in Phase I A typical type-(k − 1) country producing a group-(k − 1) good has wage wk−1 and price pk−1 . Rearranging equation (13) as 1 − wk−1 /pk−1 = w/[ pk−1 ( Nk−1 − 1)] and plugging this into equation (15) yields: N=m w p2k−1 ( Nk−1 − 1) S or p k −1 = w mS N Nk−1 − 1 1/2 (150 ) . Likewise re-arranging equation (14) as 1 − wk /pk = [wu/vk ]/[ pk Nk ] and plugging this into equation (16) yields: N=m w S (vk /u) p2k Nk or pk = w mS N (vk /u) Nk 1/2 (160 ) . For the developing country, equations (11)–(12) imply pk−1 = pk−1 and pk = (vk /u) pk . Plugging these into equation (17) yields: N 1 = mS p k −1 1− w p k −1 1 + (vk /u) pk w 1− (vk /u) pk . (170 ) Plugging equations (150 )–(160 ) into equation (170 ) establishes that the equilibrium wage in the developing country is: w(vk ) m = S(vk ) N " #2 p ( Nk−1 − 1)1/2 + Nk1/2 / vk /u) . Nk−1 + Nk /(vk /u) (23) Plugging this back into equations (150 )–(160 ) yields closed-form solutions for the equilibrium prices pk−1 (vk )/S(vk ) and pk (vk )/S(vk ). (Recall that all solutions are normalized by the numeraire S(vk ).) Plugging w(vk ), pk−1 (vk ) and pk (vk ) into equations (11)–(14) yields closed-form solutions for the remaining equilibrium wages and prices, namely, pk−1 (vk ), pk (vk ), wk−1 (vk ), and wk (vk ), all relative to the numeraire S(vk ). Appendix A.3. Equilibrium Conditions and Outcomes in Phase II Plugging pk−1 from equation (13–II) into equation (15) and rearranging yields: w k −1 ( v k ) m Nk−1 − 2 . = S(vk ) N ( Nk−1 − 1)2 33 (15–II) For closed-form solutions to all of the equilibrium prices and wages in Phase II proceed as follows. Plugging equation (15–II) into (13–II) yields a closed-form solution for pk−1 (vk )/S(vk ). The equation (160 ) expression for pk holds in Phase II, as does the equation (11) expression pk = 1/2 w (vk /u) pk . Hence, pk = (vk /u) pk = (vk /u) mS . Plugging this into equation (17–II) and N (v /u) N k k rearranging yields: w(vk ) m α v /u = where α ≡ k <1. 2 S(vk ) N (1 + α ) Nk (24) Plugging equation (24) into equation (160 ) yields a closed-form equilibrium solution for pk (vk )/S(vk ). Plugging this and w/S from equation (24) into (14) yields a closed-form equilibrium solution for wk (vk )/S(vK ). Thus, we have provided closed-form equilibrium solutions for all of the endogenous prices in Phase II. Appendix A.4. Proof of Lemma 2 Let vkL be the value of vk at which the developing country is about to produce a type-k good, but has not yet started. At this point its wage w(vkL ) is that of a typical type-(k − 1) country i.e., w(vkL ) = wk−1 where wk−1 is given in equation (6). Equating the equation (6) expression for wk−1 with the equation (23) expression for w(vkL ) and solving for vkL yields: vkL = ( Nk−1 − 1) Nk u. Nk2−1 (25) Let vkH be the value of vk at which the developing country stops producing goods of type-(k − 1) n o w(v H ) i.e., xk−1 (vkH ) = 0. From equation (30 ), xk−1 (vkH ) = p 1(v H ) 1 − p (kv H ) so that xk−1 (vkH ) = 0 k −1 k k −1 k implies pk−1 (vkH ) = w(vkH ). Plugging this into equation (150 ) yields: w(vkH ) 1 m . = H N ( Nk−1 − 1) S(vk ) (26) Equating this to w(vkH )/S(vkH ) from equation (23) yields: r p vk p vk − Nk−1 − 1 Nk + Nk = 0 (27) u u √ which is quadratic in vk /u. The larger of the two roots exceeds 1 and hence is not a candidate for √ √ √ calculating vkH because we require vk < u.24 The other root is ( Nk /2) Nk−1 − 1 − Nk−1 − 5 so that vkH = u 2 p Nk p Nk−1 − 1 − Nk−1 − 5 . 4 (28) p p √ 24. This root is ( Nk /2) Nk−1 − 1 + Nk−1 − 5 . Since Nk ≥ 2, assumption 1 implies Nk−1 ≥ 6. Together these imply that this root exceeds 1. 34 It is straightforward but tedious to verify that vkL < vkH < u. See the on-line Appendix E.1 for a step-by-step derivation. Appendix A.5. Monotonicity of Wages in Phase I and II Lemma 4 (1) w(vk )/S(vk ) is increasing in vk for vk in the range vkL < vk < u. (2) w(vk )/S(vk ) vk /u is decreasing in vk for vk in the range vkL < vk < vkH . Proof of part: Consider Phase I where vkL < vk < vkH . Differentiating equation (23) with respect to vk yields √ √ ∂w(vk )/S(vk ) = αk ξ + − vk vk − ξ − ∂vk (29) where αk > 0 is a constant (i.e., independent of vk ) and ξ+ = u i Nk1/2 h ( Nk−1 − 1)1/2 − (2Nk−1 − 1)1/2 < 0 Nk−1 (30) u i Nk1/2 h ( Nk−1 − 1)1/2 + (2Nk−1 − 1)1/2 > 0. Nk−1 (31) √ ξ− = √ It is tedious but straightforward to show that Nk ≥ 2 and assumption 1 imply q vkH < ξ + . See q < vkL . Hence on-line Appendix E.2 for a step-by-step derivation. Since ξ − < 0, it must be that ξ − q q √ vkH ). Hence, ξ − < vk < ξ + in Phase I. Hence the interval (ξ − , ξ + ) contains the interval ( vkL , the right-hand side of equation (29) is positive and w(vk )/S(vk ) is increasing in vk . Next consider Phase II where vkH < vk < u. From equation (24), w(vk )/S(vk ) is increasing in α and hence in vk . Proof of part 2: (vkL < vk < vkH ): Dividing equation (23) by (vk /u)S(vk ) yields: w(vk )/S(vk ) m = vk /u N " #2 " #2 √ √ ( Nk−1 − 1)1/2 + Nk1/2 / vk /u m ( Nk−1 − 1)1/2 vk /u + Nk1/2 √ √ = . N Nk−1 (vk /u) + Nk Nk−1 vk /u + Nk / vk /u Differentiating, ∂ w(vk )/S(vk ) vk /u ∂vk s ) r vk Nk vk Nk = −αk (vk ) +2 − u Nk−1 − 1 u Nk−1 − 1 s s "r # "r # √ √ vk Nk vk Nk = −αk (vk ) + ( 2 + 1) − ( 2 − 1) u Nk−1 − 1 u Nk−1 − 1 where αk (vk ) ≡ 2(m/N)1/2 ( (25) and Nk−1 > 2 imply that ( w(vk )/S(vk ) 1/2 ) Nk−1 ( Nk−1 vk /u q − 1)1/2 /[ Nk−1 /(vk /u) + Nk ]2 > 0. Equation p √ vkL /u > ( 2 − 1) Nk /( Nk−1 − 1). Hence in Phase I both bracketed terms are positive and the derivative is negative. 35 Appendix A.6. Proof of Proposition 2 Part 1 (wk−1 /S increasing in vk ): Equating the right-hand sides of equations (13) and (150 ) yields: 1/2 1/2 m w k −1 ( v k ) 1 w(vk ) w(vk ) 1 = . (32) − S(vk ) N Nk−1 − 1 S(vk ) Nk−1 − 1 S(vk ) √ √ √ The right-hand side of equation (32) is quadratic in w/S and thus increasing in w/S for w/S ≤ √ √ m/N Nk−1 − 1/2. It thus suffices to show that this last inequality is satisfied in Phase I. Since w/S is increasing in vk (lemma 4), it suffices to show that the inequality holds at vk = vkH , which m 1 1/2 ≤ [ m ( N 1/2 /2 or 4 ≤ ( N 2 from equation (26) implies [ N k −1 − 1)] k −1 − 1) or Nk −1 ≥ 3. But Nk−1 −1 ] N Nk ≥ 2 and assumption 1 together imply Nk−1 > 3. Thus, the inequality holds at vkH . Part 1 (wk /S decreasing in vk ): We start with a preliminary result. At vk = vkL the developing country does not produce xk . Hence we are in the perfect sorting equilibrium of section 3: w(vkL ) = wk−1 (vkL ) which, from equation (6), gives us w(vkL )/S(vkL ) = Nk−1 −1 m . Nk2−1 N Further, from equation (25), w(vkL )/S(vkL ) N − 1 m ( Nk−1 − 1) Nk 1 m = k−21 / = . Nk N Nk−1 N Nk2−1 vkL /u (33) We now turn to the core of the proof. Equating the right-hand sides of equations (14) and (160 ) yields: wk ( v k ) = S(vk ) m 1 w(vk )/S(vk ) N Nk vk /u 1/2 − 1 w(vk )/S(vk ) . Nk vk /u This is just equation (32) with Nk replacing Nk−1 − 1 and (w/S)/(vk /u) replacing w/S. Thus, p p wk /S is increasing in (w/S)/(vk /u) if (w/S)/(vk /u) ≤ (m/N) Nk−1 /2. Since (w/S)/(vk /u) is decreasing in vk (part 2 of lemma 4), the inequality holds throughout Phase I if it holds at vk = vkL . But from equation (33), this means 1 m Nk N ≤ Nk m 4 N or 4 ≤ Nk2 , which always holds. Hence wk /S is increasing in (w/S)/(vk /u). But (w/S)/(vk /u) is decreasing in vk (part 2 of lemma 4). Hence wk (vk )/S(vk ) is decreasing in vk . Part 2: Dividing both sides of equation (32) by w(vk )/S(vk ) establishes that wk−1 /w is decreasing in w/S and hence, by lemma 4, decreasing in vk . Appendix A.7. Proof of Proposition 1 Part 1: From equation (30 ), , xk−1 = (S/pk−1 )(1 − w/pk−1 ). From equation (150 ), pk−1 /S is increasing in w/S. Dividing equation (150 ) by w(vk ), p k −1 ( v k ) = w(vk ) m 1 S(vk ) N ( Nk−1 − 1) w(vk ) 36 1/2 (34) which is decreasing in w/S. Hence xk−1 (vk ) is decreasing in w/S and, by lemma 4, decreasing in vk . By the definition of vkH , xk−1 (vkH ) = 0 so that xk−1 (vk ) falls to 0. By labour-market clearing, mxk−1 + mxk = N so that xk is increasing in vk . By the definition of vkL , xk (vkL ) = 0 so that xk (vk ) rises from 0. Part 2: From equations (11) and (34), pk−1 (vk )/w(vk ) is decreasing in w/S and hence, by lemma 4, pk−1 (vk )/w(vk ) is decreasing in vk . From equation (160 ), pk (vk ) = w(vk ) m 1 1 1 N Nk vk /u w(vk )/S(vk ) 1/2 . (35) From equation (11), pk = (vk /u) pk . Hence: (vk /u) pk (vk ) pk (vk ) = = w(vk ) w(vk ) m 1 N Nk 1/2 w(vk )/S(vk ) vk /u −1/2 (36) which, by lemma 4, is increasing in vk . Part 3: See lemma 4. Part 4: Revenue for any firm i is Ri = pi xi . Multiplying the expression for xi in equation (30 ) by pi yields Ri /S = pi xi /S = 1 − wi /pi . Hence, for a developing country producing a type-(k − 1) good, R k −1 ( v k ) w(vk ) = 1− = 1− S(vk ) p k −1 ( v k ) N w(vk ) ( Nk−1 − 1) m S(vk ) 1/2 , (37) where the second equality follows from equation (11) (pk−1 = pk−1 ) and the equation (150 ) expression for pk−1 (vk ). For a developing-country firm producing a type-k good, w(vk ) Rk (vk ) = 1− = 1− S(vk ) pk (vk ) N w(vk )/S(vk ) N m k vk /u 1/2 (38) where the second equality follows from equation (12) (pk = (vk /u) pk ) and the equation (160 ) expression for pk−1 (vk ). GDP per capita of the developing country is given by y = (mRk−1 + mRk )/N (see equation 10) so R k −1 ( v k ) R k ( v k ) + S(vk ) S(vk ) √ 1/2 m m 1/2 p N w(vk ) =2 − Nk−1 − 1 + √ k . N N S(vk ) vk /u y(vk ) m = S(vk ) N 37 Substituting into this the equation (23) expression for w(vk )/S(vk ) and simplifying yields: √ √ √ m m 1 − 2 Nk−1 − 1 Nk u/vk y(vk ) = + · . S(vk ) N N Nk−1 + Nk (u/vk ) bx The second expression on the right-hand side is a rational function of the form ca−−dx 2 with a, b, c, √ d > 0. It is positive and decreasing in x (= u/vk ) on the domain 0 ≤ x ≤ a/b, or 0 ≤ vuk ≤ 4( Nk−1 − 1) Nk and so y(vk )/S(vk ) is increasing in vk on this domain. Finally, we note that the relevant domain (vkL ,vkH ) is a subset of this domain; for this it suffices to note from the expression (25) for vkL that vkL ( Nk−1 − 1) Nk = ≤ 4( Nk−1 − 1) Nk u Nk2−1 which follows from the fact that Nk−1 ≥ 2. This completes the proof of Proposition 1. Appendix A.8. Proof of Proposition 3 Part 2 follows immediately from part 1 of lemma 4. Since equations (11) and (160 ) hold in Phase II, so do equations (36) and (38). Part 1 follows from equation (36) and part 2 of lemma 4. For part 3, start by observing that Rk (vk )/S(vk ) rises in vk , which follows from equation (38) and part 2 of lemma 4. In Phase II all income comes from production of type-k goods so that y(vk )/S(vk ) = (m/N) Rk (vk )/S(vk ), which is also rising in vk . Appendix A.9. Proof of Lemma 3 Consumers spend a fraction δ of income on each good. It follows that the value of exports of a typical type-k country is X k = pk x k − δmpk x k where mpk x k is income or GDP. Labour market equilibrium is N = mx k . It follows that X k (vk ) N pk (vk ) X k −1 ( v k ) N p k −1 ( v k ) = (1 − δm) and = (1 − δm) S(vk ) m S(vk ) S(vk ) m S(vk ) (39) where the right-hand equation follows from the left-hand equation by symmetry. Turning to the developing country, consider Phase II where only group-k goods are produced and exported. By the logic of equation (39), Xk /S = (1 − δm)(N/m)( pk /S), which is increasing in pk /S and hence, by parts 1 and 2 of proposition 3, in vk . Further, Xk /X k = ( Xk /S)/( X k /S) = pk /pk = vk /u where the last equality follows from equation (12). Hence Xk /X k is increasing in vk . Next consider Phase I, where both goods are produced. Let Rk−1 ≡ pk−1 xk−1 and Rk ≡ pk xk be revenues so that income or GDP is m ( Rk−1 + Rk ). It follows that the value of the developing 38 country’s exports of group-(k − 1) and group-k goods are given by: Xk−1 (vk ) = max {0, Rk−1 (vk ) − δm [ Rk−1 (vk ) + Rk (vk )]} = max {0, (1 − δm) Rk−1 (vk ) − δmRk (vk )} Xk (vk ) = max {0, (1 − δm) Rk (vk ) − δmRk−1 (vk )} . (40) (The max operator is needed because it is possible that one of the two goods is imported.) Revenue for any firm i is Ri = pi xi . Multiplying the expression for xi in equation (30 ) by pi yields Ri = pi xi = {1 − wi /pi }S. Hence for the developing country producing group-(k − 1) and group-k goods, Rk−1 /S = (1 − w/pk−1 ) and Rk /S = (1 − w/pk ). From part 2 of proposition 1, a Phase I rise in vk raises w/pk−1 and lowers w/pk so that Rk−1 /S falls and Rk /S rises. Hence from equation (40), Xk−1 (vk )/S(vk ) falls and Xk (vk )/S(vk ) rises in Phase I. Consider X k−1 /S. Equation (13) and proposition 2 imply that pk−1 /S is increasing in vk in Phase I. Hence from equation (39), X k−1 /S is increasing in vk in Phase I, as required. But we have already seen that Xk−1 /S falls. Hence Xk−1 /X k−1 falls. Finally, in Phase I, Rk−1 /S is falling so that by equation (40), Xk /S is rising faster than (1 − δm) pk xk /S. Compare this to X k /S = (1 − δm)( pk /S)(N/m) from equation (39). By equation (12), pk /S is rising faster than pk /S. Also, xk is rising faster than the constant N/m. Hence Xk /X k is rising. Appendix B. Trade Data COMTRADE reports each bilateral transaction twice, once by the importer and once by the exporter. We always use the importer’s data as this is known to be more reliable for most countries. The countries in our sample are (using ISO codes for brevity25 ): AFG, AGO, ALB, ARG, AUS, AUT, BDI, BEN, BFA, BGD, BGR, BOL, BRA, CAN, CHE, CHL, CHN, CMR, COL, CRI, CUB, DEU, DNK, DOM, DZA, ECU, EGY, ESP, FIN, FRA, GBR, GHA, GIN, GRC, GTM, HND, HTI, HUN, IDN, IND, IRL, IRQ, ISR, ITA, JAM, JOR, JPN, KEN, KHM, LBN, LKA, MAR, MDG, MEX, MLI, MMR, MOZ, MWI, MYS, NER, NGA, NIC, NLD, NOR, NPL, NZL, PAK, PER, PHL, PNG, POL, PRT, PRY, ROU, RWA, SAU, SDN, SEN, SGP, SLE, SLV, SOM, SWE, TCD, TGO, THA, TUN, TUR, UGA, URY, USA, VEN, ZMB, and ZWE. The only major countries not included in our list are Taiwan and Honk Kong. Taiwan is excluded because there are no 1980 data. Hong Kong is excluded because, for our purposes, it should be merged with China in 2005 and be by itself in 1980. None of our 2005 25. See on-line appendix table A1 for a full list of country names and GDPs per capita. 39 cross-section results are affected by the inclusion of Taiwan and Hong Kong (the latter either by itself or merged with China). We exclude live animals, meat, fish and dairy. These goods account for only 2.1% of trade and including them does not affect our results at all; however, it is hard to relate trade in these goods to the issues raised in this paper. Price data p gk are from the U.S. historical imports CD, 2001–2005. This CD only reports what is called the ‘first quantity’ and ‘first value’ so that all observations within an HS10 product have the same quantity units. We sum U.S. imports and quantities by HS10 product and trading partner (exporter to the United States). We calculate unit values with the summed data. In addition, we winsorize the unit values below the 10th within-HS10 percentile and above the 90th within-HS10 percentile. Winsorizing makes virtually no difference to our results. Appendix C. Price Ranges The theory predicts that, for a single good, all producers of the good will share the same pricequality ratio. Since richer countries have higher quality, they should have higher prices. That is, prices should be increasing in the income of the exporter. Since price data are not available, we follow Schott (2004) in proxying for prices using HS10 unit values from the 2005 U.S. import file. We emphasize that unit values are extremely noisy so that caution must be exercised in interpreting them as prices. See Appendix B for a discussion of the data. Let p gk be the unit value of good g exported by country k to the United States. We are interested in how the p gk vary as we move through product ranges. The most familiar way of doing this is Schott’s (2004, table V) famous regression ln p gk = α g + β ln yk where α g is an HS10 product fixed effect. Re-estimating Schott’s regression using 2005 U.S. imports from our 94 exporters (187,363 observations), the OLS estimate of β is 0.29 (clustered t = 8.05) so that, as in Schott, there is indeed a statistically significant positive correlation between unit values and exporter incomes. A sharper prediction of our theory is as follows. Consider a single HS10 good g. Recall that in the HS10 panel of figure 5 we plotted the income of the poorest and richest countries that had significant exports of g i.e. we plotted (ln ymin,g , ln ymax,g ). Since for each g we know the identity of the poorest and richest countries, we know these countries’ unit values. We denote them in obvious 40 fashion by pmin,g and pmax,g . We expect that ∆ g ≡ ln pmax ,g − ln pmin ,g > 0. This inequality is sharp in that it is directly related to our product ranges, that is, to the poorest (min) and richest (max) exporters that define the boundaries of our product ranges. It is also an inequality that is unlikely to hold because we are examining two specific unit values (pmax ,g and pmin ,g ) even though we know that such unit values are extremely noisy. ∆ g > 0 defines one inequality for each product range in the HS10 panel of figure 5. A nonparametric test of ∆ g > 0 is the sign test, which easily rejects the null hypothesis that the signs of the ∆ g are random (p-value of less than 0.0001). The mean value of ∆ g is 0.63 (t = 27.23) and, more robustly with noisy data, the median value of ∆ g is 0.45. Since e0.45 − 1 = 0.57, this implies that the richest significant exporter of the median product has a unit value that is 57% higher than the corresponding unit value of the poorest significant exporter. Cautiously interpreting unit values as prices, this means that prices are increasing as one moves through a product range in figure 5. It is tempting to examine an even stronger prediction, namely, that unit-value ranges ∆ g are large when product ranges ln ymax ,g − ln ymin ,g are large. While this is not a prediction of the model, it can be generated by adding more restrictions on how scarcity varies across countries and products. To examine this prediction we estimate the following regression: (ln pmax ,g − ln pmin ,g ) = 0.15 + 0.18(ln ymax ,g − ln ymin ,g ) (clustered t = 11.23). Thus, large product ranges in figure 5 are associated with large unit-value ranges. 41 On-Line Appendix – Not for Publication Online Appendix to “Deductions from the Export Basket: Capabilities, Wealth and Trade” On-Line Appendix – Not for Publication Figure A1: Product Ranges: Sensitivity to Cut-offs 12 12 11 Richest GDP per Capita (ymax) 10 9 8 7 6 5 9 8 7 6 5 SITC4, 1% Cut‐Off or $5M in Exports SITC4, 0.1% Cut‐Off 4 4 4 5 6 12 7 8 9 10 11 Poorest GDP per Capita (ymin) 12 Richest GDP per Capita (yymax) 11 Richest GDP per Capita (y (ymax) 10 10 9 8 7 6 5 4 6 7 8 9 7 8 9 10 11 12 12 11 11 10 9 8 7 6 10 Poorest GDP per Capita (ymin) 11 12 10 9 8 7 6 5 HS10, 1% Cut‐Off SITC4, 1% Cut‐Off or $50M in Exports 4 5 6 5 SITC4, 10% Cut‐Off 4 4 5 12 Richest GDP per Capita (yymax) Richest GDP per Capita (ymax) 11 4 5 6 7 8 9 10 11 Poorest GDP per Capita (ymin) 4 12 4 5 6 7 8 9 10 11 12 Poorest GDP per Capita (ymin) On-Line Appendix A. Robustness of Product Range Diagrams There were several choices made in constructing the product ranges of figures 4–5. When defining the product ranges we only looked at country-product pairs for which the product accounted for a significant share of the country’s exports. As in the main text, let x g` /x` be the share of country `’s exports accounted for by product g. We considered country-product pairs with x g` /x` ≥ α maxg0 x g0 ` /x` where α = 0.01. In the top left and bottom left panels of figure A1 we choose α = 0.001 and α = 0.10, respectively. As is apparent, the impression of the pervasiveness of wide product ranges does not change. An alternative to choosing cut-offs based on x g` /xk ≥ α maxg0 x g0 ` /x` is to use a dollar value cut-off: x g` > $5,000,000 or x g` > $50,000,000. This has the advantage that it keeps more of the trade of rich countries, and the disadvantage that it eliminates more of the trade of poor countries. We therefore combine this dollar criterion with our previous α (or percentage) criterion. That is, an observation ( g,`) is included if it meets either of the two criteria. The results appear in the right-hand panels of figure A1. The top and bottom panels use $5,000,000 and $50,000,000, respectively. On-Line Appendix B. Robustness of Market-Share Diagrams In section 8 we claimed that the figure 6 market-share diagram was not very sensitive to the choice of cut-offs ln ymin,g = 6, ln ymin,g = 8.25 and ln ymax,g = 10. Figure A2 shows the basis for this claim. In the top pair of panels, ln ymin,g = 6 is lowered to 5.25. In the second pair of panels, ln ymin,g = 6 is raised to 6.75. Here the monotonicity of the Low group is lost, but only because the Low group now absorbs so many Middle-group goods. It remains true that the Low-group peak is far to the left of the Middle-group peak. In the third pair of panels, ln ymin,g = 8.25 is lowered to 7.25. Monotonicity i On-Line Appendix – Not for Publication is lost for the High group, but only because it now includes so many Middle-group goods. (Also, the downturn at the far right of the High panel is entirely associated with Norway.) In the fourth pair of panels, ln ymin,g = 8.25 is raised up to 9.25. In the bottom pair of panels, ln ymax,g = 10 is lowered to 9. From figure A2 it is clear that the market-share predictions of figure 6 are not sensitive to the choice of cut-offs, or where they are, this ‘sensitivity’ is entirely explainable. The market-share diagram (figure 6) was drawn using SITC4 data. In figure A3 the figure is redrawn using HS6 data. Because there is much more data, we report the 95th quantile rather than the 90th quantile that was used for the SITC4 data. As is apparent, figures 6 and A3 are almost identical. The only difference lies in the Low group where the peak of the quantile is no longer at the far left; however, its peak remains close to the extreme left and certainly much further left than for the middle group.26 26. On a minor technical note, at the HS6 level there are a few goods that meet the criteria for being in both the Medium and High groups. Where this is the case the goods are put in the High group; however, since there are so few of these goods it makes no difference where they are put. ii On-Line Appendix – Not for Publication Figure A2: Normalized World Market Shares, Sensitivity to Cut-Offs er ln(ymin,g) = 6 to ln(ym Lowe min,g) = 5.25 Low Middle Raise ln(ymin,g) = 6 to ln(yymin,g) = 6.75 Low Middle Lower ln n(ymin,g) = 8.25 to ln(yymin,g) = 7.25 High Middle Raise ln(ymin,g) = 8.25 to ln n(ymax,g) = 9.25 Middle High Lower ln(ymax,g) = 10 to ln(yymax,g) = 9 Middle Low Notes: This figure reports the sensitivity of figure 6 to the choice of cut-offs defining the Low, Middle and High groups of goods. SITC4 data are used. The only sensitivity occurs when ln ymin,g = 6 is raised to 6.75, in which case the Low group has an interior peak. However, this peak is far to the left, at 0.20, and only appears because the Low group now contains so many Middle group goods. iii On-Line Appendix – Not for Publication Figure A3: Normalized World Market Shares, HS6 Data Normalized World Export Shares k ‐ 1 or Low w‐Group Goods Normalized World Export Shares k or Mediu um‐Group Goods Normalized World Export Shares k + 1 or High h‐Group Goods Normalized G GDP per Capita Notes: This figure repeats figure 6, but using HS6 COMTRADE data and the 95 quantile (since the HS6 data are much finer than the SITC4 data). iv On-Line Appendix – Not for Publication Table A1: List of Countries GDP per Capita (2005) Code Country $US BDI MWI ZWE RWA NER MMR SLE AFG NPL SOM MDG UGA MOZ GIN TGO BFA BGD HTI KHM MLI GHA BEN KEN TCD ZMB SDN IND SEN NGA PAK NIC PNG CMR BOL PHL IRQ HND IDN LKA PRY EGY CHN MAR AGO GTM JOR SLV 101 157 170 226 245 248 273 273 276 283 283 317 323 325 337 387 422 429 444 473 475 513 526 580 637 675 713 730 803 820 899 928 955 1,028 1,163 1,213 1,225 1,244 1,253 1,266 1,392 1,766 1,906 2,039 2,147 2,293 2,545 Burundi Malawi Zimbabwe Rwanda Niger Myanmar Sierra Leone Afghanistan Nepal Somalia Madagascar Uganda Mozambique Guinea Togo Burkina Faso Bangladesh Haiti Cambodia Mali Ghana Benin Kenya Chad Zambia Sudan India Senegal Nigeria Pakistan Nicaragua Papua New Guinea Cameroon Bolivia Philippines Iraq Honduras Indonesia Sri Lanka Papua New Guin. Egypt China Morocco Angola Guatemala Jordan El Salvador GDP per Capita (2005) ln y k 4.62 5.06 5.13 5.42 5.50 5.51 5.61 5.61 5.62 5.64 5.65 5.76 5.78 5.78 5.82 5.96 6.05 6.06 6.10 6.16 6.16 6.24 6.27 6.36 6.46 6.52 6.57 6.59 6.69 6.71 6.80 6.83 6.86 6.94 7.06 7.10 7.11 7.13 7.13 7.14 7.24 7.48 7.55 7.62 7.67 7.74 7.84 Code Country ALB COL ECU THA TUN PER DOM DZA BGR JAM CUB BRA ROU CRI ARG TUR URY MYS VEN LBN CHL MEX POL HUN SAU PRT ISR GRC ESP NZL SGP ITA DEU FRA CAN JPN AUS AUT GBR FIN NLD SWE USA DNK IRL CHE NOR Albania Colombia Ecuador Thailand Tunisia Peru Dominican Rep. Algeria Bulgaria Jamaica Cuba Brazil Romania Costa Rica Argentina Turkey Uruguay Malaysia Venezuela Lebanon Chile Mexico Poland Hungary Saudi Arabia Portugal Israel Greece Spain New Zealand Singapore Italy Germany France Canada Japan Australia Austria United Kingdom Finland Netherlands Sweden USA Denmark Ireland Switzerland Norway $US 2,691 2,739 2,794 2,797 2,846 2,911 3,073 3,115 3,441 3,622 4,093 4,260 4,557 4,616 4,728 4,969 4,996 5,098 5,374 5,436 7,297 7,365 7,923 10,942 13,119 17,457 19,389 25,562 25,947 26,789 26,968 30,053 33,718 33,862 35,071 35,646 36,321 36,760 36,954 37,307 38,512 39,539 41,348 47,839 48,373 49,282 63,704 ln y k 7.90 7.92 7.94 7.94 7.95 7.98 8.03 8.04 8.14 8.19 8.32 8.36 8.42 8.44 8.46 8.51 8.52 8.54 8.59 8.60 8.90 8.90 8.98 9.30 9.48 9.77 9.87 10.15 10.16 10.20 10.20 10.31 10.43 10.43 10.47 10.48 10.50 10.51 10.52 10.53 10.56 10.59 10.63 10.78 10.79 10.81 11.06 On-Line Appendix C. Full Country Names and GDP per Capita Table A1 provide full country names for the ISO codes listed in Appendix B. On-Line Appendix D. Additional Justification of Our Modelling Approach D.1. Alternative Models Used in the International Trade Literature There are a number of models of international trade that could have been used to deliver our results e.g. Eaton and Kortum (2002), Bernard, Eaton, Jensen and Kortum (2003) and Melitz (2003). Why have we not used these? The core of our model has two components. First, we assumed that quality capabilities are scarce and asymmetrically distributed across countries. Second, in equilibrium wages adjust to changes in quality capabilities so that rich countries are priced out of low-k goods. One could obtain our main results in these other models, but it would be less straightforward. v On-Line Appendix – Not for Publication For one, in these other models, scarcity is described by the distribution of productivities (G ( a) in Melitz, 2003; the Type II extreme value distribution in Eaton and Kortum, 2002 and Bernard et al., 2003). To use these other models one would have to provide a detailed specification of how these distributions vary across both countries and industries. This can be done, but it would require so much asymmetry that the elegance of these models would be lost. Re-stated, these other models are designed to handle within-industry heterogeneity and are less concerned with standard betweenindustry comparative advantage. In contrast, we have made the extreme assumption that there is at most one firm per country and focussed instead on the cross-country, cross-industry distribution of capabilities that are central to our Ricardian logic. For another, in these other models wage adjustment plays a role in determining entry thresholds. Beyond this, there is little discussion of the comparative advantage implications of wages or of why some countries are rich and others poor. It is these latter issues that are our main concern. In short, we have chosen the simplest model possible that focuses on (1) Ricardian asymmetries in the distribution of quality capabilities and (2) the role of wages as an adjustment mechanism in a multi-industry world populated by rich and poor countries. D.2. The Endogeneity of Quality Capabilities The driving assumptions on this paper are that (a) some capabilities are relatively scarce, and (b) the relatively scarce capabilities are distributed asymmetrically across countries. (We have chosen to take (a) and (b) as given, and explore their consequences.) It might seem natural to endogenize the entry process, and so derive (a) and (b) from more primitive assumptions. We have chosen not to do this for the following reasons. Endogenizing (a) is straightforward, and has been done elsewhere (Sutton, 1991, 1998, 2007b). To model (b), however, requires that some assumption be made regarding asymmetries between countries. This could be done by assuming that the (unobservable) relationship between the fixed and sunk costs of product development, and product quality, differ across countries; but to do this would simply beg the question, why? This leads, then, into the broad economic history of industrial development, and so to the issues that lie far beyond our present scope; no single way of modelling the origin of these cross-country differences could hope to command general acceptance. And so we have chosen to present the theory in its simplest form, staying close to the empirical observables, by taking (a) and (b) as our primitives. On-Line Appendix E. Detailed Proofs E.1 Details of Proof That vkL < vkH < u 2 √ √ Simplifying N√ k −1 − 1 + √Nk −1 − 5 √ √equation (28) using 2 Nk−1 − 1 Nk−1 − 5 = 2 Nk−1 − 3 + Nk−1 − 1 Nk−1 − 5 , = ( Nk−1 − 1) + ( Nk−1 − 5) + p p vkH N = k ( Nk−1 − 3) + Nk−1 − 1 Nk−1 − 5 . u 2 √ √ √ Define z ≡ vk /u. Re-write equation (27) as ( Nk + z) = Nk−1 − 1 Nk z or ( Nk + z)2 = Nk ( Nk−1 − 1)z. Cross-multiplying this equation, it can be further re-written as LHS(z) ≡ 1 Nk z = ≡ RHS. ( Nk + z)2 Nk−1 − 1 vi On-Line Appendix – Not for Publication LHS(z) is increasing in z for z ∈ [0,1] and LHS(0) = 0 < RHS. I will show that LHS(1) > RHS, which will establish that there is only one solution on the unit interval to this equation. LHS(1) − RHS = 1 ( Nk + 1)2 ( Nk−1 Nk ( Nk−1 − 1) − ( Nk + 1)2 Nk ( Nk + 3) − ( Nk + 1)2 Nk2 + 3Nk − Nk2 − 1 − 2Nk − 1) 1 2 ( Nk + 1) ( Nk−1 − 1) 1 = 2 ( Nk + 1) ( Nk−1 − 1) 1 = 2 ( Nk + 1) ( Nk−1 − 1) >0 ≥ [ Nk − 1] where the inequality follows from our scarcity assumption: Nk−1 − 1 ≥ ( Nk + 4) − 1. Hence vkH /u = 2 p Nk p Nk−1 − 1 − Nk−1 − 5 < 1. 4 (41) We complete the proof by showing that vkL < vkH . From equations (25) and (41), vkL < vkH if and only if 2 p Nk p ( Nk−1 − 1) Nk < N − 1 − N − 5 k − 1 k − 1 4 N2 √ k −1 p 1 p Nk−1 − 1 < Nk−1 − 1 − Nk−1 − 5 Nk−1 2 p p p 2 Nk−1 − 1 < Nk−1 Nk−1 − 1 − Nk−1 − 5 p p ( Nk−1 − 2) Nk−1 − 1 > Nk−1 − 5 which Nk−1 ≥ √ must hold because √ our scarcity√assumption implies √ √ Nk + 4 ≥ 6 so that ( Nk−1 − 2) Nk−1 − 1 ≥ (6 − 2) Nk−1 − 1 > 4 Nk−1 − 1 > 4 Nk−1 − 5 > Nk−1 − 5. E.2 Details of Proof That q vkH /u < ξ + From equations (28) and (31), vkH /u < ξ + if √ h √ i p p Nk p Nk p Nk−1 − 1 − Nk−1 − 5 < Nk−1 − 1 + 2Nk−1 − 1 . 2 Nk−1 √ √ Multiplying both sides by Nk−1 − 1 + Nk−1 − 5 and simplifying: h i p p Nk−1 [( Nk−1 − 1) − ( Nk−1 − 5)] < 2 ( Nk−1 − 1) + Nk−1 − 1 2Nk−1 − 1 hp ip p +2 Nk−1 − 1 + 2Nk−1 − 1 Nk−1 − 5 p p 4Nk−1 < 2( Nk−1 − 1) + 2 Nk−1 − 1 2Nk−1 − 1 hp ip p +2 Nk−1 − 1 + 2Nk−1 − 1 Nk−1 − 5 p p ( Nk−1 + 1) < Nk−1 − 1 2Nk−1 − 1 hp ip p + Nk−1 − 1 + 2Nk−1 − 1 Nk−1 − 5 vii On-Line Appendix – Not for Publication But 2Nk−1 − 1 > Nk−1 − 1 so that the first term on the RHS exceeds Nk−1 − 1. Further, since √our scarcity assumption implies that Nk−1 ≥ 6, the second term on the right hand side exceeds 5 + √ 1/2 < ξ +. 11 > 5. Hence the RHS exceeds Nk−1 − 1 + 5 = Nk−1 + 4. This establishes that vkH /u viii