Tradable Services: Understanding the Scope and Impact of Services Offshoring [with Comments and Discussion] Author(s): J. Bradford Jensen, Lori G. Kletzer, Jared Bernstein and Robert C. Feenstra Source: Brookings Trade Forum, Offshoring White-Collar Work (2005), pp. 75-133 Published by: Brookings Institution Press Stable URL: http://www.jstor.org/stable/25058763 Accessed: 23-03-2017 10:39 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://about.jstor.org/terms Brookings Institution Press is collaborating with JSTOR to digitize, preserve and extend access to Brookings Trade Forum This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. BRADFORD JENSEN Institute for International Economics LORI G. KLETZER University of California-Santa Cruz and Institute for International Economi Tradable Services: Understanding the Scope and Impa of Services Off shoring Globalization, particularly globalized production, is evolving and broaden from manufacturing into services. Services activities now account f larger share of global trade than in the past. Services trade has almost doubl over the past decade: in the period 1992 to 2002, exports increased fro $163 billion to $279 billion, and imports increased from $102 billio $205 billion. These changes, and their implications for American firms workers, have attracted widespread attention. Coincident with the broadening of global economic integration from manu facturing to services, the face of job displacement in the United States is ch ing. While manufacturing workers have historically accounted for more tha half of displaced workers, over the period 2001-03, nonmanufacturing worke accounted for 70 percent of displaced workers.1 The share of job loss account for by workers displaced from information, financial services, and professio and business services nearly tripled, from 15 percent during the 1979-82 rec sion to 43 percent over the 2001-03 period. The industrial and occupational s We appreciate the comments and suggestions of our Brookings Trade Forum discussants, J Bernstein and Robert Feenstra, as well as those of Andrew Bernard, Catherine Mann, Michael Mus Dave Richardson, Peter Schott, and seminar participants at the Institute for International Eco ics; the University of California, Santa Cruz; and the 2004 Empirical Investigations in Internat Trade conference. We gratefully acknowledge the support of the Alfred P. Sloan Foundation. 1. The shift in job loss from manufacturing and production workers toward service and w collar (nonproduction) workers has been in evidence since the recession of the early 1990s. At time, concerns about downsizing and reengineering were coincident with a rise in the share white-collar and service sector job loss. See Podgursky (1992); F rber (1993); Gardner (19 and Kletzer (1995, 1998). 75 This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 76 Brookings Trade Forum: 2005 in job loss has been associated with a rise in the probability of job loss for more educated workers.2 Bringing these two trends together, the changing mix of industries exposed to international trade in services may have deep implications for the structure of U.S. industry and labor markets in the future. Currently, there is little clear understanding of the role of services globalization in domestic employment change and job loss. More fundamentally, there is little clear understanding of the size and extent of services offshoring, how large it is likely to become in the near-term future, or what impact it is having on the U.S. economy. Fueled by the 2004 presidential race and continued slack in the labor market, the services offshoring debate became headline material. The literature on ser vices offshoring is expanding rapidly. A nonexhaustive list of recent contributors includes: Amiti and Wei (2004); Arora and Gambardella (2004); Bardhan and Kroll (2003): Bhagwati, Panagariya, and Srinivasan (2004); Brainard and Litan (2004); Bronfenbrenner and Luce (2004); Dossani and Kenney (2003, 2004); Kirkegaard (2004); Mann (2003); Samuelson (2004); and Schultze (2004). Despite the attention, relatively little is known about how many jobs may be at risk of relocation or how much job loss is associated with the business decisions to offshore and outsource. There are a few prominent projections, advanced mostly by consulting firms. The dominant and most widely quoted projection of future job losses due to movement of jobs offshore is Forrester Research's estimate of 3.3 million.3 Oth ers include: Deloitte Research's estimate that by 2008 the world's largest finan cial service companies will have relocated up to 2 million jobs to low-cost coun tries offshore; Gartner Research's prediction that by the end of 2004 10 percent of IT jobs at U.S. IT companies and 5 percent of IT jobs at non-IT companies will have moved offshore; and Goldman Sachs's estimate that 300,000 to 400,000 services jobs have moved offshore in the past three years, and that 15,000 to 30,000 jobs a month, in manufacturing and services combined, will be subject to offshoring in the future.4 It is clear that changes in technology are enabling more activities to be traded internationally. What is unclear is how large these trends are likely to become, 2. It is still the case that less-educated workers have the highest rates of job loss overall. Over the 2001-03 period, the rate of job loss for workers with a high school diploma or less was .141; for workers with at least some college experience, the rate of job loss was .096 (estimates from the 2004 Displaced Worker Survey). See F rber (2005) for a more detailed examination of worker characteristics and the risk of job loss. 3. See McCarthy (2002). The Forrester projection was updated in 2004 to 3.4 million. 4. See, in order, Gentle (2003); Gartner Research (2004); and Tilton (2003). This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms /. Bradford Jensen and Lori G. Kletzer 11 the sectors and occupations affected to date and going forward, and the impact on workers of the resulting dislocations. Without understanding the nature and scope of the changes, it is difficult to formulate effective public policy to address emerging needs. This paper develops a new empirical approach to identifying, at a detailed level, service activities that are potentially exposed to international trade. We use the geographic concentration of service activities within the United States to identify which service activities are traded domestically. We classify activities that are traded domestically as potentially tradable internationally. Using the identified industries and occupations, we develop estimates of the number of workers who are in tradable activities for all sectors of the economy. We com pare the demographic characteristics of workers in tradable and nontradable activities and employment growth in traded and nontraded service activities. We also examine the risk of job loss and other employment outcomes for workers in tradable activities. To preview the results, we find considerable employment shares in tradable service industries and occupations. Based on our estimates, there are more work ers in tradable professional and business service industries than in tradable man ufacturing industries. We also examine the characteristics of workers in tradable and nontradable activities and find that workers in tradable sectors have higher skills and significantly higher wages. Within specific sectors like professional services, the earnings differentials are even larger, approaching 20 percent. When we examine employment growth trends across traded and nontraded activities, tradable activities have lower growth rates, due primarily to employ ment losses in manufacturing. Within services, tradable and nontradable activi ties have similar growth rates except at the lowest end of the skill distribution. Low-skill tradable industries and occupations have negative average employ ment growth, whereas employment growth in nontraded, low-skill services is positive (though low). We also examine worker displacement rates in tradable and nontradable service activities. We see some evidence that displacement rates are higher from tradable service industries than from nontradable. We also find higher dis placement rates from tradable white-collar occupations than from nontradable. Consistent with the characteristics of employed workers, we find that workers displaced from tradable service activities are more educated, with higher earn ings, than workers displaced from nontradable activities. Job loss from tradable and nontradable service activities is costly to workers in terms of earnings losses (comparing new job earnings to old job earnings). Taken together, the results are consistent with the view that economic activity within the United This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 78 Brookings Trade Forum: 2005 States is moving toward a U.S. comparative advantage in services, similar to manufacturing. In the next section we describe our empirical approach to identifying tradable activities. The following sections describe the tradable and nontradable cate gories for both manufacturing and services activities; compare worker charac teristics in tradable and nontradable services; explore the employment trends in tradable and nontradable services; and consider the most recent evidence on job displacement from tradable activities. Empirical Approach Historically, services have been considered nontradable, with a paucity of empirical work examining trade in services relative to empirical work on man ufacturing. To examine the potential impact of trade in services on the U.S. econ omy, we wanted to identify the size and scope of services trade at as detailed a level as possible. As many observers and researchers have noted, gathering detailed data on the extent of services offshoring is quite difficult. While the Bureau of Economic Analysis (BEA) provides data on international trade in services, the data on international trade in services that BEA publishes do not provide particularly detailed industry-level data. Table 1 shows the level of industry detail available from BEA. Our interest in examining trade in services in more detail than what is avail able through the BEA services trade data necessitated an alternative empirical approach to identifying tradable service activities. Our approach to identifying service activities that are potentially tradable is novel: we use the geographic concentration of service activities in the United States to identify industries and occupations that appear to be traded domestically. From this domestic informa tion, we infer that service activities that can be traded within the United States are also potentially tradable internationally. Framework The economic intuition we rely on to develop our baseline measure of trad able services is that nontraded services will not exhibit geographic concentration in production. We observe that goods that are traded tend to be geographically concentrated (to capitalize on increasing returns to scale, access to inputs such as natural resources, etc.), while goods that are not traded tend to be more ubiq uitously distributed. We apply this same intuition to service production. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 79 Helpman and Krugman (1985) present a model that demonstrates this intu ition. They model a world with two goods, two countries, and three industries, where the first industry is assumed to be a nontradable constant-returns sector, the second industry is an industry with differentiated varieties that are assumed to be costlessly traded, and the third industry is a tradable constant-returns sec tor. Helpman and Krugman derive the input vectors V(l), V(2), and V(3) for the integrated world equilibrium. With homothetic and identical tastes, if country y has a share sj of world income, it must allocate resources sjY(l) to the nontrad able industry; that is, the production of the nontraded good must be allocated between countries in proportion to their shares of world income. Nontraded goods are distributed uniformly according to population and income. This intuition is revealed more descriptively by Paul Krugman, who notes, "In the late twentieth century the great bulk of our labor force makes services rather than goods. Many of these services are nontradable and simply follow the geo graphical distribution of the goods-producing population fast-food outlets, day care providers, divorce lawyers surely have locational Ginis pretty close to zero. Some services, however, especially in the financial sector, can be traded. Hartford is an insurance city; Chicago the center of futures trading; Los Angeles the enter tainment capital; and so on_The most spectacular examples of localization in today's world are, in fact, services rather than manufacturing_Transportation of goods has not gotten much cheaper in the past eighty years_But the ability to transmit information has grown spectacularly, with telecommunications, com puters, fiber optics, etc."5 The idea is that when something is traded the produc tion of the activity is concentrated in a particular region to take advantage of some economies in production. As a result, not all regions will support local production of the good, and some regions will devote a disproportionate share of productive activity to a good and then trade it.6 We use the geographic concentration of ser vice activity within the United States as an indicator that the service is traded within the United States and thus potentially tradable internationally. The "locational Gini" referred to by Krugman is one of several ways to meas ure geographic concentration.7 The measures compare a region's share of 5.Krugman(1991,p.65). 6. The relationship between geographic concentration of production and trade, particularly exports, has a long tradition in both economic geography (where the measure used is the location quotient) and trade analysis (where the measure used is revealed comparative advantage). The measures of economic concentration used in this paper are different from the location quotient and revealed comparative advantage measures, but all the measures have a similar flavor in that they compare the share of production (or exports) in a particular region to an "expected" baseline. 7. Among the different empirical approaches to measuring geographic concentration and agglomeration are Duranton and Overman (2004). This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 80 Brookings Trade Forum: 2005 Table 1. Private Services Trade by Type, 2002 Millions of dollars Trade type Exports, 2002 Imports, 2002 Travel Overseas Canada Mexico Passenger fares Other transportation Freight Port services Royalties and license fees Affiliated U.S. parents' transactions U.S. affiliates' transactions Unaffiliated Industrial processes Other Other private services Affiliated services U.S. parents' transactions U.S. affiliates' transactions Unaffiliated services Education Financial services Insurance services Telecommunications Business, professional, and technical services Accounting, auditing, and bookkeeping services Advertising 66,547 54,772 6,268 5,507 17,046 29,166 12,330 16,836 44,142 32,218 29,066 3,152 11,924 3,900 8,024 122,594 43,500 25,194 18,306 79,094 12,759 15,859 2,839 4,137 28,799 360 633 58,044 44,494 6,489 7,061 19,969 38,527 25,973 12,554 19,258 15,132 2,958 12,174 4,126 1,935 2,192 69,436 32,367 17,529 14,838 37,069 2,466 3,665 15,348 4,180 10,732 716 1,360 (continued) employment in or output of an activity with the region's share of overall eco nomic activity. We make use of two common measures of geographic concentra tion; but before turning to those measures we address one more conceptual issue. Demand-Induced Agglomeration and Intermediate Services Measures of geographic concentration are a way to implement the intuition described above. Most measures of concentration use the region's share of employment in an industry relative to the region's share of total employment. The measures of concentration do not differentiate the reasons activity is con centrated. It does not matter whether production is concentrated because of the This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms /. Bradford Jensen and Lori G. Kletzer 81 Table 1. Private Services Trade by Type, 1992-2002 (Continued) Millions of dollars Exports, Imports, Trade type 2002 2002 Agricultural, mining, and onAgricultural and mining service Waste treatment and depollution Architectural, engineering, and Computer and data proc Construction, architectural, engin mining services n.a. n.a. Construction 654 226 Data base and other infor Industrial engineering 749 185 Installation, maintenance, and Legal services 3,270 768 Management, consulting, an Medical services 1,901 n.a. Miscellaneous disbursements 62 Operational leasing 3,573 190 Research, development, and Sports and performing arts 175 1 Trade-related services 353 95 Training services 501 361 Other business, professional, Other unaffiliated services 14,7 Source: Bureau of Economic Analysis, n.a. = not available. location of natural resources, in to the agglomeration of worker the good or service is produce sumed. So, in general, the reason except in one instance. If a serv concentrated (that is, if industr cally concentrated), the service we would incorrectly infer that To incorporate this case into ou framework. If a nontradable ind stream industry, we would expe intermediate industry to follow Instead of being distributed wit proportion to the geographic dis This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 82 Brookings Trade Forum: 2005 We construct region-specific measures of demand for each industry using the 1999 input-output use tables produced by the Bureau of Economic Analysis.8 This measure of industry demand share (IDSip) represents how much geographic concentration there is in demand for a good or service i in a particular region p. We construct the demand for industry / in Place of Work Metro Area/? by: IDSip = X, (VY, * InEMP^/InEMP,), (1) where Yu = the output of industry i used by industry j (including government and private households as "industries"); Y i = total output of industry /; InEMP p = industry j employment in region/?; InEMP, = total employment in industry/ We include both direct use and investment in the "use" of industry / output by industry/ To construct the region-specific measures of demand for each occupation, we use the industry-region-specific demand measures described above and weight those by the share of occupation employment in an industry. 0DSo>p = Xj (IDS,, * 0cEMPo ,/OcEMPJ, (2) where lDSjp = industry demand share for industry j in region/?; OcEMPOJ = occupation o employment in industry j; and OcEMP0 = total employment in occupation o. These adjustments take account of the concentration of downstream industry concentration and adjust the "denominator" in the geographic concentration measures that follow. Measuring Geographic Concentration The first measure of economic concentration, as described in Ellison and Glaeser(1997),is: EC^is^-Xpf. (3) 8. For more information, see www.bea.doc.gov/bea/dn2/i-o.htm. We aggregate some BEA input-output (10) industries to a level consistent with the industry classification used by the Cen sus Bureau on the 2000 Decennial PUMS (Public Use Micro Sample). This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 83 The measure is an index for comparing a region's share of industry employment (sip) with the area's share of aggregate activity/employment (xp). When an area's employment share in an activity is significantly greater than the area's share of aggregate employment, this is interpreted as indicating a concentration, or spe cialization, in the given activity. The index EC provides a national index for each industry, and measures of EC indicating geographic concentration are inter preted as indicative of trade in that activity, in the sense that "local" employment exceeds "local" demand in some areas and the difference is traded outside the area. We modify the EC measure to look at the difference between the region's share of industry employment and the region's share of industry demand, as noted above: EQ^is^-IDS^f. (4) The new measure of EC is an index for comparing a region's share of an indus try's employment (s ) with the region's share of demand for that industry (IDSJ. We do not make the Herfindahl adjustment that Ellison and Glaeser (1999) use in their index of agglomeration because we are not interested in agglomer ation (the co-location of different firms in the same industry), but are interested in pure geographic concentration (whether the concentration is due to one firm or a number of firms). If economic activity is concentrated because significant scale economies are captured within a firm, we do not want to discount this concentration. The second measure of geographic concentration we use is the Gini coeffi cient. The Gini coefficient (G) for the concentration of industry activity is given by: Gt = \ 1 -M Yi,P-i + <^) * (aX,,_7-aXp) I, (5) where /?'s index regions (sorted by the region's share of industry employment), ten years 0.23 0.12 0.14 Educational attainment (share) High school dropout 0.14 0.05 0.11 High school graduate 0.40 0.19 0.31 Some college 0.24 0.30 0.33 College + 0.22 0.45 0.25 Male 0.61 0.54 0.45 In predisplacement job Share with health insurance 0.75 0.66 0.47 Full-time 0.96 0.90 0.82 If full-time, real weekly earnings (dollars) 342.70 443 Standard deviation (dollars) 300.54 Share reemployed 0.64 0.77 0.75 Of reemployed, share full-time 0.80 0.78 0.72 All reemployed Change in In earnings (mean) -0.32 -0.30 -0.14 Standard deviation 0.89 0.98 1.02 Median change -0.15 -0.11 -0.03 Share with no loss in earnings 0.42 0.45 0.51 Full-time to full-time Change in In earnings (mean) -0.21 -0.21 -0.12 Standard deviation 0.76 0.69 0.97 Median change -0.10 -0.07 -0.03 Share with no loss in earnings 0.42 0.46 0.52 Source: Authors' calculations from the 2004 Displaced Worker Survey, using sampling weights. Agriculture, Mining, an tion omitted. Conclusions This paper develops a new empirical approach to identifying, at a level for the entire economy, industries and occupations that are tradab the methodology, we find substantial employment in tradable service in and occupations. Workers in these industries and occupations are mo skilled and have higher earnings than workers in the manufacturing se nontradable service activities. The higher earnings are not solely a higher skill levels: in regressions controlling for observable charact This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms Table 17. Characteristics of Selected Service Sector Displaced Workers, by Industry and Tradabilit Information Tradable Nontradable Financial, insura real estate Tradable Nontr Job tenure (mean in years) Standard deviation Job tenure > ten years Educational attainment (share) High school dropout High school graduate Some college College + Male In predisplacement job Share with health insurance Full-time If full-time, real weekly earnings (dollars) Standard deviation (dollars) Share reemployed Of reemployed, share full-time All reemployed Change in In earnings (mean) Standard deviation Median change Share with no loss in earnings Full-time to full-time Change in In earnings (mean) Standard deviation Median change Share with no loss in earnings 5.80 7.37 0.192 0.032 0.207 0.262 0.499 0.559 0.82 0.93 530.82 409.45 0.72 0.76 -0.57 1.07 -0.34 0.346 -0.40 0.82 -0.25 0.36 4.51 7.25 0.16 0.00 0.038 0.45 0.512 0.668 0.62 0.87 387.98 350.69 0.81 0.87 -0.72 2.97 -0.024 0.469 -1.003 3.328 -0.07 0.344 5.82 7.00 0.167 0.04 0.179 0.389 0.392 0.47 0.62 0.91 409.88 380.43 0.61 0.80 -0.16 1.09 -0.08 0.456 -0.15 0.51 -0.047 0.457 54 45 Source: Authors' calculations from the 2004 Displaced Worker Survey, using sampling weights. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms Table 18. Characteristics of Displaced Workers in Selected Service Occupations, by Occupation and T Worker characteristics Management, business, andfinancial Tradable Nontradable Professional and related Tradable Nontra Job tenure (mean in years) Standard deviation Job tenure > ten years Educational attainment (share) High school dropout High school graduate Some college College + Male In pre-displacement job Share with health insurance Full-time If full-time, real weekly earnings (dollars) Standard deviation (dollars) Share reemployed Of reemployed, share full-time All reemployed Change in In earnings (mean) Standard deviation Median change Share with no loss in earnings Full-time to full-time Change in In earnings (mean) Standard deviation Median change Share with no loss in earnings 6.72 8.04 0.204 0.008 0.132 0.269 0.591 0.466 0.775 0.965 554.78 434.23 0.786 0.791 -0.374 1.08 -0.127 0.492 -0.205 0.852 -0.045 0.528 5.03 4.99 0.143 0.012 0.272 0.28 0.436 0.633 0.588 0.927 426.02 336.05 0.72 0.726 -0.364 1.144 -0.165 0.389 -0.357 1.165 -0.109 0.351 4.82 6.09 0.111 0.003 0.092 0.198 0.708 0.717 0.794 0.93 523.24 369.44 0.80 0.805 -0.34 1.155 -0.084 0.455 -0.318 1.176 -0.068 0.462 4. 5 0 0 0 0 0 0 0 0 323 226 0 0 -0 0. -0. 0. -0. 0. -0. 0. Source: Authors' calculations from the 2004 Displaced Worker Survey, using sampling weights. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 114 Brookings Trade Forum: 2005 workers in selected tradable service activities earn 16-17 percent higher incomes than similar workers in nontradable activities in the same sector. Examining employment growth across industries and occupations, there is lit tle evidence that tradable service industries or occupations grow more slowly than nontradable industries or occupations overall, though at the low end of the skill distribution employment growth is negative for tradable services. High skill service activities have the highest employment growth rates. There is job insecurity associated with employment in tradable activities, including service activities. We find a higher rate of job loss from tradable indus tries than from nontradable industries, with the greatest difference outside of manufacturing. In comparison with an overall rate of job loss of .103 for 2001-03, tradable nonmanufacturing industries have a rate of job loss of .128 and nontradable industries .073 (though we note the possibility that these dif ferences are driven by the tech bubble). Also within occupations, workers in tradable jobs faced a higher rate of job loss than workers in nontradable jobs, with the greatest difference within white-collar occupations. These results have several implications. First, it seems inappropriate to con sider all service activities as inherently nontradable. The geographic concentra tion of some service activities within the United States is as great as in manu facturing and is consistent with the view that a number of service industries and occupations are tradable. The share of employment in tradable services is large enough that a better understanding of the forces shaping trade in services war rants our attention. At a minimum, more resources should be devoted to collect ing and publishing considerably more detail on international service flows. Con tinuing to increase the amount of information collected on the use of intermediate service inputs within the United States would also increase our ability to track and understand developments in this large and growing sector. Second, the results presented in this paper suggest that tradable services are consistent with U.S. comparative advantage. While professional and business services jobs require higher skills and pay higher wages than manufacturing jobs in general, tradable services jobs in these sectors require even higher skills and are more highly paid than nontradable service activities. We would expect that as technological and organizational change increases the potential for trade in services, economic activity in the United States will shift to activities consistent with U.S. comparative advantage.26 It is therefore possible that further liberal ization in international services trade would directly benefit workers and firms in 26. The United States maintains a positive trade balance in service activities; see table 1. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms /. Bradford Jensen and Lori G. Kletzer 115 the United States. The policy community should devote more attention to under standing the impediments to services trade. Third, although tradable services have relatively high employment growth rates overall, at the low end of the skill distribution tradable service activities have negative employment growth. The potential for reallocation across activi ties in response to shifting trade patterns in services is real. Policymakers should prepare for additional reallocation among this group of workers. The process of adjustment to job displacement might be eased by service worker characteristics. For the most part, workers displaced from tradable serv ices are different, in terms of job tenure and educational attainment, from work ers displaced from (tradable) manufacturing industries. Generalizing from what we know from studies of manufacturing worker job loss, lower levels of job tenure and higher levels of educational attainment may be advantages in seeking reemployment. Given the current availability of data, it is too early to tell. We need data beyond the time period of the "jobless recovery." We also need more information to discern whether workers in tradable activities face different reem ployment outcomes than workers in nontradable activities. The evidence we do have tells us that job loss for services workers is costly. These costs underscore the need to have a less porous safety net (for example, by extending Trade Ad justment Assistance [TAA] to services workers and extending wage insurance beyond TAA). Lower rates of employment growth at the lower end of the skill distribution in tradable service activities may have implications for the retrain ing strategies and opportunities for displaced low-skill workers in both manu facturing and services. Appendix: Displaced Worker Survey The Displaced Worker Survey is administered biennially as a supplement to the Current Population Survey (CPS). The first survey was administered in Jan uary 1984 and the most recent in January 2004. In each survey, adults (aged 20 years and older) in the regular monthly CPS were asked if they had lost a job in the preceding three- or five-year period due to "a plant closing, an employer going out of business, a layoff from which he/she was not recalled, or other sim ilar reasons."27 If the answer was yes, a series of questions followed concerning 27. For the 1984-92 surveys, the recall period was five years. Starting in 1994, the recall period was shortened to three years. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 116 Brookings Trade Forum: 2005 the lost job and the period of joblessness. Other causes of job loss, such as quit ting and firing, are not considered displacements.28 This categorization is con sistent with our common understanding of job displacement: it occurs without personal prejudice in that terminations are related to the operating decisions of the employer and are independent of individual job performance. This opera tional definition is not without ambiguity: the displacements are "job" displace ments, in the sense that an individual displaced from a job and rehired into a dif ferent job with the same employer is considered displaced. A key advantage of the DWS is its large-scale representative nature. As part of the CPS, it draws on a random sample of 60,000 households, which is weighted to be representative of the U.S. workforce. As a result, the surveys yield responses from large numbers of displaced workers in a wide set of indus tries. In exchange for breadth of coverage, the DWSs have two weaknesses rel evant to any study of the costs of job loss. The first is the relatively short-term horizon. Individuals are surveyed just once, providing information about one postdisplacement point in time, rather than about their experiences over time. The second weakness is the lack of a readily available comparison group of nondisplaced workers. Without such a comparison group, we cannot investigate what would have happened to these workers if they had not been displaced. The lack of a comparison group leads to some unavoidable errors in measuring out comes such as postdisplacement reemployment and earnings losses. The rate of job loss reported in the tables is calculated as in F rber (1993, 2003, 2005): it is the ratio of the (weighted) number of reported displacements divided by the (weighted) number of workers who were either employed at the survey date or reported a job loss but were not employed at the survey date. See Kletzer (2001) for more discussion of the issues that arise when using the DWSs to measure the incidence of job loss. 28. Individuals who respond that their job loss was due to the end of a seasonal job or the fail ure of a self-employed business are also not included. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms Comments and Discussion Jared Bernstein: Jensen and Kletzer have written a refreshingly clear and insightful paper that readers will find to be one of more useful contributions to the often fuzzy literature on offshoring. Much of this work has tried to identify the service or white-collar jobs at risk to offshore competition, but we have been stymied by the difficulty of using trade data on service flows for this purpose. These authors derive a clever method using geographical clustering for doing so, and while they may need to work a bit harder to convince skeptics, many will find their approach convincing, as I do. This innovative classification scheme sets the stage for the paper's other main contribution: a description of the char acteristics and earnings of those in tradable services relative to those in nontrad able services. One criticism of the paper is that the title promises more than the authors, or anyone else for that matter, can yet deliver. That is, while they go further than others toward identifying the industries and occupations directly affected by off shoring, to truly capture the "scope and impact" of this growing competitive challenge, researchers need to go beyond the direct effects. The authors do point out that displaced workers in tradable services suffer large wage losses relative to other displaced workers, but (a) it is not clear that this is because they are in tradable services, and (b) surely the impact of offshoring goes beyond this sub group. This latter point is critical. The implicit supply shock from adding mil lions of skilled workers to a relatively concentrated set of occupations and indus tries may have a significant negative impact on the wage structure of white-collar workers, much as the increase of trade in manufacturing goods with low-labor cost competitors has structurally altered the wage distribution of blue-collar 117 This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 118 Brookings Trade Forum: 2005 workers. In short, a white-collar worker needn't get displaced to feel the impact of this growing phenomenon. Using Geographic Clustering to Identify Tradable Services The credibility of their paper rests on the authors' novel method for identify ing tradable services. They point out that BEA data on international trade flows in services are not disaggregated enough by industry to serve this purpose. But the problem goes deeper than this. As my EPI colleague Josh Bivens points out, these data, especially the highly relevant parts relating to information technology, are getting a bit hard to believe, given what so many firms are telling us about their service imports and what some other countries' service export data suggest. Take, for example, data on the value of imports of computer-related services, which includes software writing, from India. Even with recent large upward revisions, the tiny magnitudes of the BEA numbers for example, $330 million in 2003 are hard to believe. The Indian tech trade group NASSCOM puts this value at $4.7 billion. This is not to suggest that NASSCOM's data capacity is superior to BEA's. Rather, if you're out to identify service jobs affected by offshoring, most analysts are suspicious of the quality of our data on the import of some key services asso ciated with offshoring. At any rate, Jensen and Kletzer use the assumption that tradable firms exhibit geographic concentration. This assumption comes from research on the goods sector, where returns to scale, access to transportation nodes, and proximity to natural resources lead goods producers to congregate near each other. Is it rea sonable to extend this to service production? Empirically, we can, without much effort, observe this concentration, or lack thereof. Silicon valleys and "research triangles" have appeared in numerous places over the past decades. Meanwhile, bowling alleys and child-care centers are scattered pretty much all over the place. In this regard, their transporting of this method of identifying tradable industries from goods to services does not seem a big stretch. There are, however, some differences between goods and services that will lead some readers to wonder if scale economies and access issues loom large enough in services to motivate geographic clustering. For example, to transport cars or steel, manufactures have historically needed to locate near waterways. But it is hard to see why this constraint would hold for, to take a very relevant case, transmitting information across the Internet. In fact, it is the sharp decline This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 119 in such costs that has allegedly motivated service firms to offshore data to extremely distant places. So they may need to work a little harder to convince skeptics. What are the specific benefits they have in mind that motivate tradable services to locate near each other? Are there some case studies they could cite? As mentioned, it is not hard to point to areas where high-tech firms are concentrated, but there could be lots of reasons for that, including niche education and labor markets: California's Silicon Valley and North Carolina's Research Triangle, for example, are both near universities with specialties in computer science. And where I live, in north ern Virginia, our silicon alley, out Route 66 in the Dulles corridor, likely grew out of the desire to be close to federal government contractors and purchasers. What is the connection to international trade? And why shouldn't nongeo graphically clustered service industries offshore some of their jobs? Hospitals, for example, score in the authors' least geographically concentrated category, presumably because they are pretty pervasive across localities in our economy. But anecdotes suggest that hospitals are beginning to offshore some of their accounting services, certainly a plausible scenario (anecdotes also suggest hos pitals are offshoring high-tech functions, like radiology services, but as the con ference paper by Frank Levy and Ari Goelman (this volume) finds, this does not appear to be occurring).1 While I encourage them to work a little harder to convince the reader that their classification scheme is up to the task, a close look at their tables and fig ures reveals strong face validity. There are a few industries, such as hospitals, that seem questionably classified as nontradable (accounting, tax preparation, bookkeeping, and payroll services is another), but no such system will be per fect. In the case of the two examples I just mentioned, they are services that by their nature tend to be demanded in most localities and thus fly under the radar of their test. So perhaps Jensen and Kletzer can think of an added filter that would help address such industries.2 They presumably pick up some of these jobs in their occupational analysis. Their table 8, for example, shows that 11 percent of total employment is in trad able occupations in nontradable industries. Still, the apparent misclassification of a few industries may unsettle some readers. 1. Levy and Goelman show that both gatekeeper actions by U.S. radiologists and malpractice regulations explain why hospitals are hard-pressed to offshore such services. 2.1 doubt anyone would squawk if they just added a few industries like hospitals and tax prepa ration services that are widely reported to be tradable services, even though they are not geo graphically concentrated. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 120 Brookings Trade Forum: 2005 They are careful to avoid the following mistake: suppose a nontradable upstream service provides an intermediate service to a tradable downstream service industry. If the upstream firms need to locate near the downstream firms, they will be misclassified as tradable services. For example, if a computer firm both offshores programming tasks to India and outsources payroll services to a nearby firm, the authors could end up mistakenly labeling the upstream industry as a tradable service. To avoid this, they use input/output tables to parse the upstream services from the downstream ones. A final concern is in regard to the role of productivity growth in their method of using workers to identify where firms are clustered. If demand is constant, falling, or not growing too quickly, as was arguably the case over their period of study, firms with fast-growing productivity might be shedding workers. The impact of this on their analysis is not necessarily problematic, as long as the firms in such industries remain clustered (and it is hard to see why they would not). But this may be one reason why this type of analysis is usually based on more direct measures of industry output (one reason they are sticking with work ers is because they want to examine occupations as well as industries). Comparing the Characteristics of Workers in Tradable and Nontradable Jobs As one might have expected, given the anecdotes in the newspapers, jobs in tradable services pay more than those in nontradable services: a 35 percent annual earnings differential in tradable services, unadjusted for worker differ ences, and a large adjusted differential, discussed next. Such workers are also more likely to be male and have higher educational attainment. With a set of earnings regressions, the authors find a statistically and eco nomically large premium associated with being in a tradable industry, a tradable occupation, and a combination of the two (in their later analysis on displace ment, we see the downside of this workers displaced from such jobs experi ence large relative losses). Relative to those in nontradable industries and occu pations, the premium amounts to between 10 and 17 percent, depending on the sample. What is interesting here is that the impact of being in such industries and occupations is modeled as a sort of interaction, as the regressions already con trol for industries and occupations. The coefficient of interest thus tests whether an earnings premium exists above that already accounted for by the underlying industries or occupations that are also included in the tradable services indicator. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 121 Such interactions are difficult to interpret. The descriptive statistics reveal that workers in tradable services have characteristics that by themselves are all associated with positive and significant coefficients in such regressions: they are disproportionately male, nonminority, and have higher educational attain ment. Combine these characteristics and you get a fairly hefty wage boost beyond that accounted for by any one characteristic alone. Are such workers truly more productive, or are there other factors, such as bargaining power and discrimination, that might explain their premium relative to those who lack this set of characteristics? The result is also curious in relation to the tradable service categorization. One might expect that the wages of such workers face downward pressure from international competition relative to the wages of other workers with similar skill sets in nontradable industries and occupations. At least in these static regressions, that is not the case. It will be interesting to track the premium over time to see if this pressure develops. At any rate, the important point is that service workers exposed to trade com petition have a lot to lose. The last section helps to quantify that point. This part of the paper includes two tables on changes in employment levels by industry and occupation. The goal here is to determine the extent to which job losses have occurred in recent years in tradable services, a question that is a bit of a holy grail, given the nervousness regarding the impact of offshoring services. As such, I thought the section got short shrift. This part of the analysis would have benefited from more discussion of the data and trying a little harder to separate out cyclical effects. On the first point, their sources for employment data are the Census Bureau's County Business Patterns and the BLS Occupational Employment Statistics (OES). Neither of these sources is typically used to track aggregate employment changes, and readers will legitimately wonder whether they reflect the stylized facts of employment trends over the years in question (1998-2003). In fact, given the difference in employment trends between the two surveys that are universally used for such analysis the BLS Establishment and Household surveys some will question whether the facts are "stylized" at all. I took a cursory look at the total OES employment counts from 2000 through 2003, which seem to show a large growth of jobs over these years, which is hard to square with data from more reliable sources of aggregate employment growth (such as the Establishment survey). Also, one of the biggest challenges regarding the question of the impact of offshoring on job loss over recent years is separating an offshoring effect from that of the cycle. This is particularly tough given the burst of the IT bubble in late This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 122 Brookings Trade Forum: 2005 2000 and the resulting spike in layoffs in this sector. In table 12, the authors ex amine changes from 1998-2002, a period including a strong run-up in employ ment growth (1998-2000) and a recession (2001) and jobless recovery (2002). At the least, the authors might consider breaking out these two periods to add some accounting for these cyclical effects. Better yet, given the caveats regard ing these data sets for this purpose and the difficulty untangling cycle from off shoring effects, they might want to be more cautious about their claims here. For example, claims comparing the employment growth of tradable and nontrad able services made it into their abstract and could be widely cited. There is also a claim here regarding employment losses at the lower end of the skill distribu tion in tradable services, but this change is essentially zero in table 12 and (if I calculated the standard error correctly) statistically insignificant (at the 5 percent level) in table 13.3 Displaced Workers in Tradable Services The final section of the paper uses the Displaced Workers Survey (DWS) to examine the extent to which being in a tradable job raises a worker's chance of displacement. Because of coding changes on industries and occupations, the authors cannot do comparisons across this biennial survey. But using the most recent survey, covering the years 2001-03, they find that those in tradable ser vices face significantly higher displacement rates than those in nontradable ser vices. For example, 31.7 percent of those in the tradable sectors of information services were laid off (not for cause) over these years, but only 7.5 percent of those in the nontradable sectors. Here again, the concern is that we are catching the cycle and the bursting of the tech bubble in the analysis, and thus not really isolating an offshoring effect. Information services includes both newspaper publishing (a nontradable service) and Internet publishing (a tradable service), and it is surely the case that a post bubble, large negative spike in domestic demand affected the former more than the latter. A simple difference-in-difference estimator might help to difference out the cycle, say using the changes in displacement in services that were nontradable. The problem is the introduction of new industry and occupation codes in the most recent DWS. However, the BLS has a version of the monthly CPS with 3.1 divided the standard deviation by the square root of the number of industries, both given in the table (0.111/3.16) for a standard error of 0.035, which returns a t statistic of -1.85. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms /. Bradford Jensen and Lori G. Kletzer 123 new sectoral codes starting in 2000, and although they could not track displace ments, the authors should see if these files might enable them to compare wage and employment changes in tradable and nontradable services controlling for the cycle. The DWS has long showed that among displaced workers who are reem ployed at the time of the survey, blue-collar production workers take the biggest hit in wages (the pay gap between their old and new jobs is above the average loss). But Jensen and Kletzer find negative effects of a similar magnitude for dis placed workers in tradable services. The difference between the old and new wage was, on average, about -30 percent for workers displaced from tradable jobs in both manufacturing and services, and about -14 percent for those dis placed from nontradable services. So workers in tradable services were more likely to be displaced during the recent downturn/jobless recovery, and for those who found new jobs at the time of the survey, these displacements were quite costly relative to nontradable services. Summary Faced with the question of how we identify service workers directly affected by offshoring, Jensen and Kletzer come up with an elegant solution: borrow the observation from the goods-producing literature that firms engaged in trade exhibit geographic concentration. While some might question how well this assumption travels across these different sectors, their results are, for the most part, intuitively satisfying and believable. This aspect of the paper makes a useful contribution to what has been a major stumbling block in this fledgling literature, namely, identifying affected workers in tradable services. The paper's other major contribution is its documentation of the characteristics of these workers, including their relative earnings. The paper has two shortcomings, both of which are evident in much work on offshoring. First, barring some attempt to control for cyclical effects, it is hard to know whether the job and wage loss effects they identify for workers in trad able services are due to their exposure to offshoring competition or to the pro tracted labor-market downturn over this period. While they get some traction in this argument by comparing tradable and nontradable services, the problem is that the negative cyclical demand shock was particularly acute in some of the same industries and occupations that have heavy weights in their tradable serv ice category (like IT). This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 124 Brookings Trade Forum: 2005 Second, from a policy perspective, economists need to look far beyond those directly affected by offshoring to grasp the magnitude of the challenge it poses. Compared to the number who are and will be affected in some way by the com petitive pressures from this form of trade, the number of workers who lose their jobs is surely very small. This by no means should lead us to give up on those who take the "direct hit" workers displaced by service trade. Their needs are often the most acute, and in this regard, ideas like wage insurance and expand ing Trade Adjustment Assistance are meritorious. But as Richard Freeman has discussed (this volume), the implicit supply shock from the introduction of millions of skilled workers into a relatively con centrated set of occupations and industries may have a significant impact on the wage structure of white-collar workers, just as the increase in trade in manufac turing goods has structurally altered the wage distribution of blue-collar work ers, partially contributing to the post-1979 increase in wage inequality and real wage losses, particularly for men. In this sense, Jensen and Kletzer may be overstating the breadth of their work by giving their piece the subtitle: "Understanding the Scope and Impact of Ser vices Offshoring." They get us a long way, further than any previous forays, toward identifying the most visible victims of offshoring: those who lose their jobs. But if Samuelson and others are right about the impact of competitive pres sures on the United States from trade with low-cost countries in sectors where we have held a comparative advantage, the scope and impact of offshoring could spill over far beyond those directly affected. Robert C. Feenstra: This is a good paper that introduces a new technique for classifying service industries as tradable and nontradable and then pursues a number of applications. The technique involves looking at the geographic con centration of service industries, using the idea that a more concentrated industry is most likely tradable. Geographic concentration is measured using population census data from the PUMS files, which also allow us to track individuals' occu pations as well as their industries of employment. So the paper not only intro duces a new technique for measure of the tradability of industries or occupa tions, it also shows how it can be implemented on a dataset that is novel for trade economists. I actually thought of using the geographic concentration of industries to measure something about trade some years ago, when reading a Scientific Amer ican article (Landy 1999) dealing with the distribution of stars in the universe. The "cosmological principle" states that the universe overall is homogeneous, so galaxies have no particular pattern. That is true on a very large scale, but on This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 125 smaller scales, galaxies form into clusters that are fractal: even as the scale of observation is reduced, the basic pattern of galaxies is the same. The extent to which galaxies cluster together can be measured by their spatial correlation. When reading that article I thought that the same should be true of the loca tion of economic activity: we could use spatial correlation or some other tech nique to measure the clustering of industries. That is exactly what the authors do here, using the Gini coefficient and a second measure of concentration. They find that the clustering or concentration of many service industries is just as strong as for manufacturing industries, implying that these service activities must be traded. While my reference to astronomy is just for fun, economists also use the con centration of industries to make conclusions about trade. Jean Imbs and Romain Wacziarg (2003) have shown, for example, that for developing countries the concentration of industries first falls and later increases as the countries mature, so the Gini coefficient follows a U-shaped pattern. For China, Alwyn Young (2000) found that after trade was opened the concentration of industries across provinces fell, which seemed to be contrary to comparative advantage, where we would expect regions to specialize. But later research found that industries in China later became more specialized across provinces, so the Gini coefficient also follows a U-shaped pattern in that country (see Naughton 2003; Poncent 2003). From these examples I conclude that using the concentration of industries to measure their trade orientation is well motivated and that the application to service industries is entirely new. Let us now consider the results of the paper. Using the Gini coefficients of geographic concentration, the authors divide industries into three groups: those with a Gini of less than 0.1 being the least concentrated, and therefore nontrad able; those with a Gini above or equal to 0.3 being the most concentrated, and therefore tradable, and those with a Gini between 0.1 and 0.3 in an intermediate category, but also treated as tradable. The classification of industries into these three groups is appealing: there are only a handful of nontraded manufacturing industries, including cement and concrete, whereas service industries are evenly divided between nontraded and traded activities. There are some anomalies, however: the education sector is very diversified geographically, so it is classi fied as nontradable, despite the fact that it is a principal service export of the United States. The geographic diversification of education holds for elementary and high schools, as well as colleges and universities (see Jensen and Kletzer's table 2), perhaps because of the land grant system in higher education. Because the authors use census data on individuals from the PUMS files, they can also distinguish tradable occupations as opposed to tradable industries. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 126 Brookings Trade Forum: 2005 That is, they can measure the geographic concentration of job titles rather than just industries. These job titles are unfamiliar to trade economists, so some fur ther explanation would be desirable. For example, occupational titles within the life, physical, and social sciences are mostly tradable; that is, these persons are geographically concentrated in their employment (see table 5). About half of these persons work in nontraded industries (such as education, which is not con centrated in space), and another half work in traded industries (see table 6). So at this point I could use some examples to understand the classifications: how can most of the employment in the life, physical, and social sciences be con centrated, when a significant number of these individuals work in education, which is not concentrated? In the next part of the paper, the authors investigate the characteristics of workers as classified by the tradability of their industry and occupation. Work ers in traded industries are more highly skilled and are paid more than in non traded industries, and this is especially true in traded service industries. The same is true for occupations: workers in tradable occupations earn more and have more education than those in nontradable occupations. Even if we strip out the effect of higher education, a wage premium persists for the traded industries, especially for traded service industries: these workers command a premium over and above their education level and demographic characteristics. The premium is about 6 percent for traded manufacturing and 15 percent for traded profes sional service industries. These results reminded me of two other related studies. First, Jeffrey Sachs and Howard Shatz (1998) made the point that services really are more skill intensive than manufacturing. The characterization of service jobs as flipping hamburgers is not true on average, where the jobs are more likely to be profes sional. Second, I was reminded of the earlier studies on the wage premiums in manufacturing by Larry Katz and Larry Summers (1989a, 1989b). They found that capital-intensive industries in manufacturing pay higher wages, and since these industries have higher exports, there is a wage premium in exporting. Trade economists were always squeamish about this finding, since it runs the risk of implying that being an exporter leads to paying higher wages, therefore suggesting that a subsidy to exports might help. On the contrary, most of us would believe that being more productive at the plant level leads to being an exporter and paying higher wages, with little or no role for export subsidies (see Fernandez 1989). The authors then investigate the growth across industries and occupations. In this I did not agree with the their expectations regarding which sectors would grow the most. For example, they state: "High-skill activities are consistent with This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 127 U.S. comparative advantage, and we would expect that as trade increases, eco nomic activity would shift to activities consistent with U.S. comparative advan tage. Thus we would expect higher-skill industries and occupations to have higher rates of employment growth." My difficulty with this logic is that it all depends on whether the United States is benefiting from increased export oppor tunities in the sectors where it has comparative advantage, or, on the contrary, whether it is facing new competition in those sectors. Paul Samuelson (2004) suggests that outsourcing could cause the United States to face competition in sectors where it formerly had comparative advantage. That is different from what Jensen and Kletzer have in mind. What they actually find is that service employment expanded during the period 1998-2003 and manufacturing employment contracted, and this shift holds regardless of whether one looks at traded or nontraded industries. So on the issue of employment growth, the methods developed in this paper to meas ure tradability just do not give us any extra explanatory power. We are back to the hypothesis advanced by James Harrigan and Rita Balaban (1999) and also by Bernardo Blum (2004): namely, that it is the rise in the service sector in the United States, combined with the skill-intensity ofthat sector (Sachs and Shatz 1998), that explains the rising relative wages of skilled workers. We still do not know whether this shift toward services comes from demand pressure, trade, productivity, or some other cause. It would have been nice if the tradability of service industries gave us extra insight on this issue, but that is not what the empirical results here show. In the final section of the paper the authors examine job loss and the charac teristics of displaced workers. This is an issue that Lori Kletzer has written on extensively, and the results here complement her earlier findings. Workers in tradable industries face a notably higher rate of job loss than those in nontrad ables. That is particularly true in service industries and in white-collar occupa tions. Nevertheless, it is still true that production workers in the United States have a higher rate of job loss than those in nonproduction and white-collar occu pations, including those occupations that we believe are being affected by ser vices outsourcing. General Discussion: Many participants commended the authors for their extremely creative and useful paper. The discussion also raised a variety of issues of interpretation and suggestions for further work, with some questioning how well domestic geographic concentration could capture international trad ability. Perhaps not surprisingly, a number of speakers found the results surpris ing for particular industries or occupations. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 128 Brookings Trade Forum: 2005 Lael Brainard highlighted two reasons why the authors' concentration index approach to identifying tradability was particularly valuable. First, it can be applied across occupations as well as across industries. Second, it gets around the problem that direct measurement is more difficult for services than for goods production. Internationalization of services essentially entails linking domestic and foreign factors of production so that work moves between product or proj ect teams. It is extremely difficult to quantify the value added from each step of the process. Brainard also wondered why the authors focused on a bivariate indi cator (whether something was tradable or nontradable) in their empirical analy ses instead of exploiting the continuous variable that they constructed. She and others saw their use of an essentially arbitrary threshold as throwing away poten tially useful information. The revised version of the paper does provide the actual indicators for major sectors and occupations. Some participants suggested that it would be helpful to compare the results of the tradability measure constructed here with other available alternatives. This would be one way to explore how well it captures what we mean by tradability. Brainard noted that we have direct tradability indicators for merchandise. She expected to find that some highly tradable goods, such as sugar, are not particu larly highly concentrated. Catherine Mann wondered whether the approach by Brad Jensen and Lori Kletzer had implications similar to the work by Frank Levy and Richard Murnane, which classifies tasks in terms of routinization. Susan Collins asked how similar it was to the classification by D sir e van Wel sum and Xavier Reif. The issue of comparability is partially addressed in the introduction to this volume. Robert Lawrence advanced another way to look at the paper, focusing on agglomeration. The results show that even inside the United States, where firms are free to set up everywhere, they often choose not to, presumably because of the benefits of locating near one another. Clearly, if costs were different enough abroad, they would choose to relocate. But it may be that the more concentrated firms are now, the greater the agglomeration benefits and the less likely they are to move. From this perspective, we should see their concentration as comforting, not threatening. Lawrence also stressed that one should not jump from tradabil ity in the sense of this paper to trade. For example, his work with Martin Baily finds considerable job loss in the computer industry, which is tradable. But their input-output table analysis concludes that this is overwhelmingly due to declines in domestic demand and that trade appears to have played a relatively minor role. Other participants elaborated on Robert Feenstra's point that domestic trad ability may be very different from international tradability. In particular, T. N. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms /. Bradford Jensen and Lori G. Kletzer 129 Srinivasan noted that if transactions and transportation costs are much less for domestic than for international trade, then domestic tradability has no implica tions for international tradability. Srinivasan also raised the point that often the same industry can produce using different technologies. For example, steel, which is certainly tradable, can be produced using both integrated mills that are quite concentrated and the more recent electronic processing mini-mills, which tend to be quite dispersed. He argued that it is important to consider technology in assessing whether concen tration provides a good indicator of tradability. He also pointed out that occupa tion, and perhaps to some degree industry, is a matter of choice. Thus he sug gested controlling for selection when estimating the earnings regressions. Catherine Mann noted that regulations can play a very important role in some service sectors. This includes legal bar exams, state-specific insurance regula tions, and others. There are also significant differences in cross-country regula tions. Thus it would be interesting to explore whether changes in state-specific regulations that make a particular industry more easily traded have affected its occupational stratification or its concentration indicators. Changes in rules for interstate banking are one especially interesting recent example. Mann also asked what the results in the paper could tell us about the risk ver sus expected return associated with particular occupations. Job loss is certainly very costly. However, her casual impression was that the empirical estimates find a relatively large wage premium for jobs in risky service industries and occupations, and it was worth exploring how this compared with the probability and expected costs of job loss. In contrast, manufacturing jobs are also risky but have been commanding a much smaller wage premium. Lawrence Mishel raised concerns about drawing conclusions from simply comparing employment growth in traded and nontraded industries (or occupa tions) within a given time period. Because employment trends may be quite dif ferent, he thought it important to develop a more convincing counterfactual that incorporates information about previous trend behavior. David Richardson suggested that it would be interesting to consider other concentration measures. For instance, the Ellison-Glaeser measure comes very close to an indicator of revealed comparative advantage. He also noted that the authors should be looking for both industries and occupations with very low concentration and those with very high concentration, because unusually low ratios for production to state GDP are also an indicator of (domestic) tradability. Collins noted that it might be helpful to distinguish between different types of services, and that the domestic concentration approach could be more This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 130 Brookings Trade Forum: 2005 appropriate for some types than for others. The General Agreement on Trade in Services (GATS) distinguishes among four modes by which services are traded. For example, mode 1 includes services supplied from one country to another, such as telephone calls, while mode 2 includes consumers who use a service in another country, such as tourists and students studying at a foreign university. It seemed to her that domestic concentration might be a better indi cator of tradability for mode 1 services than for mode 2. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 131 References Amiti, Mary, and Shang-Jin Wei. 2004. "Fear of Service Outsourcing: Is It Justified?" Working Paper WP/04/186. Washington: International Monetary Fund. Arora, Ashish, and Alfonso Gambardella. 2004. "The Globalization of the Software Industry: Perspectives and Opportunities for Developed and Developing Countries." Working Paper 10538. Cambridge, Mass.: National Bureau of Economic Research. Bardhan, Ashok Deo, and Cynthia A. Kroll. 2003. 'The New Wave of Outsourcing." Working Paper 1103. University of California, Berkeley: Fisher Center for Real Estate and Urban Economics. Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott. Forthcoming 2006. "Sur vival of the Best Fit: Exposure to Low Wage Countries and the (Uneven) Growth of U.S. Manufacturing Plants." Journal of International Economics 68 (1): 219-37. Bhagwati, Jagdish, Arvind Panagariya, and T. N. Srinivasan. 2004. "The Muddles over Outsourcing." Journal of Economic Perspectives 18 (Fall): 93-114. Blum, Bernardo S. 2004. "Trade, Technology and the Rise of the Service Sector: An Empirical Assessment of the Effects on U.S. Wage Inequality." University of Toronto (September). Brainard, Lael, and Robert E. Litan. 2004. "Offshoring Service Jobs: Bane or Boon and What To Do?" Brookings Policy Brief 132 (April). Bronfenbrenner, Kate, and Stephanie Luce. 2004. "The Changing Nature of Corporate Global Restructuring: The Impact of Production Shifts on Jobs in the U.S., China, and around the Globe " U.S.-China Economic and Security Review Commission. October. Dossani, Rafiq, and Martin Kenney. 2004. "The Next Wave of Globalization? Exploring the Relocation of Service Provision to India." Working Paper 156. Stanford Univer sity: Asia Pacific Research Center. -. 2003. "Went for Cost, Stayed for Quality? Moving the Back Office to India." Stanford University: Asia-Pacific Research Center. Duranton, Gilles, and Henry G Overman. 2004. 'Testing for Localisation Using Micro Geographic Data." London School of Economics. Ellison, Glenn and Edward L. Glaeser. 1999. "The Geographic Concentration of Indus try: Does Natural Advantage Explain Agglomeration?" American Economic Review, Papers and Proceedings 89 (May): 311-16. -. 1997. "Geographic Concentration of U.S. Manufacturing Industries: A Dart board Approach." Journal of Political Economy 105 (October): 889-927. F rber, Henry S. 2005. "What Do We Know about Job Loss in the United States? Evi dence from the Displaced Worker Survey, 1984-2004." Working Paper 498. Prince ton University: Industrial Relations Section. -. 2003. "Job Loss in the United States, 1981-2001." Working Paper 471. Prince ton University: Industrial Relations Section. -. 1993. "The Incidence and Costs of Job Loss: 1982-1991." BPEA, no. 1:73-132. Fernandez, Raquel, 1989. Comment on "Industrial Wage Differential, International Competition and Trade Policy," by Lawrence F. Katz and Lawrence H. Summers. In This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms 132 Brookings Trade Forum: 2005 Trade Policies for International Competitiveness, edited by Robert Feenstra, pp. 117-22. University of Chicago Press. Gardner, Jennifer M. 1993 "Recession Swells Count of Displaced Workers." Monthly Labor Review 116 (June): 14-23. Gartner Research. 2004. "Worldwide IT Services Market Forecast, 2002-2007" (Janu ary 12). Gentle, Chris. 2003. "The Cusp of a Revolution: How Offshoring Will Transform the Financial Services Industry." Deloitte Research. Harrigan, James, and Rita A. Balaban. 1999. "U.S. Wages in General Equilibrium: The Effects of Prices, Technology, and Factor Supplies, 1963-1991." Working Paper 6981. Cambridge, Mass.: National Bureau of Economic Research. Helpman, Elhanan, and Paul R. Krugman. 1985. Market Structure and Foreign Trade: Increasing Returns, Imperfect Competition, and the International Economy. Cam bridge, Mass.: MIT Press. Imbs, Jean, and Romain Wacziarg. 2003. "Stages of Diversification." American Eco nomic Review 93 (March): 63-86. Katz, Lawrence F., and Lawrence H. Summers. 1989a. "Can Interindustry Wage Differ entials Justify Strategic Trade Policy?" In Trade Policies for International Competi tiveness, edited by Robert Feenstra, pp. 85-116. University of Chicago Press. -. 1989b. "Industry Rents: Evidence and Implications." In Brookings Papers on Economic Activity: Microeconomics 1989, edited by Martin N. Baily and Clifford Winston, pp. 209-75. Brookings. Kirkegaard, Jacob F. 2004. "Outsourcing Stains on the White Collar?" Washington: Institute for International Economics (February). Kletzer, Lori G. 2001. Job Loss from Imports: Measuring the Costs. Washington: Insti tute for International Economics. -. 1998. "Job Displacement." Journal of Economic Perspectives 12 (Winter): 115-36. -. 1995. "White Collar Job Displacement, 1983-91." In Proceedings of the 47th Annual Meeting, pp. 98-107. Madison, Wis.: Industrial Relations Research Associa tion. Krugman, Paul R. 1991. Geography and Trade. Cambridge, Mass.: MIT Press. Landy, Stephen D. 1999. "Mapping the Universe." Scientific American 280 (June): 38-45. Mann, Catherine L. 2003. "Globalization of IT Services and White Collar Jobs: The Next Wave of Productivity Growth." Policy Brief 03-11. Washington: Institute for International Economics. McCarthy, John C. 2002. "3.3 Million U.S. Services Jobs To Go Offshore." San Fran cisco: TechStrategy Research, Forrester Research (November). Naughton, Barry J. 2003. "How Much Can Regional Integration Do to Unify China's Markets?" In How Far across the River? Chinese Policy Reform at the Millennium, edited by Nicholas Hope, Dennis Tao Yang, and Mu Yang Li, pp. 204-32. Stanford University Press. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms J. Bradford Jensen and Lori G. Kletzer 133 Podgursky, Michael. 1992. "The Industrial Structure of Job Displacement 1979-89 " Monthly Labor Review 115 (September): 17-25. Poncet, Sandra. 2003. "Measuring Chinese Domestic and International Integration." China Economic Review 14 (1): 1-21. Sachs, Jeffrey D., and Howard J. Shatz. 1998. "International Trade and Wage Inequality: Some New Results." In Imports, Exports, and the American Worker, edited by Susan M. Collins, pp. 215-40. Brookings. Samuelson, Paul A. 2004. "Where Ricardo and Mill Rebut and Confirm Arguments of Mainstream Economists Supporting Globalization." Journal of Economic Perspec tives 18 (Summer): 135-46. Schultze, Charles L. 2004. "Offshoring, Import Competition, and the Jobless Recovery." Brookings Policy Brief 136 (August). Tilton, Andrew. 2003. "Offshoring: Where Have All the Jobs Gone?" U.S. Economic Analyst 03/38 (September 19). Goldman Sachs & Co. Van Welsum, D sir e, and Graham Vickery. 2005. "Potential Offshoring of ICT Intensive Using Occupations." DSTI Information Economy Working Paper DSTI/ICCP/ffi(2004)19MNAL. Paris: Organization for Economic Cooperation and Development. Young, Alwyn. 2000. "The Razor's Edge: Distortions and Incremental Reform in the People's Republic of China." Quarterly Journal of Economics 115 (November): 1091-35. This content downloaded from 147.251.185.127 on Thu, 23 Mar 2017 10:39:29 UTC All use subject to http://about.jstor.org/terms