Sources of Export Growth in Developing Countries

Transcription

1 Preliminary Not for distribution Sources of Export Growth in Developing Countries Gordon Hanson UC San Diego and NBER February 2010

2 Abstract In this paper, we examine the contribution of changes in export competitiveness and importer GDP growth to growth in exports for developing countries. Using the Eaton and Kortum (2002) model of trade as a basis for specifying a gravity model, we estimate bilateral export growth by sector as a function of exporter fixed effects, which capture changes in national export capability, importer fixed effects, which capture changes in national demand conditions, and changes in trade costs. Estimates of exporter fixed effects reveal the contribution of changes in supply conditions to export growth and how these conditions vary across exporting countries. In the 2000s, developing countries experienced pronounced improvements in export competitiveness relative to developed economies, with the strongest performance occurring in middle income nations in East Asia and the Pacific. Europe and Central Asia and Latin America and the Caribbean also performed well, though Sub-Saharan Africa suffered declines in competitiveness across the board. Changes in export competitiveness can account for over 10 percent of variation in bilateral export growth over the 2000 to 2007 period. GDP growth in exporting countries explains a small fraction of changes in export competitiveness, suggesting that changes in export capabilities are not simply a byproduct of overall economic expansion. As a second exercise, we replace importer fixed effects in the gravity model with importer GDP broken down into two components, one associated with trend GDP and a second associated with cyclical variation in GDP, where we use the Hodrick Prescott filter to decompose GDP. In the BRICs, income growth in the 2000s was associated almost entirely with the GDP trend. Other middle income countries have a more varied experience, with Asian nations (Indonesia, Malaysia, Thailand) showing high trend growth and dampened volatility (outside of the Asian financial crisis in ) and Latin American nations (Argentina, Mexico, Venezuela) showing more modest trend growth and higher volatility. The GDP trend in large high income nations is flat in comparison to middle income countries. Estimating a gravity model of trade using trend and cyclical GDP as separate regressors allows one to gauge how much of growth in trade is associated with trend growth in expenditure versus its cyclical variation. While growth in bilateral trade is strongly positively correlated with changes in trend GDP it is weakly correlated with changes in the cyclical component of GDP, suggesting that the recent growth in trade derives from structural economic growth in importing nations. If GDP trends in middle income countries persist, there would appear to be a strong basis for developing nations to be a significant source of import demand growth in the coming decade. This possibility is especially important for growth in trade, given the likelihood that the European Union, Japan, and the United States may have anemic growth in the medium run.

3 1 Introduction The world economy is beginning to recover from great recession of In emerging economies, industrial production, after falling sharply in the second half of 2008, began to grow in early 2009, at which point world trade also began to rebound. The financial crisis, though crippling in its effects on rich countries, did not infect developing economies outside of Eastern Europe (de la Torre, 2010). The exposure of emerging economies to the recession has been limited largely to a contraction in the global demand for their exports. With that contraction easing, world trade is likely to continue to expand in Somewhat lost in the mayhem surrounding the recent volatility in the global economy has been an appreciation for the profound changes in trade patterns that occurred over the last decade. Most notably, developing countries have become a prime source of growth in demand for exports from other developing countries. The BRICs (Brazil, Russia, India, China) have attracted much attention, owing to their large size and rapid rates of recent economic growth, but they are by no means the only significant buyers of developing countries products on the global market. Between 2000 and 2008, the share of exports sent to developing economies (countries designated by the World Bank as low or middle income) rose from 28% to 41% for low income nations, 23% to 41% for lower middle income nations, and 18% to 31% for upper middle income nations (Hanson, 2010). The BRICs accounted for roughly half of this expansion, with other middle income countries accounting for the rest. The flip side of developing regions accounting for a larger share of global imports is that high income countries have become relatively less important destinations for developing country goods. The proximate cause for the change in market share is a shift in the locus of economic growth from high income to middle income 1

4 nations. Because middle income countries are growing so much more rapidly than high income economies, they account for an ever larger source of global demand. In this paper, we examine whether recent changes in global trade patterns are likely to be sustained as the global economy recovers. Naturally, a pro-longed collapse in economic growth in the BRICs or other middle economies would profoundly affect global trade. But absent such an interruption can we expect growth in import absorption by middle income nations to continue? The answer depends in part on whether recent growth in middle income economies is robust; that is, whether it represents long run growth based on growth in TFP, capital investment, and skill upgrading or simply a temporary deviation from a more modest trend. The answer also depends on whether export growth in developing nations is being matched by improvements in their export capacity, allowing the expansion to be sustained. We apply the gravity model of trade to evaluate both issues. Using the Eaton and Kortum (2002) model of trade as a basis for specifying a gravity model, we estimate bilateral export growth by sector as a function of exporter fixed effects, which capture changes in national export capability, importer fixed effects, which capture changes in national demand conditions, and changes in trade costs. Estimates of exporter fixed effects reveal the contribution of changes in supply conditions to export growth and how these conditions vary across exporting countries. In the 2000s, developing countries experienced pronounced improvements in export competitiveness relative to developed economies, with the strongest performance occurring in middle income nations in East Asia and the Pacific. Europe and Central Asia and Latin America and the Caribbean also performed well, though Sub-Saharan Africa suffered declines in competitiveness across the board. Changes in export competitiveness can account for over 10 percent of variation in bilateral export growth over the 2000 to

5 period. GDP growth in exporting countries explains a small fraction of changes in export competitiveness, suggesting that changes in export capabilities are not simply a byproduct of overall economic expansion. As a second exercise, we replace importer fixed effects in the gravity model with importer GDP broken down into two components, one associated with trend GDP and a second associated with cyclical variation in GDP, where we use the Hodrick Prescott filter to decompose GDP (based on annual data from the WDI over the period 1960 to 2008). In the BRICs, income growth in the 2000s was associated almost entirely with the GDP trend. Other middle income countries have a more varied experience, with Asian nations (e.g., Indonesia, Malaysia, Thailand) showing high trend growth and dampened volatility (outside of the Asian financial crisis period in ) and Latin American nations (e.g., Argentina, Mexico, Venezuela) showing more modest trend growth and higher volatility. The GDP trend in large high income nations is flat in comparison to middle income countries (and to smaller high income nations, such as Australia, Canada, and Spain). Estimating a gravity model of trade using trend and cyclical GDP as separate regressors allows one to gauge how much of growth in trade is associated with trend growth in expenditure versus its cyclical variation. While growth in bilateral trade is strongly positively correlated with changes in trend GDP it is weakly correlated with changes in the cyclical component of GDP, suggesting that the recent growth in trade derives from structural economic growth in importing nations. If GDP trends in middle income countries persist, there is a strong basis for developing nations to be a significant source of import demand growth in the coming decade. This possibility is especially important for growth in trade, given the likelihood that the European Union, Japan, and the United States may have anemic growth in the medium run. 3

6 2 Framework for analysis To examine export growth in developing countries, we adopt the Eaton and Kortum (2002) model of trade. In their framework, which builds on the Dornbusch, Fischer, and Samuelson (1977) Ricardian model of trade, each sector consists of a large number of products over which consumers have CES preferences. Ignoring sector subscripts, country i s comparative advantage in a sector is determined by a factor, T i >0, which is common to all products in a sector and which captures the country s absolute productive capability as determined by its factor endowments, institutions, technology, etc. There is also a random component to productivity, which allows for the possibility that even in countries that have an absolute disadvantage in a sector (i.e., a low T) some firms, through innovation or other efforts, may succeed in achieving high productivity, allowing them to serve markets at home and abroad. The most productive producer in country i has productivity z i (j), such that the marginal cost of a good produced in country i delivered to country n is c i d ni /z i (j), where c i is the cost of one unit of inputs, 1/z i (j) is the number of input units needed to produce one unit of output, and d ni is an iceberg transport cost between countries. In each importing country n, consumers search across all potential manufacturers of a given product and choose the one that sells at the lowest price, a process they repeat for all products in the sector. With productivity varying across products within a sector, countries are likely to have positive exports in all sectors, although their exports may be very small in sectors in which they have an absolute disadvantage. To use this setup to derive a gravity model of trade, we need to specify a distribution for the productivity term z i (j) and impose a market structure on each sector. Eaton and Kortum adopt a Fréchet distribution for productivity (from the family of generalized extreme value distributions), which captures the fact that most potential producers in a sector are inefficient and 4

7 only a relatively small number succeed in achieving high productivity. With the Fréchet, the cdf for z i (j) (the probability that a producer from country i has productivity less than or equal to some level z) is exp(-t i z -θ ), where the parameter θ>1 captures the dispersion in productivity (a higher θ indicates less dispersion). The higher the dispersion in productivity, the more likely it is that countries with a weak absolute advantage succeed in exporting some products. Under the assumption of perfect competition, setting price equal to marginal cost implies that the share of products country n purchases from country i in a sector is XX nnnn = TT ii(cc ii dd nnnn ) θθ, (1) XX nn Φ n where X ni is purchases by country n from country i, X n is total purchases by country n, and Φ n = TT h (cc h dd nnh ) θθ h describes country n s price distribution (which reflects marginal production and distribution costs in countries that supply products to country n). Equation (1) is the basis for the gravity model of trade in the Eaton and Kortum framework. A country s exports to a particular destination are increasing in the exporter s technological capability (T i ) and the importing country s spending (X n ) and decreasing in exporter production costs (c i ), bilateral transport costs (d ni ), and the competitiveness of the importer s market (Φ n ). The importance of production costs and trade costs for trade is higher the lower is dispersion in productivity (the higher is θ), as with low dispersion it is unlikely individual producers will draw a high enough productivity to compensate for high production or trade costs. While the setup of the Eaton and Kortum model may seem restrictive (in terms of imposing perfect competition and a particular distribution on productivity), the framework is quite flexible. It can easily be generalized to allow for imperfect competition (Eaton, Kortum, and Kramarz, 2008) or more general distributions of productivity (Costinot and Komunjer, 2009). 5

8 Taking logs and expressing the gravity equation in first differences, the growth in bilateral exports becomes, ΔllllXX nnnn = Δllll TT ii cc θθ ii + ΔllllXX nn + ΔllllΦ n θθθθθθθθdd nnnn. (2) The first term on the right of equation (2) is the change in country i s competitiveness, due either to changes in its technology capability or production costs; the second term on the right is the change in country n s spending on the sector; the third term on the right is the change in the competitiveness of country n s market; and the final term is the change in bilateral trade costs. To take equation (2) to the data, we group country specific terms and write, ΔllllXX nnnn = ΔS ii + ΔSS nn θθθθθθθθdd nnnn, (3) where ΔS ii is a fixed effect for exporter i, capturing the change in its competitiveness, and ΔSS nn is a fixed effect for importer n, capturing the change in demand conditions in its market. Using equation (3) as a basis for the empirical estimation, we proceed to estimate exporter fixed effects, evaluating how they have changed over time for countries in different regions and income levels, and how important they are in explaining export growth. In a subsequent exercise, we replace importer fixed effects with the log change in importer GDP, divided into one component that represents changes in trend GDP and a second component that represents changes in cyclical GDP. This second regression is useful for gauging the differential contribution of long run versus cyclical changes in GDP to export growth. 6

9 3 Data and preliminary analysis 3.1 Data The data we use for the analysis are from the Comtrade database. Trade data are available at a high level of disaggregation. Yet, for our purposes such disaggregation is problematic. The gravity expression in equation (1) is based on the law of large numbers, in that we implicitly assume there are a sufficient number of producers and consumers in each country such that there will be positive trade for all bilateral country pairs in each sector, though some trade flows may be very small (owing to the fact that in countries that have a weak absolute advantage in a sector it is unlikely that many producers will succeed in achieving high enough productivity to export). In practice, the combination of small samples (i.e., small country sizes) and low probabilities of trade produces many zero bilateral trade flows. In the Eaton and Kortum framework, such zeros are not economically interesting; they are simply a byproduct of finite samples (though in other theoretical frameworks zeros have greater importance). To get around zero trade, we aggregate four digit HS products into eight sectors (defined in terms of aggregates over two digit HS sectors): (1) Agriculture, meat and dairy, seafood (HS 1-10, 12-14) (2) Food, beverages, tobacco, wood, paper (HS 11, 15-24, 44-48) (3) Extractive industries (HS 25-27, 68-71) (4) Chemicals, plastics, rubber (HS 28-36, 38-40) (5) Textiles, apparel, leather, footwear (HS 41-42, 50-65) (6) Iron, steel, and other metals (HS 26, 72-83) (7) Machinery, electronics, transportation equipment (HS 84-89) (8) Other industries (HS 37, 43, 49, 66-67, 90-97) 7

10 Each sector is a collection of industries that share similar factor intensities and are likely to rely on similar technological or institutional foundations as a basis for production. The first sector includes land intensive activities surrounding agriculture production; the second sector includes manufacturing activities that use agriculture, forestry, and other land intensive inputs; the third sector includes nonmetallic minerals, ores, petroleum, precious stones, precious metals, and other industries based on sub-soil resources; the fourth sector involves the manufacture of chemicals and other petroleum-based products; the fifth sector involves the production of labor intensive clothing and apparel items and the inputs for these goods (textiles and leather); the sixth sector involves the production of iron, steel, and other metals; the seventh sector involves the production of skill and capital-intensive machinery, electrical materials, electronics, and transport equipment; and the eighth sector is a collection of remaining manufacturing industries (photographic material and equipment, fur, printed material, umbrellas, hats, musical instruments, arms, furniture, toys, miscellaneous items). This aggregation scheme is similar to that utilized by Harrigan (1997) and Romalis (2004) for SIC industries. Aggregation helps remove zero trade values from the data but does not solve the problem entirely. To further remove zero trade associated with small samples, we drop from the sample very small exporters (countries with a population in 2000 of less than 1 million inhabitants) and small importers, limiting importers to the 56 countries that have a share of world GDP of at least 0.1% in These countries account collectively for 98% of global GDP in The list of importers is presented in an appendix. Included are two low income countries (Bangladesh, Vietnam), the BRICs, 25 other middle income countries (Algeria, Argentina, Chile, Colombia, 1 Missing in the WDI GDP data, and therefore from the sample of importers, are Taiwan and several Gulf States (Bahrain, Kuwait, Oman, Qatar, and the United Arab Emirates). Other nations with missing GDP data are war torn countries (Afghanistan, Iraq, Somalia, Palestinian Territories), tightly controlled economies (Cuba, Myanmar, North Korea, Zimbabwe), and very small countries (most of which are islands). 8

11 Czech Republic, Egypt, Hungary, Indonesia, Iran, Korea, Libya, Malaysia, Mexico, Morocco, Nigeria, Pakistan, Peru, Philippines, Poland, Romania, South Africa, Thailand, Turkey, Ukraine, Venezuela), four non-oecd high income countries (Hong Kong, Israel, Singapore, Saudi Arabia), and 21 OECD high income countries (Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, UK, US). A second issue in constructing the data set involves the choice of years to include in the sample. To capture peak to peak changes in trade, it is desirable to stay as close as possible to global business cycles. A natural time span to examine would be 2000 to 2008, as this roughly corresponds to the peaks of the economic expansions from 1991 to 2000 and from 2001 to While both years are available in Comtrade, there is a problem with the number of countries covered in the data in The number of importers included in the sample is 140 in 2000, 155 in 2002, 160 in 2004, 160 in 2006, and 132 in The drop off in 2008 does not reflect a contraction in trade but rather a reduction in the number of countries reporting their imports. To account for the drop off in countries in 2008, we use 2007 instead (for which there are 154 importers in the data). Of the 56 importers, seven do not report imports in one of the two years, 2000 or We drop these countries from the sample. 2 The resulting sample has over 82,000 observations on bilateral trade in the eight sectors in the two years. Even with the sample restrictions, there are 3,000 bilateral trade partners that exit the sample (show positive trade in 2000 but not 2007) and 6,900 that enter the sample (show positive trade in 2007 but not 2000). To estimate changes in exporter and importer fixed effects, we are forced to drop these 2 These countries are Bangladesh, Egypt, Iran, Libya, Morocco, Pakistan, and Ukraine, whose collective share of world GDP in 2000 is 1%. 9

12 observations from the sample, leaving a sample size of slightly more than 36,000 changes in bilateral trade flows at the sector level. 3.2 Preliminary analysis To provide a sense of trade patterns in the data, Table 1 shows export shares by sector for 2000 and 2007 where developing countries are aggregated into three groups: low income countries, lower middle income countries, and upper middle income countries (following World Bank country categories for 2008). Low income countries are heavily specialized in agriculture (13% of exports in 2007), extractive industries (24% of exports in 2007), and apparel, footwear, and textiles (34% of exports in 2007). As discussed in Hanson (2010), during the 2000s low income countries became more diversified in terms of their exports, shifting into sectors in which they had relatively low export participation in the past, including iron and steel (metals) and machinery, electronics, and transport equipment. Relative to low income countries, lower middle income countries are more diversified, with machinery, electronics and transportation equipment being the most important sector for the group (37% of exports in 2007) and extractive industries being the second most important (18% of exports in 2007). One notable shift for lower middle income countries is the decreasing importance of apparel and textiles, whose share of exports falls from 22% to 14% from 2000 to 2007, with much of this change being attributable largely to the decreasing relative importance of apparel in China s exports. Upper middle income countries exhibit similar diversification patterns as lower middle income countries, with the primary difference being that extractive industries (36% of exports in 2007) are more important than machinery and electronics (26% of 10

13 exports in 2007), due to the preponderance of oil exporters in the Middle East and North Africa in this income group. Table 2 gives the same information as Table 1, with country income groups broken down by six geographic regions. Within income groups, there is strong variation in export patterns across regions. While apparel is important for low income countries overall, much of this is due to East Asia and the Pacific and South Asia, where in 2007 apparel accounts for 37% and 89% of exports, respectively. In the Middle East and North Africa and Sub-Saharan Africa, in contrast, apparel is not a major export sector, with these regions relying more heavily on agriculture, extractive industries, and metals. Among lower middle income countries, the shift into machinery and electronics is strongest in East Asia and the Pacific, where the sector accounts for 46% of exports in Extractive industries are important for Sub-Saharan Africa, the Middle East and North Africa, Europe and Central Asia, and Latin America and the Caribbean. Finally, for upper middle income countries machinery and electronics are a major export industry for East Asia and the Pacific and Latin America and the Caribbean, while extractive industries are the major source of exports for the Middle East and North Africa, Europe and Central Asia, and Sub-Saharan Africa. To characterize the destinations for developing country exports, Table 3 shows export shares by destination market for developing country income groups. It should be emphasized that these figures understate the importance of low and middle income destinations, owing to the sample restrictions placed on the data (e.g., the exclusion of countries lacking import data in both 2000 and 2007, which drops seven middle income nations from the sample of importers, and the exclusion of importing countries lacking GDP data, which drops Taiwan and five Gulf States from the sample of importers). For all income groups, the share of exports destined for low and 11

14 middle income countries rises, with the most pronounced growth occurring in lower middle income exporters. Other developing countries are an increasingly important source of export growth for low and middle income economies. Table 4 provides similar detail as Table 3, with developing countries broken out by six geographic regions. Finally, we describe the decomposition of GDP we employ to separate changes in GDP into growth associated with the trend and growth associated with cyclical changes deriving from business cycles. The decomposition we use is due to Hodrick and Prescott (1997). While their decomposition is ad hoc, it has proved to be very useful in the literature and is widely applied. For a given smoothing parameter, λ, the HP filter removes a smooth trend from data, which allows one to identify both a variable trend (which captures long run growth) and deviations from the trend (which captures business cycle fluctuations). Let y t be the log of GDP in year t, the HP filter then estimates a trend series, τ t, based on the solution to, min ττtt TT (yy tt ττ tt ) 2 λλ (ττ tt+1 ττ tt ) (ττ tt ττ tt 1 ) 2 tt=1 (4) where cc tt = (yy tt ττ tt ) is the cyclical component of GDP. For annual data, Ravn and Uhlig (2002) recommend using a smoothing parameter of 6.25, though other literature has used values ranging from 5 to 15. We perform the analysis with λ=6.25 and also use λ=10 (which produces a trend that smooths the data more heavily and leaves less cyclical variation) to check robustness of results. In the gravity estimation, we will use the change in GDP, and not its level, as a regressor, which is decomposed as yy tt = ττ tt + cc tt, (5) where the first term on the right is GDP growth associated with the GDP trend, which we refer to as trend growth in GDP, and the second term is GDP growth associated with changes in cyclical conditions, which we refer to as the cyclical variation in GDP. 12

15 Figure 1 plots the trend and cyclical components of GDP for several groups of countries. Each plot shows log GDP (left scale), the trend component of log GDP (left scale), and the cyclical deviation from trend (right scale). The series are estimated on annual data for each country covering 1960 to For ease of exposition, we limit the plots in the figures to the period 1980 to We begin with the BRICs (Figure 1.a). Evident for China and India is strong trend growth over the last three decades. After a recession in the early 1990s, there is little cyclical variation in China s GDP. The cyclical component of India s GDP is similarly modest. For both countries, most variation in GDP over time comes from the long run growth trend. In Brazil, trend growth accelerates over time, while cyclical variation dampens over time. Russia shows a sharp break in its trend growth in the late 1990s, after which point cyclical variation is reduced. Thus, for all four BRICs the 2000s were a time when most GDP growth appears to be associated with trend GDP rather than cyclical variations. Moving to other middle income nations, we see more cyclical variation in GDP and sharp differences in trend GDP growth. Korea (Figure 1.b) has a strong positive GDP trend that weakens somewhat in the 2000s. Trend growth in Mexico, Poland and Turkey accelerates after In all four countries, cyclical movements in GDP are more pronounced than in the BRICs. Trend growth is modest in Argentina and South Africa (Figure 1.c), with the former showing large cyclical variation and the latter showing relatively little. Indonesia and Thailand show large cyclical variation in GDP around the Asian financial crisis in , accompanied by a dampening of trend growth, which later accelerates in the early 2000s. Malaysia (Figure 1.d) resembles other Asian countries with volatility in the late 1990s and an uptick in trend growth in the 2000s. Egypt and Pakistan exhibit stable trend growth, with only modest cyclical variation. 13

16 Venezuela follows the Latin American pattern of strong cyclical variability and several breaks in trend growth. Turning to high income countries, trend growth is unsurprisingly much slower in the large economies of Germany, the UK, and the US (Figure 1.e). Japan shows a marked slowing in its trend growth in the early 1990s, consistent with the beginning of the country s economic stagnation. The magnitude of business cycle variations is similar in the four countries. While growth in France (Figure 1.f) resembles that in the large high income countries, Australia, Canada, and Spain exhibit more rapid trend growth in GDP. Business cycle variation in these countries is similar to the other high income group. All in all, middle and high income countries show considerable variation in the pace and stability of trend growth in GDP and in the contribution of the cyclical component to GDP changes. From these figures it is not immediately apparent which component of GDP growth, the trend or the cycle, is most important for growth in import demand. We leave the issue to the empirical estimation to resolve. 14

17 4 Empirical Results For the empirical analysis, we use the gravity model in equation (3) as a basis for estimation. In a first exercise, we regress the change in bilateral exports on exporter country dummy variables (to absorb exporter fixed effects), importer country dummy variables (to absorb importer fixed effects), and standard gravity variables (bilateral distance, sharing a border, common language, colonial history, RTA membership). Since the gravity variables are constant over time (except for RTA membership), their coefficient estimates capture changes in the importance of these variables for trade costs. The regressions are estimated separately for the eight aggregate sectors defined in section 3, such that we allow exporter and importer fixed effects to vary by broad product group. We then evaluate the exporter fixed effects for evidence of how countries export capabilities are evolving over time on a sectoral basis. In a second exercise, we estimate a modified version of equation (3) in which we drop the importer country dummies and replace these with trend growth in GDP (the change in the HP trend component), the change in the HP cyclical component of GDP, a dummy variable for whether the importer belongs to the WTO as of 1999 (which captures how WTO membership affects changes in import barriers), and a dummy variable for whether the importer joined the WTO between 2000 and 2007 (which captures how WTO ascension affects changes in trade costs). Data on WTO membership are from We use this specification to evaluate the contribution of difference components of GDP growth to trade growth. 4.1 Changes in export capabilities Table 5 reports coefficient estimates for the first regression exercise. We suppress coefficient estimates for the exporter and importer dummies, but describe these in more detail 15

18 below. From equation (2), the exporter dummies capture the sum of the change in exporter absolute advantage (ΔllllTT ii ) and the change in exporter production costs ( θθδllllcc ii ), which one can think of as describing the relative change in export competitiveness for a country in a sector. The importer dummies capture the sum of the change in importer expenditure in a sector (ΔllllXX nn ) and the change in the distribution of prices in the importer s market (-ΔllllΦ n ), which one can think of as describing the change in effective demand by an importer in a sector. To improve the efficiency of the estimation, we limit the sample to exporters that export to at least 10 importers in each sector, a restriction which drops eight small exporters from the sample. The reported parameter estimates in Table 5 show how changes in trade are associated with changes in the impact of gravity variables (i.e., the estimated coefficient is the change in the coefficient on a gravity variable from a conventional gravity regression run in levels). Data on gravity variables are from and on regional trade agreements are from The effect of distance on trade is positive and precisely estimated in two sectors, apparel and other industries (which is primarily instruments and specialized equipment), indicating that in these sectors distance became a less important impediment to trade over the period. Having a common language has a positive and precisely estimated coefficient in three sectors, agriculture, metals, and machinery and electronics, indicating that in these cases language became a more important impediment to trade over the period. And sharing a colonial relationship has a negative and precisely estimated coefficient in four cases, agriculture, chemicals, apparel, and metals, indicating colonial heritage became a more important impediment to trade in these sectors over the period. Having an existing regional trade agreement (RTA) or a new agreement post 2000 (new RTA) are uncorrelated with trade growth. 16

19 The exporter fixed effects absorb the determinants of a country s average competitiveness in a sector, holding importer characteristics constant. How important is the change in exporter capabilities to the growth of trade? To gain insight into this question, we run an additional regression to that reported in Table 5, in which we exclude exporter dummy variables but still include import dummies and the gravity variables. The difference in the adjusted R squared for these regressions and those in Table 5 gives a sense of how important changes in exporter competitiveness are for explaining trade growth. Across the eight sectors, the inclusion of exporter dummies raises the adjusted R squared by an average of 0.11, with the largest increase occurring in apparel and textiles (0.16) and the smallest in the metals industry (0.08). Thus, changes in relative exporter capabilities, be they associated with changes in technology or in production costs, can account for approximately 11% of the variation in bilateral export growth over 2000 to While the value of the estimated exporter fixed effects are not comparable across sectors (owing to their being subject to an arbitrary normalization), they are comparable across exporters within a sector. Table 6 shows mean exporter fixed effects by sector and exporter income group (where fixed effects are weighted by 2000 exports). By construction, exporter fixed effects have mean zero across all countries within a sector. The value of the fixed effect for an individual country in an individual sector shows its mean growth in sectoral exports relative to other exporters, holding importing country conditions and trade costs constant. We interpret a positive exporter fixed effect as an indication that a country enjoyed an increase in its export competitiveness relative to other countries (it is because the global baseline varies by sector that fixed effects for a given exporter are not comparable across sectors). 17

20 Over 2000 to 2007, low income countries have seen improving competitiveness in agriculture, apparel and textiles, metals, and other industries, and declining competitiveness in food processing and extractive industries (with little relative change in the other industries). Lower middle income countries have seen improving competitiveness in all sectors except extractive industries and show the strongest relative improvement among all income groups in five of the eight sectors (food processing, apparel and textiles, chemicals, metals, and machinery and electronics). Upper middle income countries have seen improvements in their competitiveness in agriculture, food processing, extractive industries, apparel and textiles, and other industries. If developing nations are seeing improvements (or no change) in their competitiveness in most sectors, then high income countries must be seeing a decline, as is the case for high income OECD countries in all sectors and for high income non-oecd countries in all sectors but one (chemicals). The 2000s, then, were a time when developing economies, and lower middle income nations in particular, enjoyed substantial improvements in their export competitiveness relative to high income nations. One may think that these rankings are influenced by the presence of the BRICs among middle income nations. However, when we exclude the BRICs from their respective income categories, as seen in the second panel of Table 6, the resulting mean exporter fixed effects are largely unchanged. This indicates that other middle income nations have enjoyed improvements in export competitiveness similar to that experienced by the BRICs. To provide regional context for changes in export competitiveness, Table 7 reports mean sectoral exporter fixed effects by geographic region, where low and middle income nations are grouped together within each region and high income nations across all regions are grouped into a single category. East Asia and the Pacific shows improvements in export competitiveness in 18

21 all sectors (with or without China included). Europe and Central Asia, Latin American and the Caribbean, and South Asia show improvements in export competitiveness in all sectors but one (metals, chemicals, and agriculture, respectively). The Middle East and North Africa shows improvements in export competitiveness in three sectors, agriculture, apparel and textiles, and machinery and electronics. And Sub-Saharan Africa, in contrast to other developing regions, shows improvements in export competitiveness in no sector. Among geographic regions, there is clearly sharp variation in export outcomes, with East Asia showing the first or second largest improvement in export competitiveness in six of the eight sectors and Sub-Saharan Africa showing the largest decline in six of the eight sectors. It is also instructive to examine outcomes in export competitiveness for individual countries. Figure 2 shows the change in export competitiveness by sector for 24 large economies, 8 of which are high income and 16 of which are middle income. The performance of the BRICs is impressive. China is one of the top five countries in terms in improving export competitiveness in all eight sectors, India is in the top five in seven sectors, Brazil is in the top five in two sectors, and Russia is in the top five in one sector. Other middle income nations have also shown strong growth in export competitiveness, with Turkey in the top five in six sectors, Poland in the top five in five sectors, and Argentina in the top five in four sectors. (Changes in export competitiveness could in part reflect unmeasured changes in trade barriers in destination countries (e.g., for Poland and Turkey) or the effects of currency devaluation during the 2000 to 2007 period (e.g., in Argentina)). High income nations, in contrast, show up as have weak improvements in export competitiveness. Japan and the United Kingdom are among the bottom five performers in six of the eight sectors and the United States is among the bottom five 19

22 performers in four sectors. Venezuela is the sole middle income national with consistent poor performance, showing up in the bottom five countries in seven of the eight sectors. In sum, the 2000s were a period in which developing countries enjoyed significant improvements in their export competitiveness, with East Asia and the Pacific exhibiting the strongest performance and Sub-Saharan Africa exhibiting the weakest. A natural question is whether changes in export competitiveness are simply a reflection of differences in GDP growth across countries. That countries vary in their sectoral rankings of improvements in export competitiveness is initial evidence this is not the case. To examine the question more formally, we run additional regressions of exporter fixed effects on the log change in exporter GDP. Table 8 shows these results (which are weighted by the number of countries to which an exporter exports). Unsurprisingly, the coefficient on GDP is positive in all regressions and precisely estimated in six of the eight sectors. However, the adjusted R squared in these regressions is over 0.09 in just one sector, extractive industries, and averages 0.05 across the eight sectors. The small explained variance suggests that GDP growth plays only a modest role in changes in countries export competitiveness. Other factors are at work, presumably related to technological advancement and savings in production costs. A final issue worth mentioning is how one interprets the exporter fixed effects. The interpretation we have given follows from equation (3), which relates export competitiveness to technological capability and production costs. However, production costs may be related to international trade. Countries that are more integrated into global production networks may have lower production costs, which would show up in terms of a higher exporter fixed effect (or in time differences in larger increases in exporter fixed effects). Eaton and Kortum (2002) demonstrate this relationship formally. Thus, nothing we have presented rules out an important 20

23 role for global production networks in changes in countries export competitiveness. Given the concentration of these networks in just two of the eight sectors (apparel/textiles and machinery/electronics/transportation), cross border trade in inputs is unlikely to be an important factor in changes in export competitiveness in the other sectors (agriculture, food processing, extractive industries, chemicals, and metals). 4.2 Trend versus cyclical variation in GDP and import demand We turn next to the contributions of different components of GDP growth to growth in demand for imports. We modify the gravity equation in (2) as follows: ΔllllXX nnnn = ΔS ii + Δττ nn + Δcc nn θθθθθθθθdd nnnn. (6) As in the previous section, we include exporter fixed effects and gravity variables as regressors. However, we replace importer fixed effects with the log change in importer GDP decomposed into the change in the HP trend (Δττ nn ) and the change in the HP cyclical component (Δcc nn ). With importer dummies left out of the regression, we can include measures of importer specific trade costs. We also include a dummy variable for whether the importer was a member of the WTO in 1995 or whether it joined the WTO in 2000 or later (which applies to just three importers in the sample: China, Saudi Arabia, and Vietnam). The regression results are presented in Table 9. In the first panel, we include the change in log importer GDP as a regressor. Consistent with abundant previous research on the gravity model, it enters positively and significantly in all sectors, with a coefficient ranging in value from 0.9 (chemicals) to 2.6 (extractive industries). (The coefficient estimates on the gravity variables change in value relative to Table 5. The differences in coefficients are difficult to interpret, owing to the absence of importer fixed effects in Table 9 and their presence in Table 21

24 5.) In panel (b), we split the change in log GDP into the change in the HP trend and the change in the HP cyclical component, based on a smoothing parameter of The change in the HP trend component captures long run economic growth; the change in the HP cyclical component captures changes in business cycle conditions. The HP trend component enters positively and significantly in all sectors. Growth in trend GDP is strongly associated with growth in import demand. The HP cyclical component, in contrast, has coefficients that range all over the map, being negative in two sectors and imprecisely estimated for seven of eight sectors. The cyclical component is positive and significant only for the metals industry. Business cycle variation in GDP appears to have a weak relationship with changes in import demand. The results in panel (b) are based on a smoothing parameter in equation (4) of 6.25, following the recommendation of Ravn and Uhlig (2002). When we instead use a smoothing parameter of 10, which dampens the HP cyclical component, the results, as shown in panel (c) of Table 9, are qualitatively very similar. The coefficient on the HP trend component is again positive and strongly significant while the coefficient on the HP cyclical component is negative in three industries in insignificant in seven of the eight (with the cyclical component again positive and significant only for metals). The results in Table 9 suggest that the primary source of variation in import demand is coming from variation in trend GDP, rather than from business cycle fluctuations. In the following section, we discuss the implications of these results for future trade growth among developing countries. 22

25 5 Discussion In this paper, we examine the contribution of changes in export competitiveness and importer GDP growth to growth in exports by developing countries over the period 2000 to The analysis is based on the Eaton and Kortum (2002) model of trade, which we use to specify a gravity model in which bilateral export growth in a sector is a function of exporter fixed effects (which capture changes in national export competitiveness in a sector), importer fixed effects (which capture changes in national demand conditions in a sector), and changes in trade costs. We also use the Hodrick-Prescott (1997) filter to decompose GDP into trend and cyclical components in order to examine how each are related to changes in import demand over the sample period. The main findings are as follows: Middle and high income countries show considerable variation in the relative contributions of trend growth in GDP and cyclical variation in GDP to overall GDP growth. In the BRICs, almost all growth is related to the trend, as it is in a number of middle income countries in East Asia and the Pacific. In Latin America, the cyclical variation tends to be a more important component of growth. In high income countries, trend growth is considerably smaller than in middle income countries. Gravity estimation reveals that changes in trend GDP are strongly positively associated with trade growth, while cyclical variation in GDP positively affects trade in just one industry, metals. These results are silent on whether the flexible trend GDP growth estimated over the 1960 to 2008 period will continue. What they do show is that it is the trend component of GDP that matters for trade growth. To the extent that GDP trends in middle income nations continue, there is every reason to believe they will continue to be an important source of import demand growth. 23

26 Changes in relative exporter capabilities, as captured by exporter fixed effects in time difference gravity regressions, can account for approximately 11% of the variation in bilateral export growth over 2000 to Thus, important sources of trade growth are improvements in exporter technology and reductions in exporter production costs (part of which may be related to global production networks). East Asia and the Pacific shows improvements in export competitiveness in all sectors (with or without China included); Europe and Central Asia, Latin American and the Caribbean, and South Asia show improvements in export competitiveness in most sectors; the Middle East and North Africa shows improvements in export competitiveness in a few sectors; and Sub-Saharan Africa shows improvements in export competitiveness in no sector. Thus, while most middle income nations have enjoyed improving export competitiveness vis-à-vis developed nations, the countries of Sub-Saharan Africa have not and have actually declined. Several caveats to the analysis apply. A first relates to the sample. Absence of GDP data and concerns over zero trade in one or both years leads to the exclusion of a number of middle income and high income non-oecd countries from the sample (Taiwan, five Gulf States, several moderate sized middle income nations). These countries are likely to have become an increasingly important source of import demand, meaning that the results, in anything, understate the importance of middle income nations for import demand growth in the 2000s. A second caveat relates to the importance of global production networks. The exporter fixed effects we estimate capture changes in export competitiveness associated with technological progress and changes in production costs. One source of changes in production costs is improved access to global production networks, which often involve intensive trade in 24

27 intermediate inputs with neighboring countries. Previous literature suggests that these networks are concentrated in two of the eight sectors we examine: apparel, textiles, footwear, and leather; and machinery electronics, and transportation equipment. Our results encompass global production networks, as their presence is incorporated into exporter fixed effects, but they do not separate their effects out from the effects of country-specific technological progress. Thus, the impressive improvement in export competitiveness in East Asia and the Pacific could be due in part to the strengthening of global production networks in the region during the 2000s. An important question for future research is how much these networks can account for the improvements in export capabilities observed in developing countries. 25

28 References Costinot, Arnaud, and Ivanna Komunjer What Goods do Countries Trade? A Structural Ricardian Model. Mimeo, UCSD. de la Torre, Augusto Update on LAC and the Global Crisis: The Worst is over but What Lies behind? Mimeo, World Bank. Dornbusch, Rudiger, Stanley Fischer, and Paul Samuelson Comparative Advantage, Trade, and Payments in a Ricardian Model with a Continuum of Goods. American Economic Review, 67: Eaton, Jonathon, and Samuel Kortum Technology, Geography, and Trade. Econometrica, 70: Eaton, Jonathon, Samuel Kortum, and Francis Kramarz An Anatomy of International Trade: Evidence from French Firms. NBER Working Paper Hanson, Gordon Export Dependence in Developing Countries. Mimeo, UC San Diego. Harrigan, James Technology, Factor Supplies, and International Specialization: Estimating the Neoclassical Model. American Economic Review, 87(4): Hodrick, Robert J., and Edward C. Prescott, Postwar U.S. Business Cycles: An Empirical Investigation, Journal of Money, Credit and Banking 29(1): Ravn, Morten O., and Harald Uhlig On Adjusting the Hodrick-Prescott Filter for the Frequency of Observations. The Review of Economics and Statistics, 84(2): Schott, Peter K "Across-Product versus Within-Product Specialization in International Trade." Quarterly Journal of Economics, 119(2):