Which of the following factors impact how much trade takes place between two countries?

4.4.3 Economic Proximity: Service Trade Flows

The analysis focuses on service trade flows as opposed to general trade flows comprising goods in order to avoid potential distortions caused by raw materials, fuel, and other commodities that are unrelated to knowledge circulation. Service trade data were kindly provided by the OECD Directorate for Trade and Agriculture based on an internal working version of the OECD Trade in Services19 database used as input for the OECD/WTO TiVA database. Different categories of services trade can be distinguished according to the 2002 Extended Balance of Payments Services classification (EBOPS, 2002). We thus explore the role of trade in more knowledge-intensive services such as computer and information services and other business services with respect to scientist flows and collaboration over 1996–2011.

4.4 Financial Frictions and International Trade

Two recent papers focus on the interaction between trade flows and capital flows. Jin (2012) presents a stochastic two-country overlapping generations model with production and capital accumulation in which factor intensities are (exogenously) different across countries. The paper combines insights from the factor proportions trade literature with those of the standard neoclassical open economy growth model. In her model, there is both an intertemporal motive for capital flows and an intratemporal motive since capital will tend to flow to countries that are more specialized in capital intensive industries. Hence two competing effects determine the direction of net capital flows: the composition effect (linked to asymmetries in specialization across countries) and the standard efficiency effect (stemming from capital scarcity). A country hit by a positive productivity shock, or experiencing a relative increase in its labor force—as was the case for many emerging economies since 1990—can nevertheless become a capital exporters if it specializes in labor intensive industries. Hence specialization is the key mechanism through which Jin (2012) may account for global imbalances.

Antràs and Caballero (2009) present a model where financial frictions determine patterns of capital flows and trade flows. Countries are heterogeneous in terms of financial development and sectors differ in their degree of financial dependence. They feature a two-country (North and South, where South is financially underdeveloped), two-factor (capital and labor), two-sector general equilibrium model where a homogeneous good is internationally traded. Under trade and financial autarky, South invests disproportionately in the sector without financial frictions. This depresses wages and rental rates of capital. If capital is now allowed to move freely, but international trade in goods remains restricted, capital will flow out of the financially underdeveloped economy toward the financially developed one, as in the models presented in this section. By contrast, if international trade in goods is also liberalized, countries will specialize along the lines of comparative advantage: the financially underdeveloped South specializes (incompletely) in the sector unaffected by the financial friction. This raises the rental rate of capital in the South because of good price equalization, while domestic wages remain depressed, and this can reverse the direction of capital flows. Hence it is the difference in production structures due to the pattern of specialization induced by comparative advantage that interacts with financial liberalization to shape the direction of net capital flows. The pattern of specialization is thus endogenously determined by cross-country differences in financial development, echoing the main theme of this section.

6.11 Fixed Effects With Large N and Large T

The gravity model is a standard approach to analyzing trade flows between countries. A typical application, with fixed effects might begin with

yi,j∗=αi+γj+β′xi,j+ɛi,j,i,j=1,…, n.

The model involves two sets of incidental parameters. Charbonneau (2017) examined the case in which yi,j = 1[yi,j⁎ > 0] indicates whether or not trade takes place, a binary response. The conditional logit approach will eliminate either αi or γj, but will retain the other. By applying the conditioning recursively, the author arrives at the conditional log likelihood,

lnL=∑i=1n∑j=1n∑l,k∈Zi,jlnexpβ′xl,j−xl,k−xi,j−xi,k1+expβ′xl,j−xl,k−xi,j−xi,k

where Zi,j is the set of all potential l,k that satisfy yl,j + yl,k = 1, yi,j + yi,k = 1 and yi,j + yl,j = 1 for the pair i,j. As the author notes, because the fixed effects have been eliminated, it is not possible to compute partial effects. To explore the effects of covariates, e.g., the log of distance, on the probability of trade, the unconditional probabilities are computed at the means of xi,j and with the fixed effects all set to zero.

The gravity model contains two sets of fixed effects that grow equally with the sample size. The incidental parameters problem, if there is one, is accommodated by using a conditional estimator. Fernandez-Val and Weidner (2016) consider a similar case with more general data observation mechanism—two cases considered are a probit binary response model and a Poisson regression. Both begin with an index function model,

yi,t∗=αi+γt+β′xi,t,i=1 ,…,n;t=1,…,T,

where for the probit model, yi,t = 1[yi,t⁎ + ɛi,t > 0] while in the Poisson model, E[yi,t | xi,t] = exp(yi,t⁎). The model extension allows both i and t to grow, such that N/T converges to a constant. The authors focus on bias-corrected unconditional estimators. This enables estimation of partial effects as well as coefficients. Consistent with Greene’s (2004a, 2005) results, they find that the bias of estimators of APEs is much smaller than that of the coefficients themselves. For their case, with biases diminishing in both n and T simultaneously, they find the biases in the partial effects to be negligible.

Interactive effects of the form

yi,t∗=αiγt+β′xi,t+ɛi,t

were examined by Bai (2009). Chen et al. (2014) treat this as a fixed effects model, and derived a two-step maximum likelihood estimator for probit and Poisson regression models. Boneva and Linton (2017) extend the model to allow multiple common latent factors.

Theoretical Foundation for Gravity Models in Goods Trade

Gravity models have long been used to explain bilateral trade in goods. They explain trade flows between countries i and j by their sizes (GDPs) and a variety of variables capturing the geographic and historical proximity between the two counties (distance, common language, common border, and other factors that affect trade barriers).

The theoretical foundation for the use of gravity models in the trade literature is provided by the work of James Anderson and Eric van Wincoop, among others. Anderson and van Wincoop develop a model with constant elasticity of substitution (CES) preferences and goods that are differentiated by country of origin. Prices of the same goods differ across locations due to trade costs that are not directly observable. These trade costs include not just transport costs but also information costs, design costs, and various legal and regulatory costs. The solution to the model yields a gravity equation of the form

(4.1) Xij=yiyjywτijPiPj1−σ

where σ is the elasticity of substitution.

The gravity equation says that the flow of exports from country i to country j (Xij) depends on the GDP of the two countries relative to world GDP (yiyj/yw) and on the bilateral trade costs between them (τij) relative to the product of their multilateral resistance variables (Pi and Pj). The multilateral resistance variables are aggregate measures of each country's trade costs relative to all its trading partners. An increase in the multilateral resistance of the importer (Pj) raises its trade with the exporter. For a given bilateral trade barrier between i and j, higher barriers between j and its other trading partners will reduce the relative price of goods from i and raise imports from i. An increase in multilateral resistance of the exporter (Pi) also raises its trade with j. Higher trade barriers faced by an exporter lower the demand for its goods. For a given bilateral trade barrier between i and j, this raises the level of trade between them.

When going from the gravity equation derived from the Anderson and van Wincoop model to the data, a key issue is how to measure the unobservable bilateral trade costs. This can be done using a variety of observable variables that affect trade barriers (e.g., the distance between the two countries, whether they share a common border or a common language). Therefore, the empirical counterpart of Eq. (4.1) would take the following form:

(4.2)ln(Xij)=− ∑m=1MφmZijm+ηi+γj +ɛij

Unobservable bilateral trade costs are captured by a set of M observable variables Zijm. Exporter and importer fixed effects (ηi and γj) capture the GDP of the two countries as well as their multilateral resistance variables.

3.1.2 Trade Growth vs Trade Elasticity and Gains from Trade

Independent of the question of how large the effects of tariffs or other trade policy frictions on trade flows are, Yi's analysis raises another interesting question: Even if trade policy matters for trade volume, does it matter for the gains from trade? In addition, are “trade flows” what we should be focusing on if we want to demonstrate the relevance of trade policy?

The reason Yi's work brings up this question is that his analysis was motivated by the observation that the growth of trade in the last three decades implied implausibly large trade elasticities if one were to explain the trade growth by tariff reductions alone. But large elasticities imply (conditional on trade volume) small gains from trade. One cannot have it both ways: if trade policy generates large increases in trade, then it also generates small (static) gains from trade. We will revisit this issue when we examine the evidence on the impact of trade policy on the aggregate gains from trade. For now, we simply point out that showing that trade policy played an important role in the increase of world trade does not imply that trade policy matters from a welfare point of view. And vice versa, a finding of small effects of trade policy changes on trade volume does not imply that the gains associated with these trade policy changes are small.

12.5.1 Trade facilitation

In an influential study of the cost of time delays at the border, Hummels (2001) estimates the economic value of timeliness by comparing the willingness to pay for the relatively more-expensive air cargo versus the less-expensive ocean shipping services. Using data of US imports of manufacturing goods from the rest of the world, he finds that each additional day spent on transportation corresponds to an ad valorem tariff equivalent of 0.8% on manufactured goods and reduces the probability that the US will import from a country by 1–1.5%. In light of the fact that poor trade facilitation can easily delay shipments by 10 days, these are large numbers.

Hertel et al. (2001) use Hummels’ estimates of the value of time in trade to analyze the economic impact of the Japan–Singapore Free Trade Agreement. This FTA was notable in that one of the partners (Singapore) had almost no tariffs. The bulk of the agreement focused on trade facilitation and improved movement of people and investment. The authors estimate that the expansion of e-business and automation of customs procedures between Japan and Singapore under this FTA were likely to greatly expand bilateral trade between these countries as well as their trade with the rest of the world. The latter result stems from the fact that improved customs procedures benefit imports from – and exports to – all countries, not just the FTA partners. Therefore the authors conclude that such reforms are more consistent with “open-regionalism” than are preferential tariff cuts, which tend to promote the diversion of trade from the rest of the world to FTA trading partners.

Walkenhorst and Yasui (2003) undertook a global study at the OECD using the GTAP framework in which they decompose the costs of border barriers into direct and indirect costs, where direct costs measure the logistic barriers of moving goods across borders, such as efficiency of customs services, transparency and integrity of administrative processes, and indirect costs measures the cost to firms due to delays in freight movement, border waiting times, and so on. Since the indirect cost measures timeliness and increases with waiting time, it is modeled using the iceberg approach, whereas direct costs are modeled as taxes since they generate revenue for private and public firms that provide customs facilities, shipping services, amongst others. Incorporating these into the CGE model the authors simulate the effect of reducing border-related transaction costs. They find that a reduction these transactions costs equal to 1% of world trade generates about $40 billion in welfare gains.

In many cases, these non-tariff barriers to trade dominate the impact of traditional tariffs on trade flows. Mirza (2009) utilizes a variant of the GTAP model – supplemented by econometric results relating trade facilitation to trade volume changes – in order to compare the likely impacts of tariff and trade facilitation reforms in South Asia. She finds that modest improvements in trade facilitation – measured either through a reduction in number of days delay in international shipments or through improvements as measured by the World Bank’s Logistics Perception Index – offer a much more powerful stimulus to trade in this region of the world than do tariff reforms. For example, Bangladeshi exports increase by more than $3 billion under the envisioned trade facilitation reforms, whereas full tariff removal in the region boosts trade by less than half that amount.24

Many non-tariff barriers to trade arise from domestic regulations put in place to address issues such a product safety, consumer health and environmental quality. Regulatory heterogeneity across countries in these areas often leads to trade frictions, as imports are forced to comply with new domestic requirements. At what point non-tariff measures become trade barriers is conceptually and empirically difficult to ascertain. Unlike for tariffs, “reduction to zero” of these regulatory measures is not an option. So the estimation of trade and welfare effects depends crucially of an intelligent design of scenarios (Dee et al., 2011).

International trade is the physical movement and electronic transfer of goods and services across national borders. The attributes of trade flows of interest to geographers include the direction, composition, and magnitude of trade. To analyze the geographical aspects, or ‘geographies,’ of international trade patterns, geographers make use of the following core spatial concepts: place, location, distribution, spatial interaction, spatial scale, change, region, and a variety of potentially constraining or limiting forces. Prior to the early 1990s, most of the research by geographers was focused upon commodity movements between countries, with little attention given to the evolving theory of international trade or to the interrelationships that exist between the changing global economy and the geographies of sub-national places. More recently, geographers have begun to intensify their research on international trade, and have been focusing this work at both the macro- and microlevel spatial scales. This research has important implications for world competitiveness, the internationalization and changing spatial structure of economic activities, global–local relationships, government policymaking, and national and regional economic growth and development strategies.

5.1 The Dataset for Greek Exporting Firms

This section presents an empirical application on export pricing using a new dataset that combines data from two sources: International trade flows of Greek firms are identified at the firm-level and are matched through the ICAP database to firms’ characteristics. More specifically, firm-level data on nonoil products for Greek manufacturing exporters from 2003 to 2015 are obtained from the Eurostat’s Extra-EU and Intra-EU trade statistics.2 When goods are declared to the EU statistics, they are classified according to the eight-digit Combined Nomenclature or CN8, which is the most detailed product classification system for keeping foreign trade statistics in the EU. Free-on-board export values are collected at a monthly frequency and are aggregated to annual values and quantities traded by a company to obtain export prices (unit values) as the ratio of these two variables. Defining products at a highly disaggregated level minimizes the scope for factors such as product quality differences, which are likely to determine price variations within firms.

Following the main studies surveyed in the previous section, the behavior of aggregate export prices is examined by running gravity-type regressions of export prices on main features of the destination country, namely bilateral distance to Greece, overall economic remoteness, income, and size. Following related empirical literature, data about GDP and GDP per capita are obtained from the World Bank’s World Development Indicators. Bilateral distances from Greece are obtained from CEPII. Remoteness, which proxies geographical isolation from most other nations or closeness to small countries but far away from big economies, is measured as a weighted average of a country’s bilateral distance to all other countries in the world using countries’ GDP as weights. The degree of financial constraints for each company is proxied by the credit rating score from the ICAP database, which is a 10-scale indicator controlling for insolvency, excessive and/or bad debts, overdue accounts, and other typical commercial risks, and is used routinely by Greek banks to make decisions about whether to supply credit to firms. It also is used by firms in assessing the credibility of their clients and thus provides a form of extra liquidity through short-term financing from suppliers.

To address noise or error, data are trimmed using the following rules. First, quantities equal to one or two on a monthly basis are excluded to deal with rounding errors and misreporting. To avoid product classification inconsistencies, product codes that do not match through years are excluded. Also, the data are cleaned by dropping observations with large price variations across destinations. Large variations refer to prices that differ from a factor of five or more from the mean across all destinations. Finally, the remaining price-based outliers are identified and excluded from the sample by censoring the extreme quantiles of each firm’s price distribution, below the 1st and above the 99th percentile.

After dropping these outliers, the matched sample from 2003 to 2015 contains 161,045 firm-product-destination matched observations. In Table 1, the summary statistics are shown. For the period 2003–15, there is an average of 898.3 firms employing 110.4 workers with an export volume of 5.85 million euros per year, exporting 1908.3 CN8 products to 129.3 destinations. On average, for the same period, each company exports 5.8 CN8 products to 6.5 destination markets. These figures are not far from those reported in other datasets and ensure that the dataset is comparable to those used in other empirical studies.3

Table 1. Summary Statistics for Greek Exporters, 2003–15

1 Introduction

As the name suggests, gravity equations are a model of bilateral interactions in which size and distance effects enter multiplicatively. They have been used as a workhorse for analyzing the determinants of bilateral trade flows for 50 years since being introduced by Tinbergen (1962). Krugman (1997) referred to gravity equations as examples of “social physics,” the relatively few law-like empirical regularities that characterize social interactions.1 Over the last decade, concentrated efforts of trade theorists have established that gravity equations emerge from mainstream modeling frameworks in economics and should no longer be thought of as deriving from some murky analogy with Newtonian physics. Meanwhile empirical work—guided in varying degrees by the new theory—has proceeded to lay down a raft of stylized facts about the determinants of bilateral trade. As a result of recent modeling, we now know that gravity estimates can be combined with trade policy experiments to calculate implied welfare changes.

This chapter focuses on the estimation and interpretation of gravity equations for bilateral trade. This necessarily involves a careful consideration of the theoretical underpinnings since it has become clear that naive approaches to estimation lead to biased and frequently misinterpreted results. There are now several theory-consistent estimation methods and we argue against sole reliance on any one method and instead advocate a toolkit approach. One estimator may be preferred for certain types of data or research questions but more often the methods should be used in concert to establish robustness. In recent years, estimation has become just a first step before a deeper analysis of the implications of the results, notably in terms of welfare. We try to facilitate diffusion of best-practice methods by illustrating their application in a step-by-step cookbook mode of exposition.

1.1 Gravity Features of Trade Data

Before considering theory, we use graphical displays to lay out the factual basis for taking gravity equations seriously. The first key feature of trade data that mirrors the physical gravity equation is that exports rise proportionately with the economic size of the destination and imports rise in proportion to the size of the origin economy. Using GDP as the economy size measure, we illustrate this proportionality using trade flows between Japan and the European Union. The idea is that the European Union’s area is small enough and sufficiently far from Japan that differences in distance to Japan can be ignored. Similarly because the EU is a customs union, each member applies the same trade policies on Japanese imports. Japan does not share a language, religion, currency, or colonial history with any EU members either.

Figure 3.1(a) shows Japan’s bilateral exports on the vertical axis and (b) shows its imports. The horizontal axes of both figures show the GDP (using market exchange rates) of the EU trade partner. The trade flows and GDPs are normalized by dividing by the corresponding value for Greece (a mid-size economy).2 The lines show the predicted values from a simple regression of log trade flow on log GDP. For Japan’s exports, the GDP elasticity is 1.00 and it is 1.03 for Japan’s imports. The near unit elasticity is not unique to the 2006 data. Over the decade 2000–2009, the export elasticity averaged 0.98 and its confidence intervals always included 1.0. Import elasticities averaged a somewhat higher 1.11 but the confidence intervals included 1.0 in every year except 2000 (when 10 of the EU25 had yet to join). The gravity equation is sometimes disparaged on the grounds that any model of trade should exhibit size effects for the exporter and importer. What these figures and regression results show is that the size relationship takes a relatively precise form—one that is predicted by most, but not all, models.

Figure 3.1. Trade is Proportional to Size; (a) Japan’s Exports to EU, 2006; (b) Japan’s Imports from EU, 2006. GRC: Greece

Figure 3.2 illustrates the second key empirical relationship embodied in gravity equations—the strong negative relationship between physical distance and trade. Since we have just seen that GDPs enter gravity with a coefficient very close to one, one can pass GDP to the left-handside, and show how bilateral imports or exports as a fraction of GDP varies with distance. Panels (a) and (b) of Figure 3.2 graph recent export and import data from France. These panels show deviations from the distance effect associated with Francophone countries, former colonies, and other members of the EU or of the Eurozone. The graph expresses the “spirit” of gravity: it identifies deviations from a benchmark taking into account GDP proportionality and systematic negative distance effects. Those deviations have become the subject of many separate investigations.

Figure 3.2. Trade is Inversely Proportional to Distance; (a) France’s Exports (2006); (b) France’s Imports (2006)

This chapter is mainly organized around topics with little attention paid to the chronology of when ideas appeared in the literature. But we do not think the history of idea development should be overlooked entirely. Therefore in the next section we give our account of how gravity equations went from being nearly ignored by trade economists to becoming a focus of research published in the top general interest journals.

5.6 Globalization and deindustrialization60

A popular culprit for rising labor market inequalities in developed countries is the increased globalization of economic activity arising from reductions in barriers to trade and reduced costs to international economic transactions. Increased trade with developing countries is commonly viewed as a driving force behind “deindustrialization” (a sharp decline in the share of employment in production jobs in manufacturing) and the woes of less-skilled workers in advanced economies (e.g., Wood, 1994, 1995, 1998). US manufacturing imports from less-developed countries (LDCs) increased from 0.8% of GNP in 1970 to 2.3% in 1980 to 2.8% in 1990 to 4.1% in 1996 (Borjas et al., 1997). Increased international capital mobility, reduced costs of international technology transfer, and greater foreign outsourcing opportunities also may increase the effective elasticity of demand facing workers in bargaining, erode their bargaining power, and reduce the extent to which internal labor markets insulate them from product market and labor market shocks (e.g., Borjas and Ramey, 1995; Rodrik, 1997; Bertrand, 1998).

A common (but controversial) method for estimating the effects of trade on labor markets is factor content analysis (Borjas et al., 1992, 1997; Sachs and Shatz, 1994; Wood, 1994, 1995; Lawrence, 1996). The basic approach is to determine how much of different types of labor (e.g., skilled and unskilled labor) are used to produce a country’s exports, and how much would have been used in produce its imports (or the domestic goods that would have been produced in the absence of imports). The difference between the supplies of labor used in exports and imports provides an estimate of the implicit change in the relative supply of unskilled labor from trade, or, equivalently, the impact of trade on the relative demand for the unskilled. An estimate of the aggregate elasticity of substitution between skilled and unskilled labor can then be used to simulate the impact of the implicit change in relative skill supplies from trade. Increased trade will tend to have an adverse effect on less-skilled workers to the extent that import-competing industries disproportionately employ less-skilled workers and export sectors are relatively more skill-intensive. This pattern is strongly present for US trade with LDCs, but the characteristics of workers in industries with high imports and exports with other developed countries are fairly similar (Sachs and Shatz, 1994; Borjas et al., 1997).

The factor content of observed changes in net exports can provide an accurate input to assessing how changes in trade affect relative wages in limited circumstances (see the chapter by Johnson and Stafford in this volume). If one begins in autarky, then allows for trade, and trade is a modest proportion of the national economy, the change in national endowments due to the factor content of trade measures the pressure of trade for changes in relative wages (Deardorff and Staiger, 1988; Krugman, 1995). More generally, if the changes in net exports being examined are caused by external factors (e.g., reductions in trade barriers or reductions in transportation costs, changes in factor endowments abroad), then factor content analysis may be sensible. If changes in net exports result from domestic sources (e.g., an increase in the relative supply of skilled labor leading to greater net exports of high-skill goods and lower net exports of low-skill good), then factor content analysis can be quite misleading (Learner, 1996).

A further practical issue in factor content analysis is the how to estimate the hypothetical factor content of the domestic production that would arise to replace imports from LDCs. The standard approach is to assume LDC imports would be replaced by domestic production in the closest import-competing industry using the contemporaneous average factor proportion in the domestic import-competing industry (e.g., Sachs and Shatz, 1994). But Wood (1994, 1995) has argued persuasively that within each sector there is a wide distribution of factor proportions and labor productivity, and that LDC imports are likely to be most directly competing with the segment of an industry using the most unskilled-labor intensive production techniques. The issue is somewhat more complicated since some LDC imports may not closely compete with any domestic industry so that their absence might expand domestic demand for goods or services with quite different (and possibly even higher) skill intensities than in the assumed “import-competing” sector.

Borjas et al. (1997) examine the factor content of the growth of US trade with LDCs from 1980 to 1995. They examine the robustness of the conclusions to a wide range of assumptions concerning the factor ratios that would have been used in US industries to replace LDC imports. They find that the growth of trade with LDC’s from 1980 to 1995 to a 1.4 log point increase in the implicit relative supply of high school equivalents relative to college equivalents assuming US manufactures would use the same factor ratios that prevailed in their industries in 1980 (prior to the change in LDC trade being assessed) in the absence of LDC imports. Under our preferred estimate of σ = 1.4, this implies that growth of trade with LDCs can account for only 1 log point out of a 19 log point increase in the college wage premium from 1980 to 1995. Thus demand shifts from skill-biased technological change and domestic sources of changes in relative skill supplies appear to be much more significant factors in the recent expansion of the US college wage premium than trade’s impact as measured by factor contents. The impact is relatively larger if one focuses on the impact of trade on the high school dropouts. But Borjas et al. also find that increased unskilled immigration had a much larger impact on changing the implicit relative supply of the least skilled US workers than did LDC trade from 1980 to1995.

The factor content approach may understate the effects of globalization pressures on relative wages when the threat of trade, outsourcing, or plant relocation can lead to wage changes even in the absence of new trade flows (Rodrik, 1997).61 Borjas and Ramey (1995) explore the contribution of the erosion of industry wage differentials in trade competing durable goods manufacturing industries to increased US educational wage differentials and find it to be quite modest.

Product-price studies attempt to more directly assess the implication of the Stolper– Samuelson theorem that impacts of trade on relative wages operates through changes in the relative product price of more- and less-skill intensive. Product-price studies suffer from similar practical limitations to factor-content studies both arising from data quality issues in price data (the difficulty of separating true price from quality changes) and difficulties in trying to isolate product-price changes driven by exogenous trade-related forces rather than other sources. Slaughter (1998) provides a nice review of the emerging literature in this area and concludes that these limitations combined with a wide range of somewhat conflicting results make it difficult to draw strong conclusions from the price studies concerning the impact of international trade on wage inequality. Attempts to isolate “exogenous” international components of changes in product prices and trade flows (possibly by examining the consequences of changes in trade policy and explicit trade barriers) could be a more fruitful research strategy than standard approaches to factor content analysis and product price studies.

“Deindustrialization” (a substantial decline in manufacturing employment) is also often identified as a leading cause of poor labor market performance of less-skilled workers in advanced countries. And international trade is often viewed as the major driving force behind deindustrialization (e.g., Wood (1994, 1995, 1998). Between-industry demand shift indices (Section 5.4) do indicate that shifts out of manufacturing to moreskill intensive sectors have played some role in the decline in the relative demand for less-skilled workers. But the overall rate of between-industry demand shifts does not appear to be any larger in the period of sharp increases in wage inequality in the 1980s than in other recent decades. Nevertheless, it is striking that much of the recent increase in US wage inequality and educational wage differentials is concentrated in the period from 1979 to 1985 centered on a deep recession and containing a large appreciation of the US dollar and large decline in manufacturing employment. And the periods of extremely tight labor markets and strong demand for production workers in manufacturing during the two World Wars are the two periods of large compressions in the US wage structure during the 20th Century.62 Furthermore studies using geographic variation across US states and metropolitan areas consistently find that larger declines in manufacturing employment are strongly positively associated (at least in the short-run) with larger increases in overall wage inequality (Juhn, 1994), residual wage inequality (Bernard and Jensen, 1998), and educational wage differentials (Borjas and Ramey, 1995; Bound and Holzer, 1997).