Outsourcing and Technological Change

technological change and outsourcing. We present a dynamic model that analyzes how firms’ expectations with regards to technological change influence ...

0 downloads 34 Views 307KB Size
Outsourcing and Technological Change

Ann Bartel, Columbia University and NBER Saul Lach, The Hebrew University and CEPR Nachum Sicherman, Columbia University and IZA

February 2009

The authors gratefully acknowledge the generous support of grants from the Columbia University Institute for Social and Economic Research and Policy and the Columbia Business School’s Center for International Business Education and Research and outstanding research assistance from Ricardo Correa and Cecilia Machado. Saul Lach acknowledges the support of the European Commission grant CIT5-CT-2006-028942.

Abstract We present a model that shows that firms that expect and experience higher rates of technological change are more likely to outsource part of their production. The intuition behind the model is that as the pace of innovations in production technology increases, the less time the firm has to amortize the sunk costs associated with purchasing the new technologies. We test the predictions of the model using a panel dataset on Spanish firms for the time period 1990 through 2002 and find that a variety of measures that are likely to reflect higher rates of technological change are strongly correlated with outsourcing. In order to address endogeneity problems we allow for unobserved firmspecific effects and use an exogenous proxy for technological change, namely the number of patents granted by the U.S. patent office, matched to the Spanish firms’ industrial sector. The number of U.S. granted patents in the firm’s industry has a positive and significant effect on the likelihood that the Spanish firm outsources production.

1

I. Introduction Outsourcing, or the contracting out of activities to subcontractors outside the firm, has become more widespread. For example, Magnani (2006) reports that the cost share of purchased services in U.S. manufacturing industries grew from 4.4% in 1949 to 12% in 1998. Similar increases in outsourcing have been observed in Europe as well.1 A number of explanations for the increase in outsourcing have been proposed and tested in the literature. Among them are that outsourcing is a response to unpredictable variations in demand (Abraham and Taylor, 1996), an opportunity to take advantage of the specialized knowledge of suppliers (Abraham and Taylor, 1996), and a method to save on labor costs (Abraham and Taylor, 1996; Autor, 2003; Diaz-Mora, 2005; and Girma and Gorg, 2004). 2 The role of technology as a factor in firms’ outsourcing decisions has received increasing attention. According to the transactions costs theory (Williamson 1975, 1985), if investments in new technologies result in greater asset specificity, firms fearing expropriation of investments will reduce outsourcing. Lileeva and Van Biesebroeck (2008) found evidence of a negative relationship between specificity and the outsourcing of inputs in the Canadian manufacturing sector.3 Mol (2005), however, shows that in the Dutch manufacturing sector, R&D-intensive industries are relying more on outsourcing as they use partnership relations with outside suppliers to reduce the likelihood of

1

See Arndt and Kierzkowski (2001), Mol (2005), and Abramovsky and Griffith (2006).

2

Other researchers (Ono (2007) and Holl (2007)) have studied the effect of agglomeration economies on outsourcing. For empirical studies of the impacts of outsourcing on wages and productivity, see Feenstra and Hanson (1999), Amiti and Wei (2006), Gorg, Hanley and Strobl (2007), and Gorg and Hanley (2007).

3

Lileva and Van Biesebroeck (2008) also found that complementarities between the investments of the buyer and the seller are associated with less outsourcing.

2

specific investments being appropriated. 4 Magnani (2006) finds evidence that technological diffusion driven by R&D spillovers is in part responsible for the growth of outsourced services in the U.S. Abramovsky and Griffith (2006) focus on the role played by information and communications technology in facilitating the fragmentation of production and find that UK firms that were intensive in these technologies were more likely to purchase services in the market. In this paper we propose an alternative perspective on the relationship between technological change and outsourcing. We present a dynamic model that analyzes how firms’ expectations with regards to technological change influence the demand for outsourcing – an issue ignored in the literature – while abstracting from other considerations (e.g., transaction costs, specificity, etc.). The model shows that outsourcing becomes more beneficial to the firm when technology is changing rapidly. A firm can buy the latest technology and produce intermediate inputs in-house. Firms incur a sunk cost when adopting new technologies. Outsourcing, on the other hand, enables the firm to purchase inputs from supplying firms using the latest production technology while avoiding the new technology sunk costs. As the pace of innovations in production technology increases, the less time the firm has to amortize the sunk costs associated with purchasing the new technologies. This makes producing in-house with the latest technologies relatively more expensive than outsourcing. The model therefore provides an explanation for the recent increases in outsourcing that have taken place in an

4

Baker and Hubbard (2003) consider how information technology in the trucking industry impacts contracting possibilities and vertical integration. Baccara (2007) uses a general equilibrium model to study how information leakages could affect a firm’s outsourcing decision as well as its investments in R&D. For more on the hold-up problem see Grossman and Hart (1986), Grossman and Helpman (2002) and Antras and Helpman (2004).

3

environment of increased expectations for technological change. We test the predictions of the model using a panel dataset of Spanish firms during the period 1990 through 2002. This dataset is superior to those used in previous studies of the determinants of outsourcing in several dimensions. First, unlike other studies that used industry level data, we use a large sample of firms. Second, we have panel data that allow us to observe changes within firms over a long period of time. Third, in addition to detailed information on outsourcing, the dataset provides rich information related to technological activities, such as use of computers, investment in R&D, registration of patents, and product innovation. Our empirical analysis requires an exogenous measure of expected technological change by the firm. For this purpose, we use the number of patents granted by the U.S. patent office classified by technological class and map the technological classes to the industrial classification used in the Spanish data set. The empirical results support the main prediction of the theoretical model, namely, that all other things equal, the demand for outsourcing increases with the probability of technological change. This finding is robust to the inclusion of firm-level fixed effects, unlike the findings for many of the non-technology variables in the empirical model. Part II describes a simple dynamic model of the relationship between outsourcing and expected technological change. The complete model is given in the Appendix. Part III discusses the data and empirical specifications used to test the predictions of the model. Results are presented in Part IV. Part V concludes.

4

II.

The Outsourcing Decision In this section we sketch a simple model that shows how the outsourcing decision

is related to the probability of technological change. The complete model is presented in the Appendix. Before doing this we present a simple example to fix ideas. Suppose that a final good is produced by a machine that requires one computerskilled operator. The final good firm needs to hire one instructor to train this worker. The training itself, as well as the hiring and firing processes, involve a sunk cost. The same instructor, however, could possibly train more than one person simultaneously without incurring additional costs, but only one person is needed for production. The combination of a sunk cost and indivisibility (of the instructor) is precisely the feature being exploited by temporary employment agencies (Autor, 2003): they use the same instructor to train several workers in basic computer skills and offer them to firms at an attractive price because they can spread the sunk cost over a larger output (computer-skilled workers). In a temporal setting, some tasks are performed infrequently by firms (training, repairs, maintenance, bookkeeping, etc.). If such tasks are performed by dedicated employees of the firm, these workers will be idle substantial amounts of time. If such tasks are performed by different firms at different times, an entrepreneur can offer to perform all these tasks at an attractive price relative to the in-house alternative because it uses the same workers continuously, thereby lowering the average fixed cost of production. The model we sketch here attempts to capture the essence of these examples. We start by assuming that a firm produces a profit-maximizing amount y of a final product using an amount x of a single input. The input x can be procured in two different ways.

5

One way is to produce x in-house, while the other way is to purchase x in the market, which we refer to as outsourcing production. In order to produce x in-house, the firm uses another input which we call machines, each of which produce θ units of x. We will later allow the productivity parameter θ to change over time. The cost of each machine is pk and there is a sunk cost

s associated with installing and using the machines of a given productivity (or vintage) for the first time. s is incurred only once and includes training costs. The total cost of producing an amount x in-house is therefore CH =

( ) x + s. pk

θ

The firm can also procure x from the market by paying a price p per unit of x . The total cost of buying x in the market is CO = px. We implicitly assume that there are firms (the suppliers) willing to produce and supply x , i.e., there is an active market for x, but we do not model the supply side here. In terms of our example, x represents the number of computer-skilled workers required while the machines are the computer instructors. The firm needs to use

x

θ

instructors since each one can train θ workers. Suppose θ > x. Because of frictions in the labor market the firm needs to hire a full-time computer instructor, which is more than what is strictly required to train x workers, and incur s dollars in hiring (and firing) costs. Alternatively, the firm can contact a firm specializing in computer training and they will send an instructor who will teach the exact number of hours required. In this scenario, the firm chooses the procuring alternative that minimizes costs. For simplicity we assume constant returns to scale in production. This allows us, after a normalization, to set x = y, which is desirable for the empirical work since y is

6

observed. This assumption also ensures that the firm does not split x between in-house and outsourcing. The firm will either produce all of x in-house or will outsource all of it. 5 As a result the firm will

(

outsource if CO ≤ CH ⇔ s > p − or

pk

θ

)y (

produce in-house if CO > CH ⇔ s < p −

pk

θ

)y

Suppose that p is greater than the marginal cost of in-house production,

pk

θ

, and

that there is heterogeneity across firms in the sunk cost s. Firms would then outsource only because their sunk cost of in-house production is relatively large; firms with low sunk costs will not outsource production. Indeed, only a fraction G

(( p − ) y ) of the pk

θ

firms will produce in-house, where G( ⋅ ) is the distribution of s. A larger production size decreases the likelihood of outsourcing because the sunk cost per unit of production decreases. Two remarks are in order. If p is below the marginal cost of in-house production, all firms will outsource x . This may characterize cases where the production of x is subject to increasing returns or learning effects which confer a cost advantage to suppliers producing (relatively) large amounts of x to serve a large market. 6 The interesting case is when p >

pk

θ

and we therefore maintain this assumption. The second

5

In reality, however, firms usually outsource part of their production, so that we should interpret x as one of the multiple input components of the final output. 6

For example, it is very rare to see a restaurant that grows the vegetables it uses for cooking. It is so rare that we do not call it outsourcing.

7

point is that if we do not allow for heterogeneity in s then all firms of a given size would be either outsourcing or producing in-house which is in general contrary to the facts. 7 We extend this simple model of outsourcing to allow for technological change in the production of x . There are now two periods and we let the productivity of machines change in period 2. Specifically, in the first period θ is given by θ1 , while in period 2 θ can increase to θ 2 > θ1 with probability λ or remain at θ1 with the complementary probability ( 1− λ ). Sunk costs are drawn at the beginning of period 1 and known to the firm. We view the probability of technological change as exogenous to the firm producing the final good y. One interpretation of this assumption is that technological change in embodied in the machines used in producing x . We solve the extended model from the second period backwards (details are in the Appendix). Suppose there is technological change in the second period. The firm faces three alternatives: it can outsource, it can produce in-house with the old technology or it can pay the sunk cost and upgrade to the new vintage of machines. To simplify the analysis we make the assumption that upgrading always dominates keeping the old technology. This requires a not too large sunk cost and we bound the possible values that

s can take in order to make upgrading optimal. A-fortiori, supplier firms should always be producing with the latest technology, and we assume they do. 8

7

Equivalently, we could also have added an additional term to C0 to reflect heterogeneity in the cost of

outsourcing among firms. What matters for the decision to outsource is the difference between C0 and

CH . In this case s represents the additional cost of using a standardized input. This additional cost is absent when x is produced in-house because the firm can perfectly tailor the input to its specific needs. 8

Supplier firms can spread the sunk cost of upgrading over a larger quantity produced.

8

Under this assumption, the firm's decision in the second period when technological change occurs is essentially the same as the decision in the static model: either to outsource or to produce in-house with the new technology. 9 The price of x in the market may change as suppliers are producing with lower marginal costs and therefore can charge a lower price for x and still make profits. Whether they will do this or not depends on the competitive environment. If there is no technological change, however, the firm's outsourcing decision in period 2 depends on its outsourcing decision in period 1. In particular, if the firm produced in-house in period 1 then it will continue producing in-house in period 2 since the sunk costs are already paid and the marginal cost of producing in-house is smaller than the price of x. If the firm outsourced in period 1 the firm will continue outsourcing

(

if its sunk costs are high enough. Indeed, it will outsource as long as s ≥ p −

pk

θ

) y where

p is now the price of x in period 2 in the absence of technological change. Thus, the

decision to outsource in period 2 when no technology change occurs depends on the outsourcing decision in period 1. The decision to outsource in period 1 is forward looking and takes into account its effects on period 2 decision. In this simple model when λ increases and a new technology is more likely to appear in the future the firm will be more reluctant to buy the current technology today and produce in-house because it will soon be obsolete. Upgrading the technology -- which is the optimal thing to do -- involves incurring a sunk

9

Because producing in-house with the new technology in period 2 requires incurring the sunk cost s , the firm's decision in the first period is irrelevant to its current decision.

9

cost di novo. The higher is λ , the more frequently the new machines arrive and the less time the firm has to amortize the sunk costs. Instead, the firm can use outsourcing to obtain x from supplying firms using the latest technology and avoid the sunk costs. In terms of our (first?) example, when a new and better software product appears in the market, the firm needs to hire a new full-time computer instructor who specializes in the new software. When the probability of software upgrades increases, the firm will anticipate that it will need to incur the sunk costs of finding and hiring a new instructor more frequently and will therefore decide to outsource its training activities to a specialized firm. All other things equal, an increase in λ decreases the cost of outsourcing relative to that of in-house production making the firm more likely to decide to outsource in the first period. This is the main prediction of the model that we take to the data. 10 III. Data and Empirical Specification We use the Encuesta sobre Estrategias Empresariales (ESEE, or Survey on Business Strategies), a panel of approximately 1800 Spanish manufacturing companies, surveyed annually since 1990. Data are currently available through 2002. The survey is conducted by the Fundacion SEPI with the support of the Ministry of Industry, Tourism and Trade. We use data from the 1990, 1994, 1998 and 2002 surveys. 11 In each year, the firms were asked if they contracted with third parties for the manufacture of custom-made finished products or parts, and if so, the value of the 10

In the Appendix we show that the decision to outsource in the first period depends also on current and future prices of x as well as on the sunk costs.

10

outsourced products or parts. We use this information to create two indicators of outsourcing, one a dummy for whether or not the firm did engage in outsourcing, and two, the value of outsourcing divided by total costs. 12 Table 1 shows the percentage of firms that reported outsourcing at least some part of production during the 1990 – 2002 period and the mean value of the outsourced production as a percentage of total cost. On average, 41% of firms reported that they outsourced production during this time period. The outsourcing percentage rose from 35% in 1990 to 43% in 2002. There is significant variation in the likelihood of outsourcing across industries ranging from a low of 16% for “drinks” to a high of 61% for “machinery and mechanical goods”. The value of the outsourced production as a percentage of total costs is approximately 5 percent during this time period; for firms that did outsource production, the mean value of outsourced production as a percentage of total costs is 11 percent. In the previous section, we derived the demand for outsourcing in period t as a function of expected technological change. In the Appendix we also show that the decision to outsource depends on the level of output and on the various input prices. Under the assumption that input prices and technology parameters are common to all firms in an industry, these can be captured by industry dummies (ID). Thus, defining χ t as the outsourcing decision in period t , we get: E ( χ t ) = h(λ , yt , IDt )

(1)

11

We did not acquire data for all of the years between 1990 and 2002 because a number of key variables were not available in the intervening years. 12 Lopez (2002) describes the evolution of outsourcing of services and of production in Spanish manufacturing firms using a sub-sample of the ESSE data for the period 1990-1999. He finds significant differences in outsourcing between small and large firms and a positive effect of outsourcing on productivity.

11

Equation (1) is a reduced-form expression because all the arguments in the h function are exogenous in the model. The main implication of the model presented in Part II is that, all other things equal, the demand for outsourcing χ t increases with the exogenous probability of technological change, λ. We make separability and linearity assumptions on the function h,

χ it = β λit + α y it + ∑ δ j IDij + uit

(2)

j

In order to test the prediction of the model, we need a proxy for λ, the firm’s expectations with regard to exogenous technological change. In spite of the richness of our data, this is a difficult task. We use two approaches. First, we rely on several variables from the dataset that are likely to be correlated with the technological change expected by a firm. Second, we use a variable outside of our dataset that is clearly exogenous. With respect to the first approach, we proxy λ by the following variables: binary variables for each year indicating whether the firm engaged in R&D, registered a patent, introduced product innovations, or frequently changes the products it offers. The rationale for these proxies is, for example, that a firm that engages in R&D is more likely to experience technological change (new processes and/or new products) as compared to a firm that does not engage in R&D. There are a number of problems with using these variables as proxies for expected technological change. First, although the model assumes λ to be exogenous, a serious concern is that the probability that the firm will experience technological change is determined by unobserved factors that also affect the decision to outsource. More

12

innovative firms may be doing more R&D, registering patents, engaging in product innovations and may also be adopting new production methods that require more outsourcing. The variables we use to proxy for λ, therefore, may be endogenous in the outsourcing equation. The estimated coefficients on the proxies will not capture the causal effect of technological change on outsourcing. We start to address this concern by exploiting the panel nature of our data which allows us to specify a time-invariant component in uit that may affect both the decision, for example, to do R&D (and therefore λ) and outsourcing. Consistent estimates under this assumption can be obtained from Logit estimation of equation (2) after timedemeaning (including fixed effects) or OLS estimation using first-differencing. Doing this, however, addresses only some of the possible causes for the endogeneity of λ (or its proxies). Causality could also possibly go in the other direction. For example, the decision to outsource could induce the firm to shift from secrecy to patenting in order to protect an innovation. 13 We address this concern by using a proxy for λ which, by construction, is exogenous to the firm. For that purpose we use the number of patents granted by the .U.S. Patent Office classified by year and industry of origin. In order to assign a yearly number of patents to each firm that will reflect changes in the technological frontier we take the U.S. patents counts by year and technological class (as codified by the International Patent Classification) and match these counts by IPC to the 44 technological fields provided by Schmoch et. al. (2003). Schmoch et. al. (2003) also provide a concordance between these technological fields and

13

For this reason, technological change might deter firms from outsourcing the production of a product or component for which competitors could more easily copy or steal an innovation (Williamson, 1985).

13

the Spanish industry classification. 14 In this manner, we assign to each Spanish firm the average number of patents issued in the US over the previous five years and use this variable as a proxy for λ. These data are shown in Appendix Table A-2. 15 In addition to expected technological change, there may be other time-varying factors that are likely to affect the decision to outsource. Our estimating equation includes measures of the firm’s current technological intensity: whether it uses computerized digital machine tools or uses robotics, whether it has registered any patents during the past year, and whether it introduced any product innovations during the past year. We also include a set of variables that were the focus of previous research on the determinants of outsourcing. Since firms may use outsourcing as a way of economizing on labor costs (see Abraham and Taylor, 1996), we include the firm’s average labor cost defined as total annual spending (wages and benefits) on “staff” divided by total employment. Outsourcing may also be used to smooth the workload of the core workforce during peaks of demand (Abraham and Taylor, 1996; Holl, 2007). Hence, we add a measure of capacity utilization defined as the average percentage of the standard production capacity used during the year. Small firms would be expected to be more likely to outsource because it may not be optimal for them to carry out all steps in the production process because of the costs of maintaining specialized equipment or skills inhouse (Abraham and Taylor, 1996). Hence we control for the size of the firm using four categories for number of employees. Another factor that can increase the propensity to

14

Our source of patent data is the “NBER Patent Citations Data File". See the Data Appendix for more details. For comprehensive information see Hall, Jaffe and Trajtenberg (2001). 15

Ideally we would prefer to count patents by the industry of use rather than by industry of origin. Such data is not readily available and we are currently in the process of obtaining such data.

14

outsource is the volatility in demand for the product (Abraham and Taylor, 1996; Holl, 2007). We proxy volatility using two dummy variables that indicate whether the main market the company serves expanded or declined during the year. It has also been argued that very young firms may be less likely to outsource because they have not had sufficient time to learn about the quality and reliability of potential subcontractors (Holl, 2007). We therefore include the age of the firm and its quadratic. Whether the firm primarily produces standardized products or custom made products could also affect the propensity to outsource if custom made production involves more frequent changes in production. We therefore add a dummy variable that equals one if the firm produces standardized products that are, in most cases, the same for all buyers. Finally, we control for the firm’s export propensity, the value of exports divided by sales and whether the firm has any foreign ownership. Both of these factors have been included in prior studies of outsourcing (Girma and Gorg, 2004; Diaz-Mora, 2005; and Holl, 2007). 16 IV. Results We estimate Equation (2) using two dependent variables: an indicator of whether or not the firm engaged in outsourcing and the value of the firm’s outsourced production as a percentage of total costs. Results using the first dependent variable are shown in Table 2 and for the second in Table 3. The equations are estimated for all years using clustered standard errors at the firm level to account for heteroskedasticity and arbitrary serial correlation.

16

The firm’s location could also serve as a proxy for the ease with which outsourcing can be done (see Ono, 2007). In our first differences model, this variable disappears.

15

A. Expected Technological Change In Table 2 we report the estimation results using logit regressions (columns 1, 2, and 4) and OLS first difference regressions (columns 3 and 5). For the logit regressions, the table reports marginal effects evaluated at the mean. Column (1) is a basic regression without fixed effects that uses various indicators from the survey to proxy expected technological change: engaging in R&D, frequently changing product offerings, introducing product innovation, and registering patents. All of these indicators are positively and significantly correlated with the probability of outsourcing. Column (2) adds fixed effects and column (3) estimates the regression using first differences. The R&D indicator and the product innovation indicator remain significant in both columns (2) and (3). According to the estimates in column (3), firms that engage in R&D are 5 percent more likely to outsource and firms that engage in product innovation are 4 percent more likely to outsource. As explained in Part III, however, we cannot infer causation from these results because the firm’s decision to engage in R&D and/or product innovation is endogenous. Columns (4) and (5) add the U.S. patents variable, an arguably exogenous measure of expected technological change, which is positive and significant in both regressions. According to column (5), a 1 percent increase in U.S. patents leads to a 5 percentage point increase in the probability of outsourcing. 17 In Table 3, the dependent variable is the value of the outsourced production

17

In general one would expect the estimated coefficient to decrease when estimating a linear regression using first differences, as can be seen in all the other variables listed. We do not have a clear explanation why it is the opposite with the patent data. However, a few points should be noted. First, in column (4) we report the results of a logit regression and not OLS, so the comparison is not that straightforward. Second, unlike the other variables, the patent data is exogenous, thus it is not clear that one should expect the estimated coefficient to be lower in first differences. And last, although the coefficient gets larger, one cannot reject the hypothesis that it is not different from the one estimated in column (4).

16

divided by total costs. Since 60 percent of the observations are zeroes, we use tobit regressions. The results in the first column indicate that, a one percent increase in U.S. patents is associated with a 1.3 percent increase in the probability of outsourcing; this effect is very close to the results observed in column (4) in Table 2. Column (2) in Table 3 also shows that, for firms that do outsource, the ratio of outsourcing to total costs increases by .002 when U.S. patents rise by 1 percent. Using the mean value of the dependent variable for firms that do outsource (see Table 1), the coefficient translates to a 1.7 percent increase in the share of outsourcing in total costs. B. Other Determinants of Outsourcing Unlike the proxies for expected technological change, the coefficients on the other variables in the model are not robust. In Column (1) of Table 2, only five variables (market expansion, average labor cost, age, size of firm and foreign ownership) are significant. As shown in previous research (Abraham and Taylor, 1996; Holl, 2007) the volatility of demand for the product as proxied by market expansion, is positively correlated with the probability of outsourcing. High average labor costs are also associated with outsourcing, as shown in previous work (Abraham and Taylor, 1996). The relationship between outsourcing and age is an inverted-U. 18 In our sample, large firms are more likely to outsource which is at odds with the prediction of our model since the sunk cost per unit of production decreases with firm size. 19 Finally, foreign-owned firms are less likely to outsource perhaps because of lack of sufficient knowledge about

18

If we delete the quadratic in age, the coefficient on the linear term is positive but insignificant.

19

Holl (2007) suggest that this positive and significant relationship between subcontracting and firm size may be due to large firms having greater capacity to establish and manage subcontracting relationships.

17

outsourcing opportunities in the local market. 20 None of these effects remain significant when we use fixed effects or first differences, indicating that findings from previous studies that rely on cross-sectional analyses may not be robust. V. Conclusions The role of technology as a factor in firms’ outsourcing decisions has received increasing attention. Previous research has focused on the roles played by asset specificity, technological diffusion, and information and communications technology. In this paper we propose an alternative perspective on the relationship between technological change and outsourcing. We present a dynamic model that shows that outsourcing becomes more beneficial to the firm when technology is changing rapidly. The intuition behind the model is that as the pace of innovations in production technology increases, the less time the firm has to amortize the sunk costs associated with purchasing the new technologies. This makes producing in-house with the latest technologies relatively more expensive than outsourcing. The model therefore provides an explanation for the recent increase in outsourcing that has taken place in an environment of increased expectations for technological change. We test the predictions of the model using a panel dataset on Spanish firms for the time period 1990 through 2002. Our econometric analysis controls for unobservable fixed characteristics of the firms and also uses an exogenous measure of expected technological change, i.e. the number of patents granted by the U.S. patents office, in the Spanish firm’s industrial sector. The results support the main prediction of the model, namely, 20

Girma and Gorg (2004) find that foreign-owned firms are more likely to engage in international outsourcing.

18

that all other things equal, the demand for outsourcing increases with expected technological change. Interestingly, while the existing literature has found evidence that other variables play a role in the decision to outsource, we find no such evidence here when accounting for firms’ fixed effects.

19

References Abraham, K. and S. Taylor (1996). “Firms’ Use of Outside Contractors: Theory and Evidence.” Journal of Labor Economics 14: 394-424. Abramovsky, L. and R. Griffith (2006). “Outsourcing and Offshoring of Business Services: How Important is ICT?” Journal of the European Economic Association 4:594601. Amiti, M. and S.J. Wei (2006). “Service Offshoring and Productivity: Evidence from the United States”, NBER Working Paper 11926. Antras, P. and E. Helpman (2004). “Global Sourcing”, Journal of Political Economy, 112:552-580. Arndt, S.W. and H. Kierzkowski (2001). Fragmentation: New Production Patterns in the World Economy (New York: Oxford University Press). Autor, D. (2002). “Outsourcing at Will: Unjust Dismissal Doctrine and the Growth of Temporary Help Employment.” Journal of Labor Economics Autor, D. (2003). “Why Do Temporary Help Firms Provide Free General Skills Training?” Quarterly Journal of Economics, 116 (4) November 2003. Baccara, M. (2007). “Outsourcing, Information Leakage and Consulting Firms,” The Rand Journal of Economics, 38:269-289. Baker, G. and Hubbard, T. (2003): “Make versus Buy in Trucking: Asset Ownership, Job Design and Information,” American Economic Review, June: 551-572. Diaz-Mora, C. (2005): “Determinants of Outsourcing Production: A Dynamic Panel Data Approach for Manufacturing Industries.” Foundation for Applied Economics Studies working paper. Feenstra, R.C. and G.H. Hanson (1999). “The Impact of Outsourcing and HighTechnology Capital on Wages: Estimates for the United States, 1979-1990”, Quarterly Journal of Economics 114:907-941. Girma and Gorg (2004). “Outsourcing, Foreign Ownership, and Productivity: Evidence from UK Establishment-level Data,” Review of International Economics 12:817-832. Gorg, H., A. Hanley and E. Strobl (2007). “Productivity Effects of International Outsourcing: Evidence from Plant-level Data”, Canadian Journal of Economics. Gorg, H. and A. Hanley (2007). “International Services Outsourcing and Innovation: An Empirical Investigation”. working paper. Grossman, S. and O. Hart (1986). “The Costs and Benefits of Ownership: A Theory of Vertical and Lateral Integration”, Journal of Political Economy 94(4): 691-719.

20

Grossman, S. and E. Helpman (2002). “Integration versus Outsourcing in Industry Equilibrium,” Quarterly Journal of Economics 117:85-120. Hall, B. H., A. B. Jaffe, and M. Tratjenberg (2001). "The NBER Patent Citation Data File: Lessons, Insights and Methodological Tools." NBER Working Paper 8498. Holl, A. (2007). “Production Subcontracting and Location,” Foundation for Applied Economics Studies working paper. Lileva, A. and J. Van Biesebroeck (2008). “Outsourcing When Investments are Specific and Complementary.” NBER Working Paper No. 14477, November. Lopez, A. (2002). “Subcontratacion de servicios y produccion: evidencia par alas empresas manufactureras espanolas,” Econmia Indsutrial, 348:127-140. Magnani, E (2006). “Technological Diffusion, the Diffusion of Skill and the Growth of Outsourcing in US Manufacturing,” Economics of Innovation and New Technology Mol, M. (2005). “Does Being R&D-Intensive Still Discourage Outsourcing? Evidence from Dutch Manufacturing,” Research Policy 34:571-582. Ono, Y. (2007). “Outsourcing Business Services and the Scope of Local Markets,” Regional Science and Urban Economics 37: 220-238. Williamson, O. (1975). Markets and Hierarchies: Analysis and Antitrust Implications, New York: The Free Press. Williamson, O. (1985). The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting, New York: The Free Press. Wooldridge, J. (2002). Econometric Analysis of Cross Section and Panel Data, Cambridge: MIT Press.

21

Appendix A Model of the Demand for Outsourcing

Using the framework provided in Section II, we can derive the demand for outsourcing in periods 1 and 2. The model is solved by proceeding backwards. Suppose there is technological change in the second period. The firm faces three alternatives: it can outsource, it can produce in-house with the old technology or it can pay the sunk cost and upgrade to the new vintage of machines. To simplify the analysis we make the assumption that upgrading always dominates keeping the old technology. This requires a not too large sunk cost, namely s ≤ pk y θθ21−θθ2 1 , which we assume to hold, i.e.,

(

G ( s) = 1

)

⎛ θ − θ1 ⎞ ⎟⎟ for s ≥ p k y⎜⎜ 2 ⎝ θ 2θ1 ⎠

In this case, the firm’s decision is essentially as in the static model analyzed in the text: to outsource or to produce in-house. Because producing in-house requires incurring the sunk cost, the firm’s decision in the first period does not affect its current decision. Then, as in (3) in the text, when there is a technological improvement, the outsourcing decision is

(

)

⎧1 if s ≥ p2λ − θpk y 2 ⎪⎪ χ2 = ⎨ ⎪0 else ⎪⎩

(A-1)

If there is no technological change in the second period, the firm’s outsourcing decision depends on its outsourcing decision in the first period. If the firm produced inhouse in the first period, it already paid the sunk cost s for the use of the technology and therefore it will outsource only if θp1k y > p2 y. But, if the firm outsourced in the first period, the cost of using the old technology in-house is

pk

θ1

y + s which needs to be

compared to the cost of outsourcing, p2 y. This gives the following outsourcing decision in period 2 in the absence of technological change,

(

)

⎧1 if χ1 × s ≥ p2 − θpk y 1 ⎪⎪ χ2 = ⎨ ⎪0 else ⎪⎩

(A-2)

Because of the presence of sunk costs, the decision to outsource during the first period affects future decisions if technological change does not occur. Specifically, a firm that did not outsource during the first period, χ1 = 0, will be less likely to outsource during the second period if no technological change occurs. 22

In the first period, the firm chooses to outsource or produce in-house by comparing the expected discounted costs of each alternative. These costs are, ⎧ ⎫ ⎧ ⎫ p p C ( χ1 = 1) = p1 y + β (1 − λ ) Min ⎨ p2 y, k y + s ⎬ + βλ Min ⎨ p2 λ y, k y + s ⎬ θ1 θ2 ⎩ ⎭ ⎩ ⎭ C ( χ1 = 0) =

⎧ p y + s + β (1 − λ ) Min ⎨ p2 y, k θ1 θ1 ⎩ pk

⎫ ⎧ ⎫ p y ⎬ + βλ Min ⎨ p2 λ y, k y + s ⎬ θ2 ⎭ ⎩ ⎭

where β is the discount factor. The difference between these two costs is ⎛ p ⎞ C ( χ1 = 1) − C ( χ1 = 0) = ⎜ p1 − k ⎟ y − (1 − β (1 − λ ) ) s + θ1 ⎠ ⎝ p ⎞ p ⎪⎧⎛ ⎪⎫ ⎪⎧⎛ β (1 − λ ) Min ⎨⎜ p2 − k ⎟ y − s, 0 ⎬ − β (1 − λ ) Min ⎨⎜ p2 − k θ1 ⎠ θ1 ⎩⎪⎝ ⎭⎪ ⎩⎪⎝

⎫⎪ ⎞ ⎟ y, 0 ⎬ ⎪⎭ ⎠

which can be written as,

(

)

(

)

(

)

(

)

⎧ p1 − θpk y − s p2 − θp1k y ≤ 0 1 ⎪ ⎪ ⎪⎪ p p p C ( χ1 = 1) − C ( χ1 = 0) = ⎨ p1 − θ1k y + β (1 − λ ) p2 − θ1k y − s, 0 ≤ p2 − θ1k y ≤ s (A-3) ⎪ ⎪ ⎪ pk p2 − θp1k y ≥ s ⎪⎩ p1 − θ1 y − (1 − β (1 − λ ) ) s,

(

)

(

(

)

)

We can rule out the first case because it is reasonable to assume that the price of x in period 2 (the last period) cannot be below the marginal cost of in-house production, i.e., 21 p2 ≥

pk

θ1

(A-4)

The firm decides to outsource in the first period whenever C ( χ1 = 1) − C ( χ1 = 0) < 0. The decision to outsource in the first period therefore depends on current and future prices of x as well as on the size of sunk costs and the probability of technological change =,

21

This assumption depends on the type of technology and competition among the firms producing x.

23

(

)

(

)

(

)

⎧1 if s ≥ p1 − θpk y + β (1 − λ ) p2 − θpk y and s ≥ p2 − θpk y 1 1 1 ⎪ ⎪ ⎪ p χ1 = ⎨1 if s ≥ ( p1 − θ1k ) y and s ≤ p − pk y 2 1− β (1− λ ) θ1 ⎪ ⎪ ⎪ ⎩0 else

(

)

(A-5)

From (A-5) we see that, all other things equal, an increase in λ decreases the cost of outsourcing relative to that of in-house production making the firm more likely to decide to outsource in the first period. We use this model to generate a reduced-form equation for the demand for outsourcing in any period. There are three different threshold values in (9) that determine the demand for ⎛

outsourcing in period 1: ⎜⎜ p1 − ⎝

⎛ pk ⎜⎜ p 1 − θ1 ⎝ 1 − β (1 −

⎛ pk ⎞ p ⎟⎟ y + β (1 − λ )⎜⎜ p 2 − k θ1 ⎠ θ1 ⎝

⎞ ⎛ p ⎞ ⎟⎟ y , ⎜⎜ p2 − k ⎟⎟ y and θ1 ⎠ ⎠ ⎝

⎞ ⎟⎟ y ⎠ . Because demand for outsourcing depends on s being above or λ)

below these thresholds, their relative magnitudes are important. Among the different ways these three thresholds can be ranked, only two configurations are feasible, namely, ⎛ p ⎞ ⎜⎜ p 2 − k ⎟⎟ <> θ1 ⎠ ⎝

⎛ ⎛ p ⎞ p ⎞ ⎜⎜ p1 − k ⎟⎟ + β (1 − λ )⎜⎜ p 2 − k ⎟⎟ θ1 ⎠ θ1 ⎠ ⎝ ⎝

> <

⎛ p ⎞ ⎜⎜ p1 − k ⎟⎟ θ1 ⎠ ⎝ 1 − β (1 − λ )

which can we written more compactly as ⎛ p ⎞ ⎜⎜ p 2 − k ⎟⎟ <> θ1 ⎠ ⎝

⎛ p ⎞ ⎜⎜ p1 − k ⎟⎟ θ1 ⎠ ⎝ 1 − β (1 − λ )

> ( p1 − θp1k ) defines two price regions which determine demand for < 1− β (1−λ ) outsourcing in the first period,

Thus,

(

p2 − θp1k

)

24

( p1 − θp1k ) y ( p1 − θp1k ) y ⎧ pk 1 if s when p y ≥ − ≥ 2 θ1 1− β (1− λ ) 1− β (1− λ ) ⎪ ⎪ ⎪⎪ χ1 = ⎨ ( p1 − θp1k ) y pk pk pk 1 if s p y (1 ) p y when p y ≥ − + − − − ≤ β λ 1 2 2 θ1 θ1 θ1 1− β (1− λ ) ⎪ ⎪ ⎪ ⎪⎩0 else

(

(

)

)

(

)

(

)

which implies . ⎧ ( p1 − θp1k ) ⎛ ( p1 − θp1k ) ⎞ pk 1 G y if p − − ≥ 2 − − − β (1− λ ) ⎜ ⎟ 1 (1 ) 1 β λ θ ⎪ 1 ⎝ ⎠ ⎪⎪ E ( χ1 ) = ⎨ ⎪ ⎪1 − G p1 − θp1k y + β (1 − λ ) p2 − θp1k y ⎪⎩

(

((

)

)

(

) ) if ( p

)

(p − ) pk

pk 1 2 − θ1 ≤ 1− β (1− λ ) 1

θ

Notice that, for given prices, 1 − G (⋅) increases with λ so that the demand for outsourcing in period 1 is increasing in the probability of technological change λ. The demand for outsourcing in period 2 is given by E ( χ 2 ) = λ E ( χ 2 | T2 = 1) + (1 − λ ) ⎡⎣ E ( χ 2 | T2 = 0, χ1 = 1) E ( χ1 ) + E ( χ 2 | T2 = 0, χ1 = 0 )(1 − E ( χ1 ) ) ⎤⎦ where T2 is an indicator for technological change occurring in period 2. In this simple model, E ( χ 2 | T2 = 0, χ1 = 0 ) = 0 because a firm that did not outsource in the first period will continue producing in-house in the second period if θ does not change. Expected outsourcing demand in the second period is therefore ⎡ ⎡ ⎛⎛ ⎛⎛ p ⎞ ⎞⎤ p ⎞ ⎞⎤ E ( χ 2 ) = λ ⎢1 − G ⎜⎜ ⎜ p2 λ − k ⎟ y ⎟⎟ ⎥ + (1 − λ ) ⎢1 − G ⎜⎜ ⎜ p2 − k ⎟ y ⎟⎟ ⎥ E ( χ1 ) θ 2 ⎠ ⎠ ⎦⎥ θ1 ⎠ ⎠ ⎦⎥ ⎝⎝ ⎝⎝ ⎣⎢ ⎣⎢

(A-6)

Given prices, E ( χ 2 ) is an increasing function of λ when the price margin after an innovation occurs, p2 λ − θp2k , is no larger than the price margin when there is no technology change, p2 − θp1k . This sufficient condition is likely to hold under many market structures because the number of potential outsourcers is larger after an innovation occurs prompting supplying firms to lower their markups in order to capture a larger market share.

25

Table 1 Mean Values of Outsourcing, by Industry Sector (1990-2002)

Industry

Proportion of Firms Outsourcing Production

Value of Outsourcing Divided by Total Costs

Meat-processing industry Foodstuffs and tobacco Drinks Textiles Leather and footwear Wood industry Paper Editing and Printing Chemicals Rubber and plastics Non-metallic minerals products Iron and steel Metallic Products Machinery and mechanical goods Office machinery, computers, processing, Electrical and electronic machinery Motor vehicles Other transport material Furniture Other manufacturing industries

0.162 0.242 0.161 0.451 0.339 0.291 0.296 0.56 0.388 0.458 0.258 0.284 0.463 0.613 0.589 0.578 0.521 0.588 0.367 0.503

All 0.008 0.019 0.023 0.061 0.050 0.033 0.024 0.076 0.025 0.038 0.021 0.023 0.053 0.098 0.059 0.062 0.065 0.089 0.041 0.050

If >0 0.045 0.080 0.138 0.134 0.147 0.116 0.080 0.137 0.066 0.083 0.081 0.082 0.112 0.161 0.100 0.107 0.128 0.155 0.113 0.098

All Industries (1990-2002)

0.412

0.047

0.114

1990 (N=2189) 1994 (N=1876) 1998 (N=1776) 2002 (N=1708)

0.351 0.416 0.467 0.429

0.042 0.044 0.054 0.048

0.120 0.107 0.117 0.112

26

Table 2 The Likelihood of Outsourcing Production, 1990-2002# Logit Regressions 1st diffs Using US Patents fixed effects Logit 1st difs 1 2 3 4 5 US patents (log of 5 years mean) Carried out or contracted R&D Normal for firm to change kind of products offered Obtained product innovation If registered patents Product very standardized market_expand market_decline If using computer digital machine tools If using robotic average labor cost per employee (000 of Euros) Foreign participation in firm capital export value divided by sales age of firm age2 capacity utilization Total Sales deflated for 2002 (000 of Euros) Employees: 21-50 51-200 201-500 501+ N

0.0879*** (0.0186) 0.0868*** (0.0174) 0.0821*** (0.0178) 0.0830** (0.0258) -0.0220 (0.0180) 0.0499** (0.0156) 0.0305 (0.0186) 0.0247 (0.0155) 0.0236 (0.0196) 0.0031*** (0.0008) -0.0458* (0.0221) 0.0601 (0.0364) 0.0026** (0.0009) -0.0000* (0.0000) 0.0004 (0.0005) 0.0000 (0.0000) 0.0407 (0.0218) 0.0368 (0.0284) 0.0971** (0.0311) 0.0964* (0.0393) 6911

0.0830* (0.0325) 0.0037 (0.0248) 0.0547* (0.0251) -0.0018 (0.0392) -0.0457 (0.0351) 0.0263 (0.0213) 0.0316 (0.0253) 0.0159 (0.0240) -0.0020 (0.0297) -0.0008 (0.0016) 0.0119 (0.0473) 0.1508 (0.0842)

0.0543* (0.0230) 0.0151 (0.0186) 0.0427* (0.0188) 0.0162 (0.0272) -0.0276 (0.0239) 0.0233 (0.0162) 0.0164 (0.0184) 0.0029 (0.0176) 0.0046 (0.0221) -0.0014 (0.0014) 0.0263 (0.0371) 0.1044 (0.0622)

0.0002 (0.0007) -0.0000 (0.0000) -0.0156 (0.0466) 0.0331 (0.0667) 0.0635 (0.0733) 0.1012 (0.0742) 2479

0.0004 (0.0005) 0.0000 (0.0000) -0.0000 (0.0000)

3787

0.0179** (0.0066) 0.0966*** (0.0186) 0.1221*** (0.0174) 0.0809*** (0.0175) 0.0976*** (0.0251) -0.0793*** (0.0170) 0.0376* (0.0156) 0.0503** (0.0189) 0.0481** (0.0156) 0.0203 (0.0193) 0.0014 (0.0008) -0.0255 (0.0220) 0.0578 (0.0354) 0.0027** (0.0010) -0.0000* (0.0000) 0.0010* (0.0005) -0.0000 (0.0000) 0.0362 (0.0228) 0.0354 (0.0280) 0.0696* (0.0315) 0.0454 (0.0387) 6499

0.0517* (0.0231) 0.0634** (0.0234) 0.0230 (0.0192) 0.0415* (0.0193) 0.0159 (0.0280) -0.0244 (0.0241) 0.0147 (0.0166) 0.0172 (0.0194) -0.0025 (0.0183) 0.0066 (0.0226) -0.0015 (0.0015) 0.0236 (0.0376) 0.0858 (0.0618)

0.0003 (0.0005) 0.0000 (0.0000) -0.0000 (0.0000)

3536

#

In parentheses are robust standard errors clustered by firm id. Marginal effects are reported for Logit regressions. “1st diffs” are estimated using OLS All regressions include year dummies. Regression (1) includes industry dummies. * p<0.05, ** p<0.01, *** p<0.001

27

Table 3 Tobit Regressions on the Share of Outsourcing in Total Costs* 1 US patents (log of 5 years mean) Carried out or contracted R&D Normal for firm to change kind of products offered Introduced product innovation If registered patents Product very standardized market_expand market_decline If using computer digital machine tools If using robotic average labor cost per employee (000 of Euros) Foreign participation in firm capital export value divided by sales age of firm age2 capacity utilization Total Sales deflated for 2002 (0000000 of Euros) Employees: 21-50 51-200 201-500 501+

0.0131* (0.0053) 0.0726*** (0.0143) 0.0723*** (0.0127) 0.0340** (0.0128) 0.0474* (0.0190) -0.0583*** (0.0130) 0.0346** (0.0119) 0.0347* (0.0146) 0.0263* (0.0120) 0.0089 (0.0145) 0.0015* (0.0006) -0.0315 (0.0167) 0.0675* (0.0272) 0.0008 (0.0007) -0.0000* (0.0000) 0.0008* (0.0004) 0.0912** (0.0305) 0.0234 (0.0176) 0.0268 (0.0214) 0.0435 (0.0234) 0.0168 (0.0289) 6352

2 0.0019* (0.0008) 0.0105*** (0.0021) 0.0106*** (0.0019) 0.0049** (0.0019) 0.0069* (0.0028) -0.0084*** (0.0019) 0.0050** (0.0017) 0.0050* (0.0021) 0.0038* (0.0017) 0.0013 (0.0021) 0.0002* (0.0001) -0.0045 (0.0024) 0.0097* (0.0039) 0.0001 (0.0001) -0.0000* (0.0000) 0.0001* (0.0001) 0.0131** (0.0044) 0.0034 (0.0025) 0.0039 (0.0031) 0.0063 (0.0034) 0.0024 (0.0042) 6352

3 0.0024* (0.0010) 0.0133*** (0.0027) 0.0135*** (0.0025) 0.0062** (0.0024) 0.0088* (0.0037) -0.0106*** (0.0024) 0.0063** (0.0022) 0.0064* (0.0027) 0.0047* (0.0022) 0.0016 (0.0026) 0.0003* (0.0001) -0.0056 (0.0029) 0.0121* (0.0049) 0.0002 (0.0001) -0.0000* (0.0000) 0.0001* (0.0001) 0.0164** (0.0055) 0.0043 (0.0032) 0.0049 (0.0040) 0.0080 (0.0044) 0.0031 (0.0053) 6352

N * p<0.05, ** p<0.01, *** p<0.001 * Marginal effects are reported. Column (1) shows the marginal effects for the probability of being uncensored; Column (2) shows the marginal effects for the expected value of the dependent variable conditional on being uncensored, E(y | a
All regressions include industry dummies and year dummies. Robust standard errors clustered by firm ID. The Tobit coefficients are given in Column (4) of Appendix Table A-1.

28

Appendix Table A-1 Dependent Variable: Share of Outsourcing in Total Costs Tobit Regressions

1 US patents (log of 5 years mean) Carried out or contracted R&D Normal for firm to change kind of products offered Obtained product innovation Product very standardized If registered patents market_expand market_decline If using computer digital machine tools If using robotic average labor cost per employee (000 of Euros) Foreign participation in firm capital export value divided by sales age of firm age2 capacity utilization Total Sales deflated for 2002 (0000000 of Euros) Employees: 21-50 51-200 201-500 501+ N * p<0.05, ** p<0.01, *** p<0.001

0.0316*** (0.0067) 0.0345*** (0.0061) 0.0223*** (0.0064) -0.0051 (0.0061) 0.0242** (0.0091) 0.0240*** (0.0059) 0.0083 (0.0072) 0.0066 (0.0057) 0.0040 (0.0069) 0.0015*** (0.0003) -0.0231** (0.0074) 0.0300* (0.0119) 0.0004 (0.0003) -0.0000* (0.0000) 0.0004* (0.0002) 0.0524*** (0.0133) 0.0135 (0.0075) 0.0042 (0.0091) 0.0159 (0.0099) 0.0075 (0.0124) 6765

Standard 2 0.0049* (0.0023) 0.0380*** (0.0071) 0.0540*** (0.0064) 0.0244*** (0.0067) -0.0321*** (0.0059) 0.0333*** (0.0095) 0.0202** (0.0063) 0.0207** (0.0077) 0.0174** (0.0060) 0.0050 (0.0072) 0.0007* (0.0003) -0.0208** (0.0077) 0.0341** (0.0121) 0.0007* (0.0003) -0.0000** (0.0000) 0.0006*** (0.0002) 0.0475*** (0.0136) 0.0102 (0.0083) 0.0038 (0.0096) 0.0077 (0.0105) -0.0051 (0.0129) 6352

Random Effect 4 0.0063* (0.0026) 0.0299*** 0.0346*** (0.0065) (0.0068) 0.0220*** 0.0342*** (0.0058) (0.0060) 0.0156** 0.0162** (0.0059) (0.0061) -0.0069 -0.0278*** (0.0063) (0.0062) 0.0182* 0.0224* (0.0086) (0.0089) 0.0192*** 0.0165** (0.0054) (0.0056) 0.0089 0.0165* (0.0065) (0.0069) 0.0065 0.0126* (0.0055) (0.0057) 0.0032 0.0042 (0.0067) (0.0069) 0.0013*** 0.0007* (0.0003) (0.0003) -0.0156* -0.0152 (0.0079) (0.0081) 0.0265* 0.0324* (0.0127) (0.0130) 0.0003 0.0004 (0.0003) (0.0003) -0.0000 -0.0000* (0.0000) (0.0000) 0.0002 0.0004* (0.0002) (0.0002) 0.0475** 0.0437** (0.0145) (0.0146) 0.0138 0.0112 (0.0078) (0.0083) 0.0134 0.0128 (0.0097) (0.0101) 0.0267* 0.0206 (0.0106) (0.0110) 0.0189 0.0080 (0.0132) (0.0137) 6765 6352 3

29

Appendix Table A-2 Number of Patent Applications in the US (1980-2002) Field Mean Median 1 Food, beverages 593.96 610 2 Tobacco products 88.39 100 3 Textile 157.79 169.5 4 Wearing apparel 221.13 227 5 Leather articles 180.87 175 6 Wood products 119.43 123 7 Paper 416.75 443 9 Petroleum products, nuclear fuel 529.46 572 10 Basic chemical 6456.83 6981 11 Pesticides, agro-chemical products 572.67 551.5 12 Paints, varnishes 2.2 2 13 Pharmaceuticals 5830.04 5216.5 14 Soaps, detergents, toilet preparation 284.13 238.5 15 Other chemicals 1120.63 1242.5 16 Man-made fibers 37.86 32 17 Rubber and plastics products 3335.46 3779 18 Non-metallic mineral products 2633.79 2828 19 Basic metals 1355.67 1493 20 Fabricated metal products 2512.08 2818.5 21 Energy machinery 3081.67 3437 22 Non-specific purpose machinery 3273.96 3613.5 23 Agricultural and forestry machinery 1075.88 1195.5 24 Machine-tools 2376.58 2574.5 25 Special purpose machinery 5998.04 6352.5 26 Weapons and ammunition 384.52 407 27 Domestic appliances 1586.33 1682.5 28 Office machinery and computers 9394.38 7528.5 29 Electric motors, generators, transformers 656.29 617.5 30 Electric distribution, control, wire, 1617.79 1553 31 Accumulators, battery 432.5 328.5 32 Lighting equipment 355.63 358.5 33 Other electrical equipment 1148.46 1142 34 Electronic components 4957.29 4295 35 Signal transmission, telecommunication 5057.17 4146 36 TV & radio receivers, audiovisual 1653.71 1446 electronics 37 Medical equipment 4489.79 4260.5 38 Measuring instruments 4078.46 4336 39 Industrial process control equipment 561.67 566 40 Optical instruments 2523.67 2568 41 Watches, clocks 134.18 136 42 Motor vehicles 4180.88 4428.5 43 Other transport equipment 1201.38 1352.5 44 Furniture, consumer goods 3026.21 3098 Total 2109.379 1115

StD 296.68 31.83 65.62 106.31 68.52 48.57 220.79 193.45 2389.2 281.43 1.11 3520.13 154.86 431.63 15.22 1420.08 1169 492.84 955.08 1188.41 1209 420.95 910.51 2349.41 122.81 664.14 7281.94 297.92 751.64 285.69 191.75 542.9 3476.51 3615.63 1065.79 2714.85 1728.3 237.68 1355.52 34.06 1786.76 478.6 1497.76 2763.5

30