Firm Outsourcing Decisions: Evidence from U.S. Foreign Trade Zones. Deborah L. Swenson. University of California, Davis and NBER.

Transcription

1 Firm Outsourcing Decisions: Evidence from U.S. Foreign Trade Zones Deborah L. Swenson University of California, Davis and NBER Abstract This paper examines the operations of firms located in U.S. foreign trade subzones to study the responsiveness of outsourcing to international cost changes. I find that firms reduce their reliance on foreign inputs when dollar depreciation increases the relative price of imported inputs. The effect is pervasive across industries and is economically significant. In addition, firms that rely more heavily on imported intermediate inputs reduce their overall shipments when dollar depreciation elevates their imported input costs. However, the magnitude of the shipments effect is economically small, suggesting that firms respond to exchange rate movements by adjusting their operations on other dimensions.

2 I. INTRODUCTION During the NAFTA debates U.S. industry argued that it would be placed at a competitive disadvantage if it failed to gain access to Mexican parts that benefitted from tariff reductions. In the absence of such benefits it was argued that U.S. firms might even move their production off shore. At the same time U.S. trade unions expressed great concern that jobs and high levels of economic activity would be lost if NAFTA caused a rise in the outsourcing of intermediate inputs. Although these countervailing concerns have attracted much attention, very little is known about the responsiveness of firm outsourcing to relative price changes or about its secondary effects on firm shipments, since outsourcing decisions are made at the firm level and appropriate data is rarely available to researchers. This paper seeks to address this void by analyzing a data set that provides information on these purchasing and shipment decisions. Recent aggregate evidence indicates that firms are purchasing an increasing portion of their intermediate inputs from their foreign affiliates, or unrelated parties located abroad. 1 It is argued that advances in transportation and communications have facilitated the trend towards the dispersion of production activities. While the aggregate trends are striking, little research has focused on individual firm decisions. One exception is Brainard and Riker's [1997] examination of multinational firms' employment choices as they respond to international wage changes. Brainard and Riker conclude that the degree of firm substitution among a firm's numerous foreign affiliates is much higher than that between parent firms and their foreign affiliates, which is relatively small. This labor market evidence suggests that foreign affiliate labor is not easily substituted for home labor, and that the two may even be complements. 2 In contrast a higher degree of substitutability is implied by work including Feenstra and Hanson [1997 ] and Irwin

3 [1996] for the U.S., or Campa and Goldberg [1997] for the U.S., U.K., Japan, and Canada, which documents that the usage of imported intermediate inputs has increased in almost all cases since the 1970's. 3 This paper studies the issue of input sourcing flexibility and its consequences by examining individual firm decisions. The analysis is based on the activities of a set of firms located in U.S. foreign trade subzones. I focus on this subset of firms, since the U.S. Foreign Trade subzones program provides a unique opportunity to examine firm level decisions regarding inputs and shipments. The reporting requirements of the Foreign Trade Zones program enable us to view aspects of firm sourcing decisions that are usually unobservable. I use two estimation methods to measure the general substitutability of inputs, and outsourcing flexibility. First, I develop a CES model of input demand. I then use non-linear least squares techniques to estimate the underlying demand parameters implied by the structural equations. I then perform a tobit analysis of firm sourcing decisions. The benefit of the second form of analysis is that it allows me to study additional determinants of outsourcing, and it does not require strong assumptions about the functional form of the production function. Both estimation methods indicate that U.S. manufacturing firms changed their sourcing patterns over time as exchange rate changes alter relative production costs. In particular, the data indicate that firms reduce their reliance on U.S. source inputs when dollar appreciation increases their cost relative to imported foreign inputs. While Swenson [1997] showed that automobile industry sourcing responds to changes in international costs, this paper demonstrates that the effect is widespread. Though the economic magnitude of these responses varies substantially across industries, it is significant in many. The fact that firms change input sourcing in response to 2

4 changes in relative production sourcing costs is interesting in itself. Most models of the multinational firm postulate that firms incur fixed operation costs when they extend their activities across borders. 4 While these costs are no doubt real, their magnitude is not large enough to deter frequent and large changes in sourcing over time. Firm input sourcing flexibility has implications for other firm decisions. If a firm can reduce its reliance on those inputs that become more expensive it can moderate cost increases and presumably retain sales it might lose if it were forced to raise prices. 5 If so the cost repercussions of U.S. dollar depreciation are likely to be most detrimental for U.S. producers who rely heavily on imported inputs. To investigate the economic importance of this effect, I examine the sensitivity of firm shipments volumes to exchange rate changes. The evidence indicates that a firm's exchange rate exposure is influenced by its reliance on imported intermediate inputs. 6 Nonetheless, I find that exchange rate exposure transmitted through prior firm sourcing choices is very small when compared with the direct effects of exchange rate movement on firm competitiveness. Here too, I find that the industry-specific responses to shocks transmitted through intermediate inputs vary widely in magnitude across industries. The paper is organized as follows. In the second section I model firm input demands to develop a framework that relates sourcing decisions to relative international cost conditions. Section three provides information on the firms in the sample. Here I use two estimation techniques to evaluate how firms change their sourcing when exchange rate movements alter the relative cost of U.S. inputs. The fourth section measures how prior firm sourcing decisions affect firm shipment volumes. Final issues, and discussion are included in the fifth section. 3

5 II. A MODEL AND ESTIMATION OF FIRM INPUT DEMAND To estimate the responsiveness of relative input demand to changes in relative international production costs I begin with a stylized model of input demands. I consider a representative firm that produces in the U.S. market for sales both in the U.S. and abroad. As the firm produces total output Y, it chooses inputs to minimize its total production costs. The production process is characterized by constant elasticity of substitution between inputs. There are N us varieties of U.S. inputs X us and N f varieties of imported inputs X f. We assume that each input enters the production function symmetrically and that the elasticity of substitution between any two inputs is 1 (1-1/ σ ) (1-1/ σ ) Y us us f f ] 1-1/ σ = [ N represented by σ, where σ> 1. As a result, the production function takes the form: X + N Each firm chooses its input mix of foreign and domestic parts to minimizes its costs in light of input prices. The price of a typical U.S. part is P us, while the price of a representative foreign part is X P f. Cost minimization implies that the demands for each U.S. and foreign input take the following X us = [ P AGG / P us ] σ Y ; X f = [ P AGG / P f ] σ Y, forms respectively: 1 (1-σ ) (1-σ ) P AGG = [ P us +( P f ) ] 1-σ where the composite price of inputs P AGG is represented by: The relative contribution of U.S.-origin content depends on the price and variety of U.S. and foreign-origin inputs. Since a high portion of component purchases are transacted within the firm, 4.

6 whether the parts are produced in the U.S. or abroad, I assume that the parts are priced at marginal cost. I assume further that intermediate inputs are produced by a constant returns to scale production process. As long as we assume that the price of inputs is the same for all parts, the following formula defines µ, the relative contribution of U.S.-origin inputs, or the value of US inputs relative to the firm's total inputs. µ ( P = us us us us us us f f f N X )/( P N X + P N X ) σ -1 = f us us f ) 1/[1+( N / N The equation demonstrates that the relative value of U.S. content is rising in the foreign price of inputs and declining with the U.S. price of inputs. It is important to note that the foreign price expressed in dollars P f is generated by the nominal foreign currency price of foreign parts P * f divided by the exchange rate e, or that P f = P * f /e. Since e represents the number of foreign currency units per U.S. dollar, dollar appreciation, which elevates e, is associated with an increase in the purchase of foreign inputs, and a decline in the purchase of U.S. inputs. )( P / P ] III. ESTIMATION In this section I use two techniques to measure the degree of substitution between foreign and domestic parts used in production. First I first use non-linear least squares techniques to estimate the specification that is implied by the theory of cost minimization under a CES production function. The benefit of this specification is that it facilitates the direct identification of the elasticity of demand for inputs. To verify the robustness of the results I next perform tobit analysis on a more general specification that is augmented to include other industry and firm 5

7 regressors. Data The primary variable of interest is the relative contribution of U.S. input purchases to the firm's overall sourcing plans. While it would be desirable to observe individual firm sourcing decisions for a large set of firms, this information is generally unavailable. I examine a set of firms located in U.S. foreign trade zones, since these firms provide a rare exception to this data limitation. Firms located in U.S. foreign trade zones report the annual value of their domestic and foreign input purchases. These inputs include both raw materials, and manufactured intermediate inputs. Firms in the foreign trade zones program also report their subsequent shipments. 7 For each firm f the relative value of U.S. inputs in year t, is computed as µ ft = (U.S. Source Inputs ft )/(Total Inputs ft,), where Total inputs = (U.S. Source Inputs + Foreign Source Inputs). Since µ ft reports the value of each zone's input purchases, rather than the physical quantities of inputs used, it is important to note that exchange-rate induced changes in input valuation move the observed input sourcing variable in a direction that is contrary to the previous model predictions. 8 In other words, a dollar depreciation causes the reported value of U.S. sourcing to rise, only if firms replace some of their foreign input purchases with U.S. source inputs. Table 1 summarizes the Foreign Trade Zones data. The use of foreign trade zones grew more than 10-fold over the 1984 to 1995 period I analyze. The growth of this program in the 1980's can be attributed to a U.S. Treasury ruling that limited tariffs assessed on foreign trade zone products to the portion of the product originating from abroad. This facilitated the use of foreign trade zones for manufactured products that relied on imported intermediate inputs, since 6

8 the finished products no longer faced tariff payment for the full value of the manufactured item. 9 While most inputs were of U.S. origin, the volume of imported inputs grew from roughly $13 billion in 1984, to $132 billion by The model in section 2 directly links the relative demand for U.S. source inputs to the relative price of inputs, P us/ P f, as shown in equation (4). Since foreign trade zone occupants report the value of their foreign and domestic sourcing, but do not provide any details on the product composition or country of origin of their input purchases, it is not possible to use price indices related to actual products or country of origin. Because price movements differ substantially across U.S. industries, I measure movements in the relative price of U.S. inputs by using U.S. dollar industry real exchange rates. 10 Increases in the real exchange rate variable reflect dollar appreciation, and should be associated with reductions in U.S. sourcing. The summary data indicate that the usage of U.S. source inputs fluctuated from year to year in a manner that is consistent with the exchange rate predictions. The average Foreign Trade Subzone use of U.S. inputs fell to its lowest point in a time when the dollar was especially strong. In contrast, the average use of U.S. source inputs peaked at 85.1 percent in a time that followed the large depreciation of the dollar. Otherwise, the percentage U.S. inputs used changes from year to year with no apparent trends. On the shipments side, it should be noted that almost all Foreign Trade Zone occupants export as well as selling to the U.S. market. Nonetheless, it is apparent that most zones are designed with U.S. customers in mind, as roughly 90 percent of all zone shipments are ultimately shipped to U.S. markets. Here too, there were year to year fluctuations in the relative importance of the U.S. market, though exports never comprised more than twelve percent of total 7

9 subzone output. Non-Linear Least Squares Estimation I estimate equation (4) directly using non-linear least squares. If one assumes that the elasticity is the same for all industries, the estimation yields an elasticity estimate of with a standard error of (0.270). 11 The estimated value of N f /N us is with a standard error of (0.028). However, we can not interpret the estimated value of N f /N us too literally, as it also captures any shifts in the relative demand between domestic and foreign inputs that are not captured elsewhere. For example, suppose that firms differ in their efficiency in using foreign and domestic inputs. Equation 1 could be rewritten with shift terms δ us and δ f multiplying the foreign and domestic inputs respectively to reflect these differential efficiencies. 12 If these shift terms belong in the production function, the functional form of equation (4) remains valid. However, the estimated N f /N us term now is a composite that captures the shift terms as well as the relative diversity of inputs. It may be overly restrictive to assume that the elasticity of demand for inputs is the same for all industries. For this reason, I next estimate the specification implied by equation (4), but allow the estimation to identify unique estimates of σ i and γ i for each industry. Each firm f is assigned to an industry i, based on its primary zone output. The new specification is: µ 1/[1+ γ ( P These results are displayed in the first column of table 2. Of the eleven primary industries in my analysis, the highest implied elasticities of substitution are found in the Pharmaceutical and Chemical, Oil and TV/Computer sectors - industries that are anecdotally known for their / P ft = i us f ( σ -1) ) i ] 8

10 outsourcing of inputs, or which have more homogenous inputs that we would expect to exhibit greater price sensitivity. If there are no differential efficiencies in the use of foreign and domestic inputs, γ i provides industry estimates of N f /N us. However, since the estimate of γ i is a composite measure of input diversity and product demand shifters, it is not a pure measure that we can interpret easily. Tobit Estimation While the first set of results directly measure the implied elasticity of demand, the estimating framework is not easily modified to capture all the firm level and industry level factors that may influence the relative demand for U.S. inputs. I now move to a specification that incorporates firm level and industry factors, as well as the relative cost of inputs. As before, the variable of interest is degree of U.S. sourcing, µ ft, chosen by the firms in the sample. The new specification takes the following form. (5) µ ft = 3 i θ i + ω*r t + λx ft + ε ft The first set of terms θ i are industry indicator variables designed to capture industry-level sourcing differences, the second component measures the sensitivity of US sourcing decisions to changes in relative production costs as measured by the industry real exchange rate R t, and the third set of terms captures changes in firm characteristics X ft over time. The regression error term is represented by ε ft. Since the dependent variable is bounded between zero and one, I perform a tobit analysis. The benefit of the tobit specification is that it allows one to easily add characteristics, such as subzone age or firm fixed effects, which presumably influence the relative demand for U.S. source inputs, and that can not be added naturally to the non-linear least squares specification in (4) without further assumptions about the structure of firm input 9

11 demands. The results of the tobit specification are presented in Table 3, which describes U.S. sourcing responsiveness for the full sample, as well as for different subsets of producers. Row (1) of Table 3 displays the baseline regression of US sourcing on relative costs. The general message is that firms economize on their use of U.S.-source inputs when dollar appreciation causes their relative price to rise. A basic question is whether one should use the current exchange rate, or its lagged value in capturing the effect of relative price changes. If firms require time to adjust to cost changes, then one would use lagged values of the exchange rate. When the first regression is repeated with the lagged value of the exchange rate, the estimated exchange rate coefficient is with a standard error of (0.829). The fit of the regression is slightly worse, as indicated by an increase in the log likelihood to (-474.7). If the current and lagged exchange rates are both included in the same regression, the current exchange rate enters with a value of (.1804), while the coefficient on the lagged exchange rate is (.1446). In these equations, and the later tests, it appears that the current exchange rate has greater explanatory power than do lagged exchange rates. This implies that the sourcing decisions of the firms in my sample respond quickly to exchange rate changes, and for this reason the remaining results presented do not include lagged values of the exchange rate. The age and industry variables included in specification (1) deserve some comment. Age may affect firm sourcing decisions if firms learn more about U.S. sources of supply after they establish their US facilities. In particular, sourcing of U.S. inputs will rise over time as firms shift their purchases towards recently identified U.S. suppliers. In addition, if suppliers 10

12 follow assemblers, we would expect the use of U.S. inputs will increase over time as suppliers set up new facilities that are located near the subzone occupants. 13 These potential changes over time are captured by Age, which is measured as the number of years the subzone has operated. 14 Age 2 is simply the square of the age term, and is entered to capture any non-linearities in the age effect. I include industry dummy variables to capture industry differences in U.S. comparative advantage in supplying inputs. The industry variables can also be justified by a growing literature based on industrial organization theories regarding profit-maximizing multinationals, which imply that the presence of multinational firms may alter the composition and magnitude of trade flows. Multinational firms have to decide whether to export their outputs from a centralized location, or to produce outputs in a number of widely dispersed subsidiaries. The benefits of dispersion include the ability to avoid transport costs, and linkage benefits created when the multinational's local presence stimulates demand in host markets. At the same time, the trend towards dispersed subsidiary production is inhibited by the fixed costs associated with the opening of new plants and facilities. 15 I assume that the production techniques chosen within an industry are similar, which implies that industry comparative advantage will determine a firm's baseline sourcing decisions. If so, I can capture the effects of comparative advantage through a set of industry dummy variables θ i. Each firm f is again assigned to one of eleven industries based on the primary industry activity it conducts in the foreign trade subzone. The possibility that θ i changes over time is addressed later. I take no position on the source of comparative advantage, and whether it arises from inherent cross country differences in factor endowments or technological abilities, or 11

13 whether the comparative advantage arises from economies typified by increasing returns to scale production. 16 I next examine whether U.S. firms respond differently than foreign firms. The Armington Assumption is usually used to justify consumer preferences for home products, but it can be applied to producers as well. 17 For example, it is likely that the typical supplier tailors its production with the needs of home country firms foremost in mind. If so, downstream firms will purchase more inputs from suppliers located in their own country, since these firms have created parts that are most closely customized to their specific needs and preferences. Also, human networks may cause firms to be better informed about the quality of products from their own markets, than about the quality of products from other countries. This would reinforce any firm tendencies to rely more heavily on home produced inputs. In light of these arguments, I apply the general regression specification to foreign and U.S.-owned Foreign Trade Subzones separately, and report the results in rows (2) and (3) of Table I find that the estimated responsiveness is greater for the U.S.-owned than for the foreign-owned facilities. However, the difference in the estimated exchange rate response is not statistically significant. This suggests that while the operation of domestic and foreign firms is somewhat different, it is not overwhelmingly so. For this reason, I treat production by domestic and foreign-owned subsidiaries as comparable for the remaining estimation, and retain the pooled sample. 19 To this point I have assumed a common exchange rate response for all industries. However, it is possible that the relative availability of foreign and domestic inputs differ across industries and that these differences condition industry responses to exchange rate levels. Since 12

14 auto assemblers represent a large portion of the sample, I exclude them from the specification tested in row (4) of Table 3 to see whether the exchange rate effects differ for the non-auto sectors. I find that industries outside the automobile sector respond more vigorously to exchange rate movements. The estimated coefficient of -.49 for the non-auto subsample exceeds the coefficient of -.38 for the full sample by almost 30 percent. While these two estimates are not statistically distinct, they reinforce the conclusion that cross-industry heterogeneity is important, and that it may operate on more than one dimension. I next expand the specification to allow the exchange rate variable to vary uniquely by industry. This involves the inclusion of eleven exchange rate coefficients ω i which correspond to the eleven industries in the sample. The regression now takes the form: (6) µ ft = 3 i θ i + 3 i ω i *R t + λx ft + ε ft Though I am observing data on firms f, located in foreign trade zones, the exchange response is now classified by the industry i, to which the firm belongs. The results which include the industry exchange rate interaction terms are displayed in rows (5), (6) and (7) of Table 3. Row (5) includes no industry dummy variables. Industry dummy variables are added in rows (6) and (7), and row (7) also includes industry trend variables. It should be noted that an F-tests for the joint significance of the industry-exchange rate variables are significant at better than the 1 percent level in all three specifications. The improvement in the log-likelihood of these equations also supports the inclusion of these more detailed variables. Consider specification (6) presented in Table 3. The exchange rate coefficient is negative in every industry but one, indicating that a dollar appreciation causes firm reliance on U.S.-source inputs to decline. For the 10 industries that respond as predicted, I find 13

15 that the exchange rate coefficient ranges from a low of -.07 to a high of -4.2 in the Chemical/Pharmaceutical industry. In principle, we might expect that overall economic changes cause the optimal value of θ i to change over the 12 years of estimation. While it would be possible to estimate unique values of θ it for each industry-year, this would use a large number of degrees of freedom. Instead, I capture changes in comparative advantage through a more compact measure: θ it = θ i0 + ρ i t. Baseline industry differences in comparative advantage are represented by the term θ i0, while changes over the period of estimation are assumed to evolve on an industry basis, and are measured by unique industry-trend terms ρ i. We expect ρ i to be negative if comparative advantage for a range of inputs used in the industry shifts away from the U.S. Two likely scenarios that would cause in this shift would include technology transfer from the U.S. that increased the diversity of available foreign inputs, or U.S. trade liberalization which created the same effect. 20 Industry-year trends are also important if there were secular trends during the period of estimation that changed the relative diversity of foreign and domestic inputs. The new specification that includes industry-specific time trends is presented in column (7) of table 3. While the magnitude of the exchange rate coefficients changes some, this is to be expected, as the industry time trends absorb all industry trend movements, including those attributable to the general decline of the dollar from its peak in Nonetheless, the same qualitative results remain. 21 Another firm characteristic that is likely to influence a firm's sourcing opportunities is whether the firm has facilities in other countries. Presumably, a firm that establishes worldwide operations should be able to change its international sourcing to a greater extent than a pure U.S.- 14

16 based firm. First, since the firm has already sunk fixed costs which accompany the decision to locate some operations abroad, it should be less costly for these firms to increase their foreign sourcing. A second reason relates to the firm's ability to identify relevant new sources of supply. Exposure to information about purchasing opportunities should be especially marked for firms with widely dispersed subsidiaries. 22 To capture these effects, I used corporate directories to identify the firms in my sample which had operations in Europe, South East Asia, Japan, South America or non-u.s. North America, and tested whether the presence of these facilities affected the responsiveness of the firms to changes in relative costs. I do not report these results, since I identified no significant effects. While this channel may be important generally, there is insufficient variation in my sample to identify the effect. 23 In order to check for robustness of the exchange rate results, I ran a final specification which included firm fixed effects γ f. If the exchange rate effect is assumed to be constant across industries, the estimated exchange rate coefficient of (.0452) is not much smaller than the coefficient for the comparable specification presented in row (1) of Table 3. When the firm dummy variables are used along with a full set of industry exchange rate terms, 9 of the 11 estimated industry-exchange rate coefficients are negative. Five of the 9 remain significant at better than the 5% level. Implied Sourcing Responsiveness To quantify the economic implications of the exchange rate coefficients for firm inputsourcing decisions, I performed calculations which compare actual industry sourcing decisions, with the predicted level of sourcing that would have occurred had the U.S. dollar been 10 percent stronger in These calculations are contained in Table 4. 15

17 As a reference point, the first column of Table 4, lists the actual 1995 reliance on U.S. inputs for the 11 industries in the sample. The second column shows the fitted values for input use, which are based on regression (1) in Table 3, which posits a common response to exchange rate changes. However, since the later regression analyses indicate that there was substantial heterogeneity in industry sourcing responses it is best to account for these differences directly. The last two columns of Table 4 report the predicted U.S. input usage for 1995 based on regression specification (6). This specification allows heterogeneous industry response, and is again used to see how U.S. input usage would have changed if the dollar had appreciated to a level 10 percent higher than its actual 1995 value. In this setting, the changes in the two auto sectors are relatively small. With the exception of the pharmaceutical sector, the implied percentage changes are less than 10 percent in the remaining manufacturing sectors. In thinking about the implied effects of these results for the U.S. as a whole, caution is advised. Since firms self select into the foreign trade zones program for its tariff reduction benefits, I expect that the firms in my sample rely somewhat more heavily on imported intermediate inputs than do the universe of U.S. firms more generally. Nonetheless, these results show that the responsiveness of input sourcing decisions is not only statistically significant, but economically meaningful in most industries. One possible explanation for my differential industry exchange rate findings is input specificity. Even though firms located in Foreign Trade Subzone firms do not report their import of foreign inputs at the product level, we know that some industries rely on specifically designed inputs, while other firms can purchase homogenous inputs from any supplier. At the one extreme, we think of many car parts, such as car bodies, or even car engines, as specifically 16

18 designed with the end user in mind. In contrast, though there are differences in grades of unrefined oil, it is likely that the inputs to oil refining or the pharmaceutical/chemical sector are far less differentiated, and not often tailored for particular purchasers. In the absence of other rigidities, such as long-term contracts, sourcing in these industries should respond more vigorously to cost shocks, than sourcing in industries characterized by input specificity. While I measure firm responses to exchange rate induced changes in relative production costs, there is no reason to believe that firms would respond any differently if cost shocks were caused by other factors such as changes in tariff and trade policy. As a result, I expect that firms will change their sourcing patterns as trade liberalization such as NAFTA, reduce the relative cost of imported inputs. More generally, the ongoing decline of trade barriers should influence the mix of foreign and U.S. inputs used in U.S. production. IV. LINKAGES BETWEEN INPUT FLEXIBILITY AND FIRM SHIPMENTS The previous section indicates that firms purchase a higher percentage of their inputs in the U.S. when the relative cost of U.S. inputs declines. However, while firms plan their production optimally in light of exchange rate expectations, they are often surprised by actual exchange rate realizations. In addition, even within industries exchange rate changes are likely to influence firms' competitive positions since prior firm sourcing choices lead to differential firm exchange rate exposure. As a result, I analyze whether firm exchange rate exposure transmitted by intermediate input sourcing helps explain changes in firm shipment levels. Since most of the products in the sample are differentiated manufactured products, the analysis assumes that final goods markets are imperfectly competitive. 17

19 Exchange rate movements influence firm profits, and consequently firm shipments in at least two ways. First, exchange rate changes alter the relative price of foreign and domestic products. This exerts a direct effect on the competitiveness of domestic firm output, and we expect that the adverse competitive consequences of dollar appreciation will reduce domestic firm shipments. Even if there is incomplete pass through of price changes by competing foreign firms, dollar appreciation will generally depress the shipments of domestic firms. Second, exchange rate movements have a unique effect on each firm which arises from the firm's sourcing choices. 24 In this context, we can predict that dollar depreciation, since it raises the price of imported intermediate inputs, will have the most negative consequences for those firms that purchase a greater percentage of their inputs abroad. To capture the effects mentioned, I examine the yearly shipments of individual subzones, Total Shipments, according to the following specification: (7) Total Shipments ft = α + δ*expose ft-1 + β*exchange Rate ft-1 + 3γ i + ε ft In addition to the constant term and industry-specific γ i 's that capture industry differences in firm shipment size, there are two coefficients which are designed to reflect the exchange rate effects described above. The exchange rate measures the foreign currency price of the U.S. dollar, which means that an decline in the variable Exchange Rate, represents a dollar depreciation. Dollar depreciation enhances the relative competitiveness of U.S. products (in both U.S. and foreign markets), and should cause U.S. shipments to rise. The competitiveness effect predicts a negative coefficient β on the pure exchange rate term. 18

20 At the same time, a dollar depreciation harms those U.S. producers who rely heavily on imported intermediate inputs. I capture the effect of exchange rate exposure with the term, Expose = (1-µ ft-1 )*(1/Exchange Rate). As before, µ ft-1 measures the percentage of U.S. origin inputs purchased by the individual zone user. Consequently, (1-µ ft-1 ) measures a particular zone's reliance on foreign inputs, and their relative price is indicated by (1/Exchange Rate). I expect the coefficient δ to be non-positive, since this term reflects the effect of a rise in the cost of intermediate inputs on firm shipments. 25 I experiment with various partial specifications before moving to the full specification which is displayed in Row (5) of Table 5. The coefficient on the interaction term Expose implies that dollar depreciation translates to reduced shipments for firms that rely more heavily on imported intermediate inputs. Highly significant industry dummies indicate that variation in shipment size is influenced by industry characteristics. The negative coefficient on the pure exchange rate term shows that dollar depreciation, because it causes the relative price of U.S. produced goods to fall, elevates the level of shipments from firms located in U.S. foreign trade zones. To see whether the potency of the intermediate inputs channel differs significantly across industries, I change the specification slightly to allow exchange rate sourcing exposure to exert to exert unique effects on the industries in the sample. The modified specification is: (8) Total Shipments ft = α + 3δ i *Expose ft-1 + β*exchange Rate ft-1 + 3γ i + ε ft I now estimate a unique δ i, coefficient for each industry in the sample to capture the effects of exchange rate sourcing exposure for individual industries. The results which include the industry exposure interaction terms are displayed in row (6) 19

21 of Table 5. Ten of the eleven exchange rate exposure terms enter with the predicted negative sign. Though many of the individual terms are not significant at conventional levels, an F-test for the joint significance of the exchange rate terms is significant at better than the 1 percent level. Some of the terms are not surprising. For example, the term for the oil industry, is not statistically different from zero. However, since the demand for oil is not thought to be elastic in the short run, it would be surprising to find a negative coefficient in this industry, as this would imply that adverse production cost shocks exert a strong inhibiting effect on oil shipments. To quantify the economic implications of the shipping results, I can ask how much shipment volumes would have changed if the dollar appreciated 10 percent. I find that a 10 percent dollar appreciation would reduce the representative firm's shipments by 4.1 percent. 26 While the transmission of cost shocks to shipments through the channel of imported intermediate inputs is statistically significant, it is not economically large. Direct demand effects strongly dominate instead. V. CONCLUSION In studying the international composition of input purchases by firms in the U.S. Foreign Trade Subzones program, two findings emerge. First, this paper documents that a firm's relative reliance on U.S. inputs is affected by the relative price of U.S. inputs; when the relative cost of U.S. inputs is elevated by U.S. dollar appreciation, the relative use of U.S. inputs declines. While the strength of this effect differs from industry to industry, the effect appears to be pervasive and it is economically significant. Second, the data in this paper allow one to measure individual firm exposure to exchange rate movements that are transmitted by cost shocks to imported intermediate inputs. The results indicate that those firms that rely more heavily on 20

22 imported intermediate inputs reduce their overall shipments when adverse exchange rate movements elevate their cost of imported inputs. However the magnitude of this effect is quite small, especially when it is compared with the direct effect of exchange rates on the relative competitiveness of U.S. products. In explaining why firm input cost shocks do not have a larger effect on firm shipments, it is important to remember that firms have many other decisions they may alter instead. Campa and Goldberg [1995] show that U.S. firms reduce their investment when adverse exchange rate shocks hit U.S. industries. Alternatively, firms may instead adjust employment, as is considered by Burgess and Knetter [1996]. Finally, firms may find that sourcing and pricing to market are substitutable channels for adjusting to exchange rate changes as is suggested by Rangan and Lawrence [1993] and Gron and Swenson [1996; 1998]. In any case, it is clear that firms in some industries change their sourcing of inputs significantly when relative prices change. While I present evidence that describes the effects of exchange rate induced cost shocks, there is no reason to believe that other changes that affect the relative price of imported inputs, including transportation cost reductions, or the effects of trade reforms on tariff costs, would not have a comparable effect on the relative demand for imported intermediate inputs. The fact that exposure transmitted through firm differences in sourcing does not translate into large shipment effects may simply mean that firms maintain shipment levels in the short run, while adjusting their long run operations along other dimensions including the sourcing of their input purchases. 21

23 References Bergsten, C. Fred and Marcus Noland. (1993), "Reconcilable Differences? United States-Japan Economic Conflict." Washington, D.C.: Institute for International Economics. Blonigen, Bruce A. and Wesley W. Wilson. (1996) "Explaining Armington: What Determines Substitutability Between Home and Foreign Goods?" University of Oregon Working Paper. Brainard, Lael. (1997) "An Empirical Assessment of the Proximity-Concentration Tradeoff between Trade and Multinational Sales." American Economic Review, 87(4): Brainard, S. Lael and David A. Riker, (1997), "U.S. Multinationals and Competition from Low Wage Countries," NBER Working Paper 5959, March. Burgess, Simon and Michael M. Knetter (1996), "An International Comparison of Employment Adjustment to Exchange Rate Fluctuations." NBER Working Paper 5861, December. Campa, Jose and Linda S. Goldberg. (1995) "Investment in Manufacturing, Exchange Rates and External Exposure," Journal of International Economics, 38: Campa, Jose and Linda S. Goldberg, (1997) "The Evolving External Orientation of Manufacturing Industries: Evidence from Four Countries," NBER Working Paper 5919, February. Casella, Alessandra and James E. Rauch, (1997), "Anonymous Market and Coethnic Ties in International Trade," March Working Paper. Caves, Richard E. (1996), Multinational Enterprise and Economic Analysis, Second Edition, Cambridge: Cambridge University Press. Dornbusch, Rudiger, (1987), "Exchange Rates and Prices," American Economic Review, 77: Feenstra, Robert C, (1989). "Symmetric Pass-Through of Tariffs and Exchange Rates Under Imperfect Competition: An Empirical Test", Journal of International Economics, 27: Feenstra, Robert C. (1998), "Integration of Trade and Disintegration of Production in the Global Economy," Journal of Economic Perspectives, 12(4): Feenstra, Robert C., Joseph E. Gagnon, and Michael M. Knetter, (1996), "Market Share and

24 Exchange Rate Pass-Through in World Automobile Trade," Journal of International Economics, 40: Feenstra, Robert C. and Gordon H. Hanson (1996) "Foreign Investment, Outsourcing and Relative Wages," in R.C. Feenstra, S.M. Grossman and D.A. Irwin, eds., The Political Economy of Trade Policy: Papers in Honor of Jagdish Bhagawati, MIT Press, Goldberg, Linda S. (1993)"Exchange Rates and Investment in United States Industry." Review of Economics and Statistics, 75(4): Gron, Anne and Deborah L. Swenson (1996), "Incomplete Exchange-Rate Pass-Through and Imperfect Competition: The Effect of Local Production," American Economic Review, 86(2): Gron, Anne and Deborah L. Swenson (1998), "Cost Pass-Through in the U.S. Automobile Market," manuscript. Irwin, Douglas A. (1996), "The United States in a New Global Economy? A Century's Perspective," American Economic Review, 86(2): Markusen, James R. (1995) "The Boundaries of the Multinational Enterprises and the Theory of International Trade." Journal of Economic Perspectives, 9(2): Markusen, James R. and Anthony J. Venables, (1998), "Multinational Firms and the New Trade Theory," December, 46(2): Rangan, Subramanian and Robert Z. Lawrence. (1993). "The Responses of U.S. Firms to exchange Rate Fluctuations: Piercing the Corporate Veil," Brookings Papers on Economic Activity, (2): Roberts, Mark J., Theresa A. Sullivan and James R. Tybout, (1995) "Micro Foundations of Export Booms: Evidence from Colombia, Mexico, and Morocco." Pennsylvania State University Working Paper, December. Slaughter, Matthew J., (1995), "Multinational Corporations, Outsourcing, and American Wage Divergence," National Bureau of Economic Research Working Paper # Swenson, Deborah L. (1997) "Explaining Domestic Content: Evidence from Japanese and U.S. Auto Production in the U.S.," in Robert C. Feenstra (editor), The Effects of U.S. Trade Protection and Promotion Policies. Chicago: University of Chicago Press. Zeile, William J. (1997), "U.S. Intrafirm Trade in Goods," Survey of Current Business, February,

25 Table 1: Trends in Foreign Trade Subzone Usage, Year No. of Value of Inputs %U.S. Inputs %U.S. Subzones $Million Shipments , , , , , , , , , , , , Notes: Data based on Foreign Trade Zones Board Annual Reports and Author's Calculations. Data include Foreign Trade Subzone data, but exclude General Zone users. %US Inputs = [US Inputs]/[Total Inputs]. %US Shipments = [US Shipments]/[Total Shipments]. Total Shipments = US Shipments + Exports. The percentage calculations are weighted by zone input size.

26 Table 2: NLS - Estimated Elasticity of Demand for Inputs σ i γ i Auto Assemblers (.701) (.028) Auto Parts (1.296) (.211) Oil Refineries (2.460) (.684) Food (2.635) (.260) Shipbuilding (1.613) (.134) Pharmaceutical/Chemicals (4.921) (7.049) Appliances/Electronics (3.588) (.555) Apparel (1.863) (.509) TV's/Computers/Microwaves (1.285) (.179) Heavy Machinery, Tractors, etc (1.461) (.431) Other Manufacturing (1.594) (.239) R Notes: 974 observations. Estimation is based on equation (4'). Standard errors in ( ). 25

27 Table 3: The Effects of the Exchange Rate on Percent U.S. Input Usage Tobit Estimates: Dependent Variable, Percent U.S. Input Usage Reale Age Age-Sq Industry Constant Obs Log-L (1) a a Yes.7234 a (.0913) (.0121) (.0010) *** (.0991) (2) US-Owned a a Yes.6273 a (.0997) (.0126) (.0011) *** (.1044) (3) Foreign-Owned b.0487 c c Yes.9987 a (.1696) (.0247) (.0020) *** (.2172) (4) No Auto a a Yes.8111 a Assemblers (.1382) (.0176) (.0014) *** (.1411) (5) EI.0221 c a No.8646 a *** (.0121) (.0010) (.0819) (6) EI a Yes *** (.0122) (.0010) *** (.2886) (7) Including Industry- EI b Yes Year Terms *** (.0121) (.0010) *** (2.4361) Notes: Standard Errors in Parenthesis. EI Indicates that the specification has interaction terms for the industry dummies with the exchange rate variable. EI results are contained in the following table which provides the remaining Table 3 results. *** Indicates that the set of variables is jointly significant at the 1% level. The letters a, b, and c denote statistical significance at the 1%, 5% and 10% levels, respectively.

28 Table 3 (Continued): The Effects of the Exchange Rate on Percent U.S. Input Usage Tobit Estimates: Dependent Variable, Percent U.S. Input Usage Exchange Rate Coefficients Sector Specification (5) Specification (6) Specification (7) Auto Assemblers a (.0782) (.1279) (.1940) Auto Parts a c (.1036) (.2604) (.5335) Oil Refineries a a.1834 (.1861) (.2678) (.3839) Food b a (.1336) (.5679) (1.099) Shipbuilding b (.1022) (.3483) (.6319) Pharmaceutical/Chemicals a a.2142 (.1242) (1.6243) (3.1544) Appliances/Electronics a (.1162) (.6302) (1.4938) Apparel a a (.1009) (.3118) (.5703) TV's/Computers/Microwaves a b a (.0925) (.2247) (.4624) Heavy Machinery, Tractors, etc a b (.0908) (.2739) (.4195) Other Manufacturing a (.0973) (.3472) (.5501) Notes: Standard Errors in ( ). The letters a, b, and c denote statistical significance at the 1%, 5% and 10% levels, respectively. 27

29 Table 4 Actual and Predicted Values of US Input Usage, 1995 Industry % US Inputs 1995 µ-prd(1) µ-prd(1) 10% $ apprec µ-prd(6) µ-prd(6) 10% $ apprec Auto Assemblers Auto Parts Food Shipbuilding Pharmaceutical/Chemical Appliances/Electronics Apparel TV's/Computers/Microwaves Heavy Machinery Oil Refineries Other Manufacturing %-US Inputs 1995 is the actual percentage US inputs by industry for µ-prd(1) is the percentage US inputs predicted for 1995, using the row(1) specification from Table 3. µ-prd(1) 10 % $ apprec predicts 1995 usage of US inputs if the dollar had been 10 % stronger. µ-prd(6) is the percentage US inputs predicted for 1995, using the row (6) specification of Table 3. µ-prd(6) 10% $ apprec predicts 1995 usages of US inputs if the dollar had been 10% stronger.

30 Table 5: The Effects of Exchange Rate Exposure on Shipment Volume Dependent Variable: Total Shipments Expose t-1 Real Exch t-1 Industry Industry* Constant Obs Adj-R 2 Dummies Expose t-1 (1).3774 c a (.1942) (.1577) (2) a (.0760) (.0603) (3) a b a (.0862) (.2763) (.2421) (4) a Yes a (.0816) *** (.1482) (5) a c Yes a (.0816) (.2970) *** (.2795) (6) c Yes Yes (.2971) *** *** Exchange Rate Exposure Interaction Terms 3 Ind i *Expose t-1 Ind1 Ind2 Ind3 Ind4 Ind5 Ind6 Ind7 Ind8 Ind9 Ind10 Ind a b a c (.1879) (.2471) (.1717) (.4777) (.4893) (.3454) (.3087) (.3778) (.2761) (.3188) (.2524) Note: Standard errors in ( ). Ind1 = Auto Assembly, Ind2 = Auto Parts, Ind3 = Oil Refining, Ind4 = Food, Ind5 = Shipbuilding, Ind6 = Pharmaceutical/ Chemical, Ind7 = Appliances/Electronics, Ind8 = Apparel, Ind9 = TV's/Computers/Microwaves, Ind10 = Heavy Machinery/Tractors, Ind 11 = Unclassified Mfg. *** Indicates that the set of variables is jointly significant at the 1% level. The letters a, b, and c denote statistical significance at the 1%, 5% and 10% levels, respectively.