Investment Responses to Trade Liberalization: Evidence from U.S. Industries and Establishments

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. Investment Responses to Trade Liberalization: Evidence from U.S. Industries and Establishments Justin R. Pierce and Peter K. Schott 2017-120 Please cite this paper as: Pierce, Justin R., and Peter K. Schott (2017). “Investment Responses to Trade Liberalization: Evidence from U.S. Industries and Establishments,” Finance and Economics Discussion Series 2017-120. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2017.120. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

Investment Responses to Trade Liberalization: Evidence from U.S. Industries and Establishments∗ Justin R. Pierce† Board of Governors of the Federal Reserve System Peter K. Schott‡ Yale School of Management & NBER First Draft: October 2017 This Draft: November 2017 Abstract This paper examines the e(cid:27)ect of a change in U.S. trade policy on the domestic investment of U.S. manufacturers. Using a di(cid:27)erence-in-di(cid:27)erences identi(cid:28)cation strategy, we (cid:28)nd that industries more exposed to reductions in import tari(cid:27) uncertainty exhibit relative declines in investment after the change in trade policy. Within industries, we (cid:28)nd that this relationship is concentrated among establishments with low initial levels of labor productivity, capital intensity and skill intensity. For plants with high initial levels of skill intensity, we (cid:28)nd that increasedexposureisassociatedwitharelativeincreaseininvestment. Wealso(cid:28)nd evidence that establishments’ investment activity is smoother following the policychange. Keywords: Investment; Trade Policy; Manufacturing; China; Normal Trade Relations; MFN ∗Any opinions and conclusions expressed herein are those of the authors and do not necessarily represent the views of the U.S. Census Bureau, the Board of Governors or its research sta(cid:27). All results have been reviewed to ensure that no con(cid:28)dential information has been disclosed. We thank Dave Donaldson, Gordon Hanson, Colin Hottman, Stephen Redding and seminar audiences at the NBERandGeorgetownforsuggestions, andTravisAdamsandEmilyWisniewskiforhelpfulresearch assistance. †20th&CStreetsNW,Washington,DC20551,tel: (202)452-2980,email:justin.r.pierce@frb.gov. ‡165 Whitney Avenue, New Haven, CT 06511, tel: (203) 436-4260, email: peter.schott@yale.edu.

1 Introduction The U.S. manufacturing sector has undergone profound changes since the turn of the century, when a shift in U.S. trade policy increased import competition from China. While a range of studies link this trade liberalization to employment loss and establishment exit, less is known about the extent to which survivors adapt by investing in new technologies, changing their product or input mix, or otherwise di(cid:27)erentiating 1 themselves from their new competitors. In this paper, we examine how the domestic investment and capital stocks of U.S. manufacturing establishments respond to the October, 2000 U.S. granting of Permanent Normal Trade Relations to China (PNTR), a trade liberalization that removed the threat of substantial U.S. import tari(cid:27) increases on Chinese goods. By eliminating this cost uncertainty, PNTR provided U.S. producers with greater incentives to invest in (cid:28)nding Chinese suppliers, moving production from the United States to China, or otherwise increasing their competitiveness in the face of rising Chinese import competition. We use industry- and establishment-level data on domestic investment by U.S. 2 manufacturers to examine the latter channel. Our empirical analysis takes place in three steps. First, we examine the relationship between exposure to PNTR and both investment and capital stocks at the industry-level. Second, we use con(cid:28)dential U.S. Census Bureau microdata to examine how individual establishments adjust investment in response to PNTR, with a particular focus on heterogeneous responses along a broad range of establishment-level attributes. The industry analysis serves as an important benchmark for our subsequent analysis of establishments because the expected impact of trade liberalization at the establishment-level is ambiguous: some plants may shrink or exit, lowering investment, 3 while others may alter their production processes in ways that increase investment. Finally, motivated by models of investment under uncertainty, we investigate the timing, frequency, andlumpinessofestablishments’investmentbeforeandafterthechange 1Consider, for example, this anecdote from a recent article in the Wall Street Journal (Michaels (2017)), quoted in Fort et al. (2017): (cid:16)When Drew Greenblatt bought Marlin Steel Wire Products LLC, a small Baltimore maker of wire baskets for bagel shops, he knew nothing about robotics. That was 1998, and workers made products manually using 1950s equipment....Pushed near insolvency by Chinese competition in 2001, he started investing in automation. Since then, Marlin has spent $5.5 million on modern equipment. Its revenue, sta(cid:27) and wages have surged and it now exports to China and Mexico.(cid:17) 2In prior research (Pierce and Schott (2016)), we show that goods more exposed to PNTR exhibit substantial relative increases in U.S. imports from China as well as the number of U.S. (cid:28)rms that import from China, the number of Chinese (cid:28)rms that export to the United States, and the number of U.S.-Chinese (cid:28)rm pairs engaged in a trading relationship. One interpretation of these outcomes is that they are the manifestation of investment in trading relationships that was unleashed by the elimination of cost uncertainty. 3Pierce and Schott (2012) show that industries with greater exposure to PNTR exhibit relatively higher job destruction due to plant and (cid:28)rm exit and relatively lower job creation due to suppressed plant and (cid:28)rm entry. 1

in trade policy. Weemployageneralizeddi(cid:27)erences-in-di(cid:27)erences(DID)identi(cid:28)cationstrategythat estimates how investment and capital stocks change after the granting of PNTR for industries and establishments with varying levels of exposure. The baseline speci(cid:28)cation includes controls for other factors that may a(cid:27)ect investment in manufacturing during our sample period, including changes in Chinese trade policy that occur as part of China’s accession to the WTO (e.g. liberalization of export licensing), the phasing out of the global Multi-Fiber Arrangement governing quotas on developing-country textile and clothing exports, and changes in the relationship between investment and industry characteristics (cid:21) such as capital and skill intensity (cid:21) that may be correlated spuriously with the trade liberalization. At the industry-level, we (cid:28)nd that greater exposure to PNTR is associated with a relative decline in investment, and that the timing of the decline corresponds closely to PNTR’s implementation. We (cid:28)nd little evidence of such a response with respect to the capital stock, however, an outcome that may be due to the relatively slow response of capital stocks to changes in investment (cid:29)ows. Our establishment-level analysis focuses on the investment activity of continuing plants observed in the quinquennial U.S. Census of Manufactures (CM). We (cid:28)nd that plant-level responses to PNTR vary according to their pre-PNTR characteristics. Speci(cid:28)cally, while PNTR is associated with a relative decline in investment for the average plant, establishments with higher initial levels of labor productivity and capital intensity exhibit little to no relative decline, and plants with the highest levels of initial skill intensity exhibit a positive relationship between exposure to the trade liberalization and relative investment. The relatively high investment associated with skill- and capital intensive plants is consistent with U.S. comparative advantage. To assess the potential e(cid:27)ects of PNTR’s reduction of uncertainty on the frequency and smoothness of plants’ investment, we use data from the Annual Survey of Manufactures (ASM) to compute the average, standard deviation and share of years with positive investment across years leading up to and after the change in policy. Here, too, accounting for heterogeneity in establishment responses is important. In speci(cid:28)cations that control for plants’ initial characteristics, we (cid:28)nd that, for the average plant, larger reductions in tari(cid:27) rate uncertainty are associated with relative reductions in the standard deviation of investment, though these reductions are not present for plants with initially high levels of productivity. Our (cid:28)ndings make three contributions to the literature. First, they provide a broader view of the impact of trade liberalization on (cid:28)rms. While negative relationships between import competition and employment are well-known, especially for the period we are considering (Autor et al. (2013); Pierce and Schott (2016)), our (cid:28)nding that higher exposure to PNTR is associated with lower industry-level investment helps explain the reduction in overall manufacturing investment and the (cid:29)attening of the capital stock that has been noted elsewhere (Kurz et al. (2016)). Furthermore, to 2

the extent that depressed investment has dynamic e(cid:27)ects on employment, our (cid:28)ndings provide insight into the potential persistence of weak employment and earnings growth in the years following a liberalization. Second, our results highlight the potential heterogeneity of responses to trade liberalization across plants within similarly exposed industries, and identi(cid:28)es the characteristics of establishments (cid:21) including low labor productivity and high labor intensity (cid:21) that are associated with relatively greater declines in investment in the face of increased import competition. These results relate to the large literature studying the impact of competition on innovation and investment (e.g., Aghion et al. (2005)). In this respect our research is most closely related to Gutierrez and Philippon (2017)), which uses publicly available data from Compustat to show that increased exposure to PNTR induces relative increases in investment among (cid:16)leader(cid:17) (cid:28)rms, de(cid:28)ned as (cid:28)rms with high market to book value. Compared to that paper, our contribution is twofold. First, we consider the full population of manufacturing establishments, as opposed to the publicly traded (cid:28)rms present in Compustat. Second, we examine a wider range of (cid:28)rm attributes (cid:21) such as capital and skill intensity and productivity (cid:21) that capture other dimensions of (cid:16)leadership,(cid:17) i.e., consistency with U.S. comparative advantage. Furthermore, our (cid:28)ndings also add additional context to recent research (cid:28)nding evidence in favor of trade-induced technical change (Bloom et al. (2016)) and a negative relationship between import competition and innovation (Autor et al. (2016)) among (cid:28)rmsintheUKandUnitedStates, respectively, facingimportcompetitionfromChina. Finally, we contribute to the relatively small number of empirical studies associated with the large theoretical literature on investment under uncertainty (Pindyck (1993); Rob and Vettas (2003)). Finding plausibly exogenous shocks to uncertainty is an important challenge in in these studies and several papers, including Guiso and Parigi (1999), SchwartzandZozaya-Gorostiza(2003), andBloometal.(2007), haveestimated such shocks using surveys, cost data for speci(cid:28)c information technology investments, or detailed information from (cid:28)rms’ annual reports. Here, PNTR provides a large and plausibly exogenous shock to establishments’ cost uncertainty, and we identify 4 e(cid:27)ects on investment that are consistent with Bloom et al. (2007). Furthermore, our setting is particularly relevant in the current policy environment, where the 2016 U.S. Presidential election and the U.K.’s vote to exit the European Union have created considerable uncertainty about future tari(cid:27) rates. The paper proceeds as follows: Section 2 describes the data, Section 3 describes our empirical strategy and presents industry-level results. Section 4 presents the establishment-level analysis and and Section 5 concludes. 4For other studies on uncertainty in trade, see Handley (2014) and Handley and Limao (2017). 3

2 Data 2.1 Establishment- and Industry-Level Investment Data Establishment-level investment and capital stock data are drawn from the U.S. Census Bureau’s con(cid:28)dential Census of Manufactures (CM) and Annual Survey of Manufactures (ASM). In both cases, the Census Bureau asks manufacturing establishments to break down their capital expenditures into two categories (cid:21) structures and equipment (cid:21) as well as to report their total capital expenditures. The CM collects this information, as well as data on other establishment attributes, including employment, shipments and value added, on every U.S. manufacturing establishment (i.e., plant) quinquennially in years ending in two and seven. In all of our analyses using the CM, we follow standard practice in excluding all administrative records, i.e., observations for which most of the key variables of interest are imputed. Table 1 summarizes real investment (i.e., real capital expenditures) among U.S. 5 manufacturing establishments appearing in our regression sample below. Each row of the table reports results for a di(cid:27)erent Census year, while each panel focuses on a di(cid:27)erent measure of investment: average investment per plant, average investment as a shareofestablishments’capitalstock, andaverageshareofestablishmentswithpositive investment. In each case, we report (cid:28)gures for total investment and its components, investment in structures and investment in equipment. Table 1: PNTR and Industry-Level Capital Stock As indicated in the table, total investment averages 905 thousand dollars across establishments in 1992, versus 954 thousand dollars in 2007. As a share of the capital stock, these levels of investment range from 11 percent in 1992 to 16 percent in 2007. Furthermore, the table reveals that investment in equipment accounts for roughly 85 percent of total capital expenditures, with the remaining 15 percent accounted for by investment in structures. Finally, the table indicates that most plants invest in each Census year, with 87 percent of establishments reporting positive capital expenditures 5We discuss the de(cid:29)ators used in the computation of real investment below. 4

in 1992 and 90 percent reporting investment in 2007. Investments in equipment are much more common than investment in structures, with the latter occurring at 44 percent of establishments in 1992 and 54 percent of establishments in 2007. For the portion of our analysis where we investigate attributes of investment that must be estimated across time (cid:21) e.g., the standard deviation of investment or average investment per year (cid:21) we require higher-frequency data than is available in the CM. We therefore augment the CM data with annual data from the ASM. However, because the ASM collects information from only a subset of plants, we must restrict our analysis to the establishments that are surveyed in every year across our 1992 to 2007 sample 6 period. While this sample is restricted, these long-lived plants typically account for a disproportionately large share of activity in the manufacturing sector. Figure 1: Manufacturing Employment versus Real Investment Our industry-level analysis makes use of the publicly available NBER-CES Manufacturing Industry Database assembled by Becker et al. (2013), which can be downloaded from the NBER website. This dataset tracks many of the same outcomes contained in the CM and ASM across six-digit North American Industry Classi(cid:28)cation (NAICS) categories, including employment, nominal investment and the real capital stock, including separate data for real stocks of equipment and structures. We de(cid:29)ate thenominalinvestmentinboththesedataandtheCMandASMusingindustry-speci(cid:28)c 7 investment de(cid:29)ators contained in the database. Because investment is not broken out 6This restriction arises from changes in the sampling frame for the ASM that occur every (cid:28)ve years, which prevent tracking some plants consistently over time. Furthermore, while some plants are sampled with certainty in the ASM, the threshold used for selecting these (cid:16)certainty cases(cid:17) changed several times over the period we consider. 7Beckeretal.(2013)convertthenominalinformationontotalcapitalexpendituresforeachindustry 5

by equipment versus structures in the NBER-CES database, we construct this breakdown ourselves using publicly available versions of the Census of Manufactures (CM) 8 and Annual Survey of Manufactures (ASM) available on the Census Bureau’s website. Figure 1 shows that total real investment by U.S. manufacturing (cid:28)rms in equipment and structures rises faster than trend in the late 1990s before falling substantially in the early 2000s. Indeed, the decline in manufacturing investment from 1999 to 2003 is roughlyequaltothedeclineexperiencedduringthemuch-deeperGreatRecession. Asa result, the manufacturing real capital stock fell from 2003 to 2004, the (cid:28)rst time it had registered a decline since the data have been tracked (Kurz et al. (2016)). This decline can be seen in Figure 2, which also reveals that most of the increase in manufacturing capital stock since the 1970s is in equipment versus structures. Figure 2: Manufacturing Capital Stock 2.2 Industry and Firm Exposure to PNTR Our analysis makes use of a plausibly exogenous change in U.S. trade policy (cid:21) the U.S. granting of PNTR to China in October 2000 (cid:21) that e(cid:27)ectively liberalized U.S. collected in the CM and ASM into real expenditure data using investment de(cid:29)ators produced by the Federal Reserve Board. They then construct industry-level real capital stocks using a perpetual inventory equation in conjunction with depreciation rates for each industry also developed by the Federal Reserve Board. 8For instances in which data from the CM and ASM are available only at levels of aggregation higher than the six-digit NAICS industries used in our analysis, we employ industry shares developed by the Federal Reserve Board to allocate investment to six-digit NAICS industries. Further detail is available from the authors upon request. 6

imports from China. This impact can be understood by considering the two sets of tari(cid:27) rates that comprise the U.S. tari(cid:27) schedule. The (cid:28)rst set of tari(cid:27)s, known as NTR tari(cid:27)s, are generally low and are applied to goods imported from other members of the World Trade Organization (WTO). The second, known as non-NTR tari(cid:27)s, were set by the Smoot-Hawley Tari(cid:27) Act of 1930 and are often substantially higher than the corresponding NTR rates. Imports from non-market economies such as China generally are subject to the higher non-NTR rates, but U.S. law allows the President to grant these countries access to NTR rates on a year-by-year basis, with the President’s decision subject to potential overruling by Congress. U.S. Presidents granted China such a waiver every year starting in 1980, but Congressional votes over annual renewal became politically contentious and less certain of passage following the Chinese government’s crackdown on Tiananmen Square protests in 1989 and other (cid:29)ashpoints in U.S.-China relations during the 1990s such as China’s transfer of missile technology to Pakistan in 1993 and the Taiwan Straits Missile Crisis in 1996. Uncertainty over China’s access to NTR tari(cid:27) rates ended with Congress passing a bill granting PNTR status to China in October, 2000, which formally took e(cid:27)ect upon China’s entry into the WTO in December 2001. Figure 3: Distribution of Industry-Level NTR Gaps We follow Pierce and Schott (2016) in measuring the impact of PNTR as the rise in U.S. tari(cid:27)s on Chinese goods that would have occurred in the event of a failed annual renewal of China’s NTR status prior to PNTR, NTRGap = NonNTRRate −NTRRate . (1) j j j 7

We refer to this di(cid:27)erence as the NTR gap, and compute it for each SIC industry j using ad valorem equivalent tari(cid:27) rates provided by Feenstra et al. (2002) for 1999, the year before passage of PNTR. As indicated in Figure 3, which reports the distribution of NTR gaps across six-digit NAICS industries, NTR gaps vary widely, with a mean and standard deviation of 30 and 14 percentage points, with an interquartile range of 0.21 to 0.40. Analysis of the underlying NTR and non-NTR rates in Pierce and Schott (2016) reveals that seventy-nine percent of the variation in the NTR gap across industries is due to variation in non-NTR rates, set 70 years prior to passage of PNTR. This feature of non-NTR rates e(cid:27)ectively rules out reverse causality that would arise if non-NTR rates were set to protect industries with declining employment or surging imports. Furthermore, to the extent that NTR rates were set to protect industries with declining employment prior to PNTR, these higher NTR rates would result in lower NTR gaps, biasing our results away from (cid:28)nding an e(cid:27)ect of PNTR. 2.3 Other Policy Variables Our empirical analysis includes controls for a wide range of additional factors that may a(cid:27)ect U.S. manufacturing investment. First, we allow for the possibility that the relationship between certain industry-level characteristics and investment may have changed around the time of PNTR’s passage. For example, a decline in the competitiveness of labor intensive industries in the United States or the decline of unions may have disproportionately a(cid:27)ected certain industries. We control for these explanations by including interactions of a post-PNTR indicator with initial values of industry capital and skill intensity and the industry-level share of union membership in 1990 (Hirsch and Macpherson (2013)). We also control for changes in Chinese domestic and trade policies related to its accession to the WTO. These changes include reductions in export licensing requirements, production subsidies and import tari(cid:27) rates. Our controls draw on data from work on export licensing requirements by Bai et al. (2015), on production subsidies from Khandelwal et al. (2013), and on Chinese import tari(cid:27) rates from Brandt et al. (2017). To account for the fact that reductions in barriers to foreign investment in China also declined at this time, we control for the share of industry inputs requiring relationship-speci(cid:28)city from Nunn (2007). Finally, we control for other policy and macroeconomic shifts occurring in the U.S. around 2000. The (cid:28)rst of these changes is the bursting of the 1990s tech bubble, which we control for with the interaction of the post-PNTR indicator with an indicator for whether the industry is engaged in the production of advanced technology products, as de(cid:28)nedbytheInternationalTradeCommission. Inaddition, wecontrolfortheelimination of quotas associated with the phasing out of the global Multi-Fiber Arrangement (Khandelwal et al. (2013)). 8

3 PNTR and Industry-Level Investment As indicated in the large literature on the impact of competition on innovation and investment (e.g., Aghion et al. (2005)), the relationship between PNTR and investment istheoreticallyambiguous. SomeestablishmentsmightstepupinvestmentintheirU.S. operations in an e(cid:27)ort to increase competitiveness vis a vis rising imports, while others might choose to exit the market, or cease domestic production in favor of production abroad. In this section we set the stage for our establishment-level analysis below by examining the aggregate impact of these decisions on investment at the industry level. Our baseline di(cid:27)erence-in-di(cid:27)erences (DID) speci(cid:28)cation examines whether industries with higher NTR gaps ((cid:28)rst di(cid:27)erence) experience di(cid:27)erential changes in investment after the change in U.S. trade policy (second di(cid:27)erence) versus before, y = θPostPNTR × NTRGap + (2) jt t j βX +γPostPNTR ×X + jt t j δ +δ +ε . j t jt The sample period for the industry analysis is 1990 to 2007. The dependent variable, y , represents an outcome in industry j, for example log investment. The (cid:28)rst term jt on the right hand side is the DID term of interest, an interaction of the NTR gap and an indicator for the post-PNTR period, i.e., years from 2001 forward. The second term on the right-hand side of equation 2 captures the impact of time-varying industry characteristics, such as exposure to MFA quota reductions and the NTR tari(cid:27) rate. The third term on the right hand side is an interaction of the post-PNTR dummy variable and time-invariant industry characteristics, such as initial industry capital and skill intensity, the degree to which industries encompass high-technology products and the extent of initial union membership in the industry. These interactions allow for the possibility that the relationship between employment and these characteristics changes in the post-PNTR period in ways that might spuriously be related to the trade liberalization. δ , δ and α represent industry and year (cid:28)xed e(cid:27)ects and the constant. j t An attractive feature of this DID identi(cid:28)cation strategy is its ability to isolate the role of the change in U.S. trade policy. While industries with high and low NTR gaps are not identical, comparing outcomes within industries over time isolates the di(cid:27)erential impact of China’s change in NTR status. The (cid:28)rst three columns of Table 2 report results for our main variable of interest, total investment, with standard errors clustered at the industry level. The (cid:28)rst column reports a speci(cid:28)cation with only the DID term, the second column adds interactions of the post-PNTR indicator with industry capital intensity and skill intensity, and the third column includes the full set of controls described in Section 2.3. As indicated in 9

the table, we (cid:28)nd negative and statistically signi(cid:28)cant coe(cid:30)cients on the DID term in all three cases. We assess the economic signi(cid:28)cance of the estimated DID coe(cid:30)cients in terms of the e(cid:27)ect on the dependent variable of an interquartile shift in an industry’s NTRgap(from0.214to0.402). Thesecoe(cid:30)cientsindicatethataninterquartileshiftin industry exposure to PNTR is associated with relative declines in investment of -0.131 to -0.168 log points (e.g., -0.696*0.188 to -1.017*0.188). The next six columns report analogous results for investment in structures and equipment. We (cid:28)nd a negative and signi(cid:28)cant relationship between exposure to PNTR and each of these types of investment in all three speci(cid:28)cations. Table 2: PNTR and Industry-Level Investment For the decline in investment to be attributable to PNTR, the NTR gap should be correlated with investment after PNTR, but not before. To determine whether there is a relationship between these variables in the years before 2001, we replace the PostPNTR indicator used in equation 2 with interactions of the NTR Gap and the full set of year dummies, 10

2007 (cid:88) y = (θ 1{y = t}×NTRGap )+ (3) jt y j y=1991 2007 (cid:88) (λ 1{y = t}×X )+ y j y=1991 βX +δ +δ +α+ε . jt j t it Here, we estimate equation 3 with the full set of controls noted above. Results are reported visually in Figure 4. The left panel of this (cid:28)gure traces out the impact on investment of an interquartile shift in industry exposure to PNTR implied by the estimated di(cid:27)erence-in-di(cid:27)erences coe(cid:30)cients θ . For comparison, a similar (cid:28)gure for y employment is provided in the right panel. As indicated in the left panel of the (cid:28)gure, point estimates are statistically insigni(cid:28)cant at conventional levels until after 2001, at which time they generally become negative and statistically signi(cid:28)cant. This pattern is consistent with the parallel trends assumption inherent in our di(cid:27)erence-in-di(cid:27)erences 9 analysis, lending further support for our empirical strategy. Figure 4: Implied Industry-Level Impact of PNTR Next, we examine whether the decline in investment associated with PNTR in Table 2 is also apparent in capital stocks. As indicated in Table 3, we (cid:28)nd that higher exposure to PNTR is associated with positive but statistically insigni(cid:28)cant changes in the capital stock for total capital (column 1) and equipment capital (column 2) and negative and statistically insigni(cid:28)cant changes in structures capital (column 3). One 9To further evaluate the role of the policy change versus other factors, Pierce and Schott (2016) comparestherelationshipbetweentheNTRgapandmanufacturingemploymentintheU.S.withthat in the EU, which was not subject to a similar policy change. They (cid:28)nd that the negative relationship between exposure to PNTR and manufacturing employment is only present in the United States. 11

potential reason for the lack of a relationship between PNTR and the capital stock, unlike for investment, is that the capital stock adjusts slowly to changes in investment (cid:29)ows. Table 3: PNTR and Industry-Level Capital Stock Aninterestingquestionforfurtherstudyiswhethertherelativeweakeningininvestment may help explain the persistence of the reduction in manufacturing employment associated with PNTR (Pierce and Schott (2016)). That is, while increases in investment may lead to subsequent rebounds in employment, declines in investment driven 10 by establishment exit may have a long-run dampening e(cid:27)ect on job creation. Finally, we examine the relationship between the change in U.S. trade policy and investment de(cid:29)ators using equation 3. To conserve space, we summarize the results of this regression in Figure 5, which plots the 90 percent con(cid:28)dence interval associated with an interquartile shift in industry exposure to PNTR. As indicated in the (cid:28)gure, we (cid:28)nd that industries with greater exposure to PNTR exhibit greater declines in 10Fortetal.(2017)showthatmorethaneightypercentofthedeclineinmanufacturingemployment between 1977 and 2012 is due to net establishment death. 12

their investment de(cid:29)ators. One potential explanation for this relationship relates to plant closures. Pierce and Schott (2016), for example, (cid:28)nd that plants with greater exposure to the change in policy are relatively more likely to shut down. If these closures increase the supply (on the secondary market) of the capital goods used to produce in that industry, their price might fall. Figure 5: PNTR and Industry Investment De(cid:29)ators 4 PNTR and Establishment-Level Investment In this section, we exploit the plant-level data available in the CM and ASM to determine the extent to which di(cid:27)erent plants in the same industry might vary in their responsetoPNTR.Plant-levelanalysisalsopermitsexaminationoftheextenttowhich changes in investment are driven by adjustments in average investment per year versus changes in the (cid:16)lumpiness(cid:17) of investment. We (cid:28)nd that in both cases, larger, more productive and more skill- and capital-intensive plants are less likely to adjust their investment activity in response to the change in U.S. trade policy. 4.1 Baseline Plant-level Estimates We begin by examining the average investment responses of plants to PNTR without including terms that might account for within-industry heterogeneity. We use data from the CM, which covers the population of manufacturing establishments and is 13

available every (cid:28)ve years. Our sample is composed of observations from the 1992, 1997, 2002 and 2007 CMs and this baseline speci(cid:28)cation is as follows, y = θPostPNTR × NTRGap + (4) pt t j γPostPNTR ×X +βX + t j jt +δ +δ +α+ε . p t pt where p indexes establishments, j indexes industries and t indexes years. The dependent variable is one of three real investment shares (cid:21) total investment (i.e., total capital expenditures), investment in equipment, or investment in structures, where each is divided by the establishment’s capital stock (cid:21) or the log value of the capital stock. The (cid:28)rst term on the right-hand side is the DID term representing the e(cid:27)ect of PNTR, and it consists of the interaction of a PostPNTR indicator and the time-invariant t NTRGap . The next two terms represent the additional control variables used in j Equation2. The remaining terms represent plant and year (cid:28)xed e(cid:27)ects. Note that this speci(cid:28)cation yields within-plant estimates of the relationship between exposure to PNTR and capital expenditures, but does not account for changes in investment driven by establishment entry and exit. The (cid:28)rst two columns of Table 4 report the results of estimating equation 4, (cid:28)rst with only the DID term of interest and the (cid:28)xed e(cid:27)ects required for its identi(cid:28)cation (column 1), and then with the full set of covariates (column 2). We (cid:28)nd that while the relationship between exposure to PNTR and total investment is negative, as in the industry-level estimates discussed above, it is not statistically signi(cid:28)cant at conventional levels. The next six columns indicate similar negative but statistically insigni(cid:28)cant relationships for the two broad categories of investment shares (cid:21) equipment and structures (cid:21) as well as for the log real book value of capital. The overall message of Table 4 is that the relationship between exposure to PNTR and investment within continuing plants is negative but not precisely estimated. We note that establishment-level results in Table 4 di(cid:27)er from the industry-level 11 estimates in Table 2 for two reasons. First, the samples are di(cid:27)erent. Second, the dependent variables are not the same: the industry-level analysis examines variation in log investment, while the plant-level regressions investigate investment shares because 12 establishment-level investment can be zero in some years. 11Thesamplesaredi(cid:27)erentinthreeways. First,inclusionofplant(cid:28)xede(cid:27)ectsintheestablishmentlevelanalysismeansthatidenti(cid:28)cationisrestrictedtowithin-establishmentvariation,therebyexcludingtheimpactofestablishmentexit,whichissubstantialduringthisperiod(Fortetal.(2017);Pierce and Schott (2012)). Second, these (cid:28)xed e(cid:27)ects also exclude plants that are not present in both the pre- and post-PNTR periods. Finally, the establishment-level regressions are restricted to Census years (cid:21) 1992, 1997, 2002 and 2007 (cid:21) while the industry-level sample includes data for every year from 1990 to 2007. 12Further discussion of the di(cid:27)erences between the industry- and plant-level results in appendix 14

Table 4: PNTR and Establishment-Level Investment 4.2 Plant-level Results Allowing for Heterogeneous Responses To assess the importance of plant heterogeneity within industries, we augment Equation 4 with an additional covariate that interacts the DID term with one of several normalized initial plant attributes: plant size, as measured by employment or value added; plant productivity, as measured by TFP, value added (VA) per worker or shipments per worker; and plant capital and skill intensity, y = θ PostPNTR × NTRGap + (5) pt 1 t j θ PostPNTR × NTRGap × EstabChar + 2 t j p ϕ PostPNTR × EstabChar +βX +γPostPNTR ×X + 1 t p jt t j δ +δ +α+ε p t pt section B. 15

These terms, which we refer to as (cid:16)plant heterogeneity terms(cid:17) and which we include one-at-a-time in separate regressions, appear as the triple interaction in the second line ofequation5. Thenormalizationdividesthe1992plantattributebytheaverageofthat attribute across all plants in the same industry in 1992, therefore explicitly accounting 13 for heterogeneity within industries, rather than di(cid:27)erences across industries. The third term in equation 5 represents the interaction of the plant heterogeneity term with the PostPNTR indicator required to identify the triple interaction. We do not t simultaneously include all plant heterogeneity terms in a single regression given their 14 high correlation. Estimates are reported in Table 5. To conserve space we restrict our reporting to the equipment investment share, as that form of investment accounts for the majority of total investment, and suppress the estimates for the non-DID terms. As indicated in the table, a key di(cid:27)erence between these results and those reported in Table 4 is the statistical signi(cid:28)cance of the main PNTR DID term in all columns of the (cid:28)rst row exceptthesecond(valueadded). Theplantheterogeneitytermsthemselves,reportedin the remaining rows of the table, are positive in all columns and statistically signi(cid:28)cant in the (cid:28)nal four, i.e., for both measures of labor productivity and for capital and skill intensity. These positive coe(cid:30)cients indicate that, for a given level of exposure to PNTR, plants with higher values of these attributes exhibit relatively higher levels of equipment investment after the change in trade policy, relative to those with lower values of the attributes. Asnotedinthebottompanelofthetable, aninterquartileshiftinindustryexposure to PNTR implies relative reductions in the equipment investment share of -0.004 (- 0.0229*0.188+0.0022*0.188*1.1176)inthe(cid:28)rstcolumnofthetable(whereemployment is the plant heterogeneity attribute), to -0.0014 in the last column (skill intensity). In each case, these impacts are evaluated at the mean level of the noted attribute across plants in 1992. These changes represent -1.1 to -3.2 percent of the mean equipment investment share in 1997, the prior year closest to the change in trade policy. The DID coe(cid:30)cients from these and analogous regressions for total and structures investment are summarized visually in Figures 6 and 7. The (cid:28)rst of these (cid:28)gures plots the main DID coe(cid:30)cient across speci(cid:28)cations, with each panel of the (cid:28)gure reporting theninetypercentcon(cid:28)denceintervalforthesevenseparateregressionsthatcorrespond to the inclusion of the seven plant heterogeneity attributes we examine. The bottom 13Given the (cid:28)xed e(cid:27)ects, plants are included in the regression only if they span 1997 and 2002. For plants that are not present in 1992, we divide their 1997 attribute by the relevant industry attribute in 1992. 14Onepotentialconcernwithourregressionspeci(cid:28)cationisthatplantswithdi(cid:27)erentvaluesofinitial characteristicsmayhavebeenondi(cid:27)erenttrendspriortothePNTR,andthatthesetrendsmaydrive the relationship between the policy change and the plant heterogeneity terms. In unreported but available results, we control for this possibility by augmenting equation 5 with interactions of year dummiesandtheplantcharacteristicbeingexamined. Resultsarevirtuallyidenticaltothosedescribed below. 16

Table 5: PNTR and Heterogeneity in Establishment-Level Equipment Investment right panel of this (cid:28)gure corresponds to the (cid:28)rst row of Table 5. As noted in the (cid:28)gure, the main DID terms are negative for log capital (upper left panel) and for the total (upper right panel) and equipment investment (lower right panel) shares, but generally statistically signi(cid:28)cant only for the latter two. Results for the structures investment share (lower left panel) are typically statistically insigni(cid:28)cant, with the exception of the skill intensity attribute regression, where the relationship is negative and signi(cid:28)cant. Figures7providesasimilarplotfortheplantheterogeneityDIDterms. Inthiscase, the bottom right panel reports the results from the second row of Table 5. As with the main DID terms, relationships between the change in trade policy and outcomes are stronger for the equipment and total investment shares than they are for structures investment and the capital stock. Finally, Figure 8 uses the coe(cid:30)cients reported in Figures 6 and 7, as well as information about the distribution of plants’ attributes in 1992, to quantify how the economic impact of PNTR varies across establishments with di(cid:27)erent levels of a particular attribute. In the (cid:28)gure, each pair of bars is computed using coe(cid:30)cient estimates 17

Figure 6: Main DID Coe(cid:30)cients for CM Regressions from the separate regressions described above (e.g., one of the columns of Table 5). Each bar is an evaluation of the impact of an interquartile shift in exposure to PNTR for a plant with a (cid:16)low(cid:17) versus (cid:16)high(cid:17) level of the noted attribute. We de(cid:28)ne the low level of an attribute as the mean less one standard deviation and the high level as the mean plus one standard deviation. We report results for total, structures and equipment investment shares. Economic impacts are expressed as the implied change in the investment share as a percentage of the mean investment share across plants in 1997, the prior year closest to the change in trade policy. Asindicatedinthe(cid:28)gure, we(cid:28)ndthattheimpactofgreaterexposureonlow-versus high-attribute establishments varies by attribute but that it is similar across the three types of investment. Our discussion here focuses on total and equipment investment shares given the stronger relationship between exposure to PNTR and these outcomes found above. Overall,we(cid:28)ndthatplantswithhighvaluesoftheattributesweexamineexperience smaller relative reductions in equipment investment following PNTR. This di(cid:27)erence is most noticeable for labor productivity, capital intensity, and skill intensity. For plant size and TFP, the di(cid:27)erences between low- and high-attribute plants are relatively small. AsecondnotablefeatureofFigure8isthatplantswiththehighestskillintensity actually experience a relative increase equipment investment after the change in trade policy versus before. 18

Figure 7: Plant Heterogeneity DID Coe(cid:30)cients for CM Regressions Combined, the results in this section suggest that the average continuing establishment reduces equipment (and total) investment in response to PNTR relative to the period before the change in trade policy. However, for the subset of plants with relatively high skill intensity, greater exposure to the change in trade policy is associated with relatively higher equipment (and total) investment. This increased investment could represent trade-induced technological change of the type discussed in Bloom et al. (2016). Alternatively, it could re(cid:29)ect capital expenditures used to upgrade product quality (Schott (2003, 2004)) or switch production (Bernard et al. (2006, 2011); Khandelwal (2010)) towards goods more in line with U.S. comparative advantage. Our results in this section are consistent with those reported in Gutierrez and Philippon (2017), who (cid:28)nd that while investment in property, plant and equipment is relatively lower for publicly traded (cid:28)rms after PNTR versus before, the relative decline is less substantial for (cid:16)leader(cid:17) (cid:28)rms, de(cid:28)ned as (cid:28)rms whose market-to-book value of capital is above the median. 4.3 Responses in the Timing and Frequency of Investment PNTR’s elimination of the risk of potential tari(cid:27) increases o(cid:27)ers a unique setting for examining how uncertainty a(cid:27)ects establishments’ investment behavior in ways beyond those explored above, in particular its timing and frequency. Bloom et al. (2007), 19

Figure 8: Implied Impact of PNTR With Plant Heterogeneity From CM Regressions for example, show theoretically that greater uncertainty lowers the responsiveness of (cid:28)rms’ investment to demand shocks, provided that investments are at least partially irreversible. In particular, because uncertainty drives a wedge between the marginal products of capital required for investment and disinvestment, it increases the (cid:16)zone of inaction,(cid:17) rendering it lumpier. In this section, we use the ASM to examine how the 15 timing and frequency of investment respond to PNTR. Asnotedabove,theASMhastwodrawbacksrelativetotheCM:itisasurveyrather than a census; and the survey sample is re-drawn every (cid:28)ve years, complicating one’s ability to track individual plants for a long period of time. Given these limitations, our analysis is limited to the balanced panel of plants present in every ASM from 1990 to 2007. This selected sample clearly di(cid:27)ers from the general population, as the plants included are larger, older and more likely to be exporters. We relate the patterns of establishments’ investment to the change in trade policy by collapsing the balanced panel into two periods: a pre-PNTR period encompassing the years 1990 to 2000; and a post-PNTR period comprising 2001 to 2007. A virtue of this sample interval, in addition to its spanning the passage of PNTR, is that each sub-period roughly coincides with a full business cycle, beginning around the time of a recession peak and continuing through the start of the next recession. For each period, we calculate three plant-level measures of investment activity. The (cid:28)rst measure is the average size of establishments’ investments, de(cid:28)ned as the sum of plant p’s investment for period c, divided by the number of years in the period. This measure provides a useful comparison to our results above. The second measure is the standard deviation of the level of investment, within plants, across the years in each period. Thismeasurecaptureschangesinthelumpinessofplants’investmentbehavior. 15Empirically, Bloom et al. (2007) show that publicly traded UK (cid:28)rms’ investment is negatively associated to the standard deviation of their stock returns, a potential manifestation of demand uncertainty. 20

The third measure is the share of years in each period with positive investment, a measure that captures the frequency with which establishments invest. In practice, as noted in Table 1, a high share of establishments invest each year, though the share is lower for structures investment. With these measures, we estimate the following equation: ln(y ) = θPostPNTR × NTRGap + (6) pc c j βX +γPostPNTR ×X + jc c j δ +δ +α+ε , p c pc where p indexes establishments and j indexes industries, as before, and c indexes the two time periods. The dependent variable ln(y ) is the log of one of the three pc measures of investment behavior for plant p in period c noted above, and the DID term and control variables are identical to those in Equation 2, with the exception that in equation 6, time-varying control variables are averaged over each period. Table 6: PNTR and the Lumpiness of Plant Investment 21

As in Section 4.1, Table 6 sets a baseline by reporting coe(cid:30)cient estimates and standard errors from estimating equation 6 without controls for plant heterogeneity. The (cid:28)rst three columns of the table display results for total investment, the next three columns for investment in structures, and the (cid:28)nal three columns for investment in equipment. The results indicate that higher industry-level exposure to PNTR (cid:21) and therefore a larger reduction in tari(cid:27) rate uncertainty (cid:21) is associated with smaller average investment sizes, a smaller standard deviation of investment across years, and a higher share of years with positive investment. Though the signs for the latter two variables generally are in line with the predictions from Bloom, Bond, and Van Reenen (2007), in that larger reductions in uncertainty lead to investments that are less lumpy (standard deviation) and more frequent (share positive), only one of these relationships (cid:21) the standard deviation of equipment investment (cid:21) is statistically signi(cid:28)cant at conventional levels. One potential explanation for the lack of signi(cid:28)cance may be that our data are at a relatively infrequent annual frequency, thereby masking variation in the timing of investments within calendar years. Figure 9: Main DID Coe(cid:30)cients for ASM Regressions As above, when we augment equation 6 with plant heterogeneity terms, these relationships become clearer. Results from this augmented speci(cid:28)cation are displayed in Figures 9 and 10, which take the same format as Figures 66 and 7. The (cid:28)gures display results only for equipment investment, as this is the category of investment that displays the strongest relationship with PNTR, as in Section 4.1. Analogous (cid:28)gures for total investment and investment in structures are unreported, but available. Figure 9 displays 90 percent con(cid:28)dence intervals for the coe(cid:30)cient estimates on the main DID term. The left panel of this (cid:28)gure reveals that, for the average establishment, higher exposure to PNTR is associated with a relative reduction in the standard deviation of equipment investment, but that these relationships are statistically signi(cid:28)cant at conventional levels only for the speci(cid:28)cations that examine heterogeneity in productivity. A relative decline in lumpiness is also manifest in the share of years with positive investment (right panel), but only for the speci(cid:28)cation that uses TFP to 22

Figure 10: Plant Heterogeneity DID Coe(cid:30)cients for ASM Regressions capture heterogeneity. The main DID terms for average investment (middle panel) are statistically insigni(cid:28)cant at conventional levels. This di(cid:27)erence compared to the results for the CM, above, may re(cid:29)ect the select group of (cid:28)rms present in the balanced ASM sample. Figure 10 displays 90 percent con(cid:28)dence intervals for the plant heterogeneity DID terms. As shown in the (cid:28)gure, plant heterogeneity DID terms are positive and statistically signi(cid:28)cant for the standard deviation of equipment investment (left panel) for the three productivity terms. Likewise, among the plant heterogeneity DID terms for the share of years with positive investment (right panel), the estimate is positive and statistically signi(cid:28)cant at conventional levels for TFP. Figure 11: Implied Impact of PNTR With Plant Heterogeneity from ASM Regressions As in the previous section, we combine the estimates plotted in in Figures 9 and 10 with information no the distribution of plants’ initial attributes to compare the impact of an interquartile shift in exposure to PNTR among plants with low and high values of each attribute. These results are reported in Figure 11. As indicated in the left panel of the (cid:28)gure, plants with low initial values of labor productivity exhibit a decline 23

in the standard deviation of equipment investment of approximately 2 percent relative to their 1997 levels. Plants with high labor productivity, by contrast, exhibit hardly any change relative to the period before the change in U.S. trade policy. One potential explanation of the results in this section can be found by considering theexpectationsofhigh-productivityestablishmentspriortopassageofPNTR.Ifthese establishments viewed their productivity level as being su(cid:30)ciently high to continue operating even if tari(cid:27)s increased substantially, their investment activity may not have been suppressed by uncertainty in the pre-PNTR period. As a result, less of a response to the timing of these establishments’ investment following passage of PNTR might be expected. 5 Conclusion This paper estimates the investment responses of U.S. manufacturing industries and establishments to the elimination of tari(cid:27) rate uncertainty associated with the U.S. granting of PNTR to China in October 2000. We use a di(cid:27)erences-in-di(cid:27)erences approach to examine how variation in exposure to PNTR is associated with changes in manufacturing investment and capital stock after the policy change, relative to before. At the industry-level, we (cid:28)nd that industries more exposed to PNTR experience relative declines in manufacturing investment, both for equipment and structures, and thatmore-exposedindustriesexperiencestatisticallyinsigni(cid:28)cantdeclinesinthecapital stock. Examining a (cid:29)exible speci(cid:28)cation that makes no assumptions about the timing of the e(cid:27)ects of PNTR, we (cid:28)nd that the decline in investment lines up closely with the timing of the granting of PNTR. At the establishment-level, we (cid:28)nd that there is heterogeneity within industries in termsofhowestablishmentsrespondtoPNTR’stradeliberalization. Whiletheaverage e(cid:27)ect of PNTR is to lower investment, for establishments with higher initial levels of laborproductivityandofcapitalandskilledlaborintensity, higherexposuretoPNTR’s trade liberalization is associated with increases in investment. These within-plant increases in investment are consistent with trade-induced technical change, productupgrading, orotheractivitiesthatdi(cid:27)erentiateU.S.productionfromimport-competing products. Examining the timing, frequency, and lumpiness of establishments’ investment behavior, we (cid:28)nd that larger reductions in uncertainty associated with PNTR are associated declines in the lumpiness of investment, though there is less of a change in behavior for establishments with high initial productivity levels. In sum, the (cid:28)ndings in this paper provide new information on the e(cid:27)ect of trade liberalization on investment, while highlighting the heterogeneous responses of individual plants. 24

References Aghion, P., N. Bloom, R. Blundell, R. Gri(cid:30)th, and P. Howitt (2005). Competition and innovation: Aninverted-urelationship. The Quarterly Journal of Economics 120(2), 701(cid:21)728. Autor, D., D. Dorn, G. H. Hanson, G. Pisano, and P. Shu (2016, December). Foreign competition and domestic innovation: Evidence from u.s. patents. Working Paper 22879, National Bureau of Economic Research. Autor, D. H., D. Dorn, and G. H. Hanson (2013). The china syndrome: Local labor market e(cid:27)ects of import competition in the united states. American Economic Review 103(6), 2121(cid:21)68. Bai, X., K. Krishna, and H. Ma (2015, May). How you export matters: Export mode, learning and productivity in china. Working Paper 21164, National Bureau of Economic Research. Becker, R., W. Gray, and J. Marvakov (2013). Nber-ces manufacturing industry database: Technical notes. NBER Working Paper 5809. Bernard, A. B., J. B. Jensen, S. Redding, and P. Schott (2011). Multi-product (cid:28)rms and trade liberalization. Quarterly Journal of Economics 126(3), 1271(cid:21)1318. Bernard, A.B., J.B.Jensen, andP.K.Schott(2006). Survivalofthebest(cid:28)t: Exposure to low-wage countries and the (uneven) growth of us manufacturing plants. Journal of International Economics. Bloom, N., S. Bond, and J. Van Reenen (2007). Uncertainty and investment dynamics. The Review of Economic Studies 74(2), 391(cid:21)415. Bloom, N., M. Draca, and J. V. Reenen (2016). Trade induced technical change: The impact of chinese imports on innovation, di(cid:27)usion, and productivity. Review of Economic Studies 83, 87(cid:21)117. Brandt, L., J. Van Biesebroeck, L. Wang, and Y. Zhang (2017, September). Wto accession and performance of chinese manufacturing (cid:28)rms. American Economic Review 107(9), 2784(cid:21)2820. Feenstra, R. C., J. Romalis, and P. K. Schott (2002). Us imports, exports, and tari(cid:27) data, 1989-2001. Technical report, National bureau of economic research. Fort, T. C., J. R. Pierce, and P. K. Schott (2017). New perspectives on the decline of us manufacturing employment. 25

Guiso, L. and G. Parigi (1999). Investment and demand uncertainty. The Quarterly Journal of Economics 114(1), 185(cid:21)227. Gutierrez, G. and T. Philippon (2017, July). Declining competition and investment in the u.s. Working Paper 23583, National Bureau of Economic Research. Handley, K. (2014). Exporting under trade policy uncertainty: Theory and evidence. Journal of International Economics 94(1), 50(cid:21)66. Handley, K. and N. Limao (2017, September). Policy uncertainty, trade, and welfare: Theory and evidence for china and the united states. American Economic Review 107(9), 2731(cid:21)83. Hirsch, B. T. and D. A. Macpherson (2013). Union membership and coverage database from the cps, 2011. Khandelwal, A. (2010). The long and short (of) quality ladders. Review of Economic Studies 77, 1450(cid:21)1476. Khandelwal, A. K., P. K. Schott, and S.-J. Wei (2013, October). Trade liberalization and embedded institutional reform: Evidence from chinese exporters. American Economic Review 103(6), 2169(cid:21)95. Kurz, C. J., N. J. Morin, et al. (2016). Annual data on investment and capital stocks. Technical report, Board of Governors of the Federal Reserve System (US). Michaels, D. (2017, March). Foreign robots invade american factory (cid:29)oors. Wall Street Journal. Nunn, N. (2007, May). Relationship-speci(cid:28)city, incomplete contracts, and the pattern of trade. The Quarterly Journal of Economics 122(2), 569(cid:21)600. Pierce, J. R. and P. K. Schott (2012, December). The surprisingly swift decline of u.s. manufacturing employment. Working Paper 18655, National Bureau of Economic Research. Pierce, J. R. and P. K. Schott (2016). The Surprisingly Swift Decline of U.S. Manufacturing Employment. American Economic Review 106(7), 1632(cid:21)1662. Pindyck, R. S. (1993). Investments of uncertain cost. Journal of (cid:28)nancial Economics 34(1), 53(cid:21)76. Rob, R. and N. Vettas (2003). Foreign direct investment and exports with growing demand. The Review of Economic Studies 70(3), 629(cid:21)648. Schott, P. K. (2003, June). One size (cid:28)ts all? heckscher-ohlin specialization in global production. American Economic Review 93(3), 686(cid:21)708. 26

Schott, P. K. (2004, May). Across-product versus within-product specialization in international trade. The Quarterly Journal of Economics 119(2), 647(cid:21)678. Schwartz, E. S. and C. Zozaya-Gorostiza (2003). Investment under uncertainty in information technology: acquisition and development projects. Management Science 49(1), 57(cid:21)70. 27

Online Appendix A Industry Results Through 2011 Theregressionsampleinthemaintextextendsthrough2007inordertoavoidtheGreat Recession. For comparison, Figure 12 reports results for industry-level investment and for the industry-level investment de(cid:29)ators using equation 3 for 1990 to 2011, the limit of the NBER-CES Manufacturing Productivity Database. Figure 12: Implied Industry-Level Impact of PNTR, 1991-2011 28

B Reconciling Industry and Plant Results Table 7: Aggregating Plant Results to Industry Results Table7illustratesthatthedi(cid:27)erencesbetweenindustry-andestablishment-levelresults largely disappear when they are aggregated to the same level and consider the same same dependent variable. Speci(cid:28)cally, the (cid:28)rst column of the table reports results of estimating equation 2 after starting with the establishment-level sample, aggregating to the industry-level, and using the natural log of total investment as the dependent variable. Comparing the coe(cid:30)cient estimate for the DID term in column 1 of Table 7 to the analogous estimate in column 3 of Table 2 indicates that both are negative and statistically signi(cid:28)cant, and of similar magnitudes. The implied e(cid:27)ect in Table 2 is somewhat larger than that in Table 7, as the latter excludes the e(cid:27)ect of plant exit that occurs between 1997 and 2002. Alternatively, when we estimate equation 2 with the same sample as column one of Table 7, but use the dependent variable from the establishment-level regressions(cid:21)investment divided by capital stock(cid:21)we again (cid:28)nd that themagnitudeoftheDIDcoe(cid:30)cientestimateissimilartotheanalogousestablishment- 29

level results from column 2 of Table 4. The coe(cid:30)cient in this industry-level analysis is statistically signi(cid:28)cant, likely because it captures changes in investment associated with net plant exit, which the establishment-level regressions do not. 30