Offshoring and Intellectual Property Rights Reform

Transcription

1 Claudia Canals 1 la Caixa and Fuat Şener Union College This Draft: August 5, 2012 Abstract This paper empirically assesses the responsiveness of US offshoring to intellectual property rights (IPR) reforms in 16 countries. To this end, we construct a measure of US offshoring at the industry level based on intermediate goods trade using panel data that covers 23 industries and 16 countries for the period We then conduct a difference-indifference analysis using the IPR reform years proposed in Branstetter et al. (2006, 2010). The analysis differentiates between high-tech and low-tech industries, and controls for country, time and industry effects. According to our findings, IPR reforms have a significant and substantial effect on offshoring. In particular, following IPR reform, aggregate offshoring increases by 27% in high-tech industries, although the effect on aggregate offshoring is not statistically significant in low-tech industries. Distinguishing offshoring activities by technology-content, we find that, after IPR reforms, high-tech offshoring activities increase by 35% in high-tech industries, while low-tech offshoring activities increase by 22% in low-tech industries. Hence, the results suggest that, following IPR reform, industries tend to increase their core offshoring, that is, they offshore more activities that are in line with the technology classification of their own industry. Nonetheless, when we narrow down our measure to intra-industry offshoring (that is, offshoring that takes place within the same industry), we observe that while high-tech industries do increase their intra-industry offshoring by 60%, low-tech industries decrease it, though at marginally significant levels. Key words: Intellectual Property Rights, Technology, Multinational Firms, Outsourcing, Offshoring, Subcontracting. JEL Codes: F1, F23, 033, The views expressed in this Working Paper are those of the authors only and do not necessarily represent those of la Caixa. Acknowledgements later to be inserted in this footnote.

2 1. Introduction and Literature Review Over the past three decades, global trade policies and the composition of world trade have changed dramatically. Many developing countries (the South) started raising their Intellectual Property Rights (IPRs) protection levels and building the necessary institutional framework. This transformation received an additional push in 1995 with the signing of the TRIPS agreement (Trade Related Aspects of IPRs) under the World Trade Organization umbrella, which called for establishing at least a minimum level of IPR protection by Despite the global movement toward stronger IPR protection, the issue of IPR reform has generated intense debate among policy makers. The proponents of TRIPS have argued that such a move would reduce the imitation risk faced by Multinational Firms (MNFs) and would encourage more technology transfer and overseas activities. Moreover, it was emphasized that a stronger IPR regime would boost innovation incentives for all firms and thus accelerate global technological progress. Meanwhile, opponents have argued that TRIPS would simply lead to a transfer of rents from the South to the developed world (the North) and hinder the South s ability to implement newly-invented Northern technologies. Thus, the prospects of increased trade and MNF activity were the central motivations for the South to raise their IPR protection and further sign on to TRIPS. During the past three decades, North-South trade has indeed expanded dramatically. A central feature of this new era of globalization has been the rapid rise in intermediate goods trade, a phenomenon which is referred to as offshoring. More specifically, Northern producers fragmented their production processes into a variety of intermediate goods/services, sometimes referred as tasks 3, and distributed them across the globe seeking the lowest-cost production locations. Production of intermediate products took place either within the boundaries of the MNF (through foreign affiliates) or outside the boundaries (through arm s length subcontracting to Southern producers). A large literature has emerged, documenting and investigating the acceleration in trade and offshoring. 4 Interestingly, only a few studies have empirically analyzed the effects of IPR regime changes on trade and MNF activities. Maskus and Penubarti (1995) and Smith (1999) examined the relationship between IPR protection and trade using US data at the industry and state level, respectively. Branstetter et al. (2006, 2010) examined the link between IPR and MNF 2 See Maskus (2000) for a comprehensive overview of IPRs and globalization. 3 See Grossman and Rossi-Hansberg (2008). 4 See Feenstra (1998) and Campa and Goldberg (1997) for an overview of offshoring by advanced countries. See Hummels et al. (2001) for an empirical investigation of vertical specialization in world trade. See Tang (2006) for an empirical analysis of how declining communication costs affect trade in differentiated goods. See Baier et al (2001) for an empirical analysis of rising trade levels and their causes. Most of the literature identifies reductions in tariffs, transportation and communication costs as the main driving forces behind the expansion in trade and offshoring. The notable exception is Baier et al. (2001) who find that two-thirds of the growth in world trade can be explained by income growth. C. Canals and F. Şener 2

3 activity by considering a variety of measures (such as affiliate assets, affiliate R&D, affiliate overseas sales, intra-firm royalty payments and such). However, none of these papers examine the issue by using an offshoring measure based on intermediate goods trade. In this paper, our objective is to first construct a measure of US offshoring at the industry level that is mapped against the trading partners of the US and that at the same time exploits the notion of intermediate-goods-based offshoring. In doing so, we follow a methodology similar to the one used by Feenstra and Hanson (1996, 1999). We use industry-level data from the US that covers the period We then empirically assess the responsiveness of offshoring to Southern IPR protection reforms. We identify the timing of IPR reforms in a total of 16 developing countries using the data set assembled by Branstetter et al. (2006). The years where major patent reforms were undertaken for the group of countries studied are listed in Table 1. Our empirical strategy is to conduct a difference-in-difference analysis in the spirit of Branstetter et al. (2006, 2010). More specifically, we regress our offshoring measure on a dummy variable which takes the value of zero before reform and one after reform, controlling for industry, time, and country effects, as well as country-specific time trends. Table 1 Country - Patent reform year Argentina 1996 Brazil 1997 Chile 1991 China 1993 Colombia 1994 Indonesia 1991 Japan 1987 Mexico 1991 Philippines 1997 Portugal 1992 S. Korea 1987 Spain 1986 Taiwan 1986 Thailand 1992 Turkey 1995 Venezuela 1994 Source: Branstetter et al (2006) and (2010) Our empirical findings imply that IPR reforms can exert a significant and substantial effect on aggregate offshoring at the industry-level. In particular, for high-tech industries, IPR reform results in an increase in aggregate offshoring by 27%, whereas for low-tech-industries IPR reform does not exert a statistically significant impact on aggregate offshoring. We also investigate the changes in the composition of offshoring within each industry by distinguishing between offshoring activities as high-tech and low-tech. We find that after IPR reform, hightech offshoring activities increase on average by 35% in high-tech industries, while low-tech offshoring activities increase on average by 22% in low-tech industries. In other words, following IPR reforms, industries tend to increase their core offshoring activites, which we C. Canals and F. Şener 3

4 define as offshoring activities that are in line with the technology classification of the source industry. Finally, some striking results arise when we narrow down offshoring to intra-industry offshoring, which we define as offshoring activities that take place within the same industry. We observe that while high-tech industries increase their intra-industry offshoring after IPR reform by at most 60%, low-tech industries decrease it, though at marginally significant levels. Thus, these results suggest that the increase in core offshoring by low-tech industries is driven by low-tech inter-industry offshoring and not by intra-industry offshoring. Our paper differs from the empirical literature on a number of accounts. First, we construct a quite unique intermediate goods-based offshoring measure for each US industry disaggregated at the source country level. 5 Hence, our model differs from the existing literature, which, to our knowledge, has mostly used finals goods trade, MNF related sales and payment levels but not an intermediate-goods-based measure. 6 Second, our offshoring measure captures MNF activity, albeit at the industry level, that takes place both within and outside the boundaries of the firm. This differs from Branstetter et al. (2006, 2010) who focus on intra-firm MNF activity measures and also from Javorcik (2004) who uses a binary FDI measure which equals one if a firm has invested in a country and zero otherwise. Third, we are able to distinguish between offshoring types by each industry as high-tech and low-tech and thereby investigate the response of technology content of offshoring to IPR regime shifts. Our utilization of a rich information set (IO tables coupled with bilateral import data) enables us to construct these distinct measures in a straightforward way. The existing literature does not have this type of distinction, neither at the firm nor at the industry level. We should note that the qualitative direction of our empirical findings is in line with the literature in the sense that most papers, especially the recent ones, find that IPR reform increases trade or MNF activities. 7 Our paper is also related to a large body of theoretical literature that investigates the impact of increased Southern IPR protection on MNF activity, Northern innovation and Southern imitation. This literature can be classified into two categories: quality-ladders models where innovation (R&D) aims at improving the quality of existing products, and variety expansion models where innovation aims at developing new varieties of goods. In both types of models, successful Northern innovators first produce in the North and then explore ways to shift production to the South where production costs are lower. This technology transfer effort 5 Falk and Wolfmayr (2005) use the same approach to construct offshoring by country of origin for seven EU countries. 6 See among others Maskus and Penubarti (1995), Smith (1999), Branstetter et al. (2006, 2010), Ivus (2010). 7 Branstetter et al. (2006, 2010) find that following IPR reform, MNFs expand the scale of their overseas activities. Ivus (2010) finds that strengthening IPRs in developing countries increases the value of exports by developed countries in patent sensitive sector. Javorcik (2004) finds that weak IPR protection discourages foreign investment in technology-intensive sectors. See also Smith (1999), Maskus and Penubarti (1995) and Co (2004). C. Canals and F. Şener 4

5 is usually modeled as a costly and uncertain activity similar to R&D. Success in technology transfer brings forth multinationalization, which implies full or partial transfer of production to the South through foreign direct investment (FDI) or subcontracting. Northern firms producing in the North and MNFs producing in the South can both be subject to imitation risk from the South, which can be reduced by increased Southern IPR protection. The variety-expansion models find that IPR reform increases the rate of FDI [see, among others, Helpman (1993), Lai (1998), Branstetter et al. (2009) and Gustafsson and Segerstrom (2010)] However, none of the papers explicitly considers fragmentation of production but instead assume full shifting of production to the South upon FDI success. The quality- ladders models, on the other hand, offer mixed results. For example Glass and Saggi (2002) find that IPR reform decreases the rate of FDI, whereas Dinopoulos and Segerstrom (2010) have the opposite result. 8 In the quality ladders literature, two papers indeed stand out as highly relevant to our empirical framework since they explicitly incorporate offshoring and fragmentation of production. The first is Sener (2006) who builds on the North-South quality-ladders model of Glass and Saggi (2002) with endogenous innovation, imitation and FDI, and extends this model by incorporating fragmentation of production between the North and the South within each industry. More specifically, in Sener s setting, MNFs offshore an endogenously chosen portion of their production to the South. Similar to Glass and Saggi (2002), Sener (2006) finds that stronger IPRs lowers the rates of imitation, FDI and innovation. However, new results emerge with regards to the offshoring indicators. Southern IPR reform increases both the fraction of multinational industries and the portion of offshored production within each MNF. This finding is also in line with Glass (2004), another relevant paper in the quality-ladders literature. In this paper, Glass constructs a North-South quality ladders model with exogenous offshoring and imitation. She finds that a more stringent IPR regime in the South raises the fraction of offshoring industries. 9 In short, our empirical result that IPR reform leads to more offshoring is in accord with the predictions of most of the recent theoretical work that uses a North-South growth and technology transfer framework. The rest of the paper is organized as follows. Section 2 describes the construction of the offshoring intensity variable and presents some evidence about its evolution for the period. Section 3 presents our empirical specification to estimate the effects of IPR reform on 8 A full comparative discussion of this literature is beyond the scope of our paper. We refer to reader to Glass and Wu (2007) and also Dinopoulos and Segerstrom (2010) for a detailed comparison of quality-ladders vis-à-vis variety expansion models in the context of global IPR reform. 9 Another closely related paper is by Sener and Zhao (2009) who differentiate between R&D races as local-sourcing- and foreign-sourcing -targeted R&D races. In the former type race, innovation success results in Northern production. In the latter, innovation success implies immediate outsourcing to the South (albeit a complete shifting of production to the South), a phenomenon which the authors label as the ipod cycle. Sener and Zhao find that strengthening Southern IPR leads to an increase in the frequency of ipod cycles and a larger fraction of industries engaged in outsourcing. C. Canals and F. Şener 5

6 offshoring intensity. Then, it introduces the data set and the rest of the variables used in the empirical analysis. And finally, it shows the empirical results. Section 4 concludes. 2. Offshoring variable A key aspect of our empirical approach is the construction of a unique measure of offshoring. We follow the spirit of Feenstra and Hanson who in several papers (1996 and 1999, among others) define foreign outsourcing (what we call offshoring) as the import of intermediate inputs. 10 In particular, our offshoring variable measures at each point in time the intermediate goods that a US industry imports from a particular country to produce one dollar worth of value. More specifically, our variable is a measure of offshoring intensity. 11 To construct this measure, we combine Input-Output (IO) tables with bilateral imports for the US. Data on IO coefficients were obtained from the US Bureau of Economic Analysis (BEA). Data on bilateral imports are from the Center for International Data at UC Davis, which were assembled by Robert Feenstra (see Appendix for detailed information regarding the construction of the offshoring variable). 12 For each industry and year, we obtain, from the IO tables, the dollar value of input that industry i gets from industry j to produce one dollar worth of i product at time t, ( a ijt ). From the same source, we can calculate the total consumption by industry j ( C ). Total imports by country c and industry j at each point in time ( M cjt ) come from the bilateral imports dataset. With all this information, we are able to construct the following offshoring variable: jt O cit M = cjt a (1) ijt j C jt Thus, as already mentioned, we are using a slightly extended version of the import proportionality assumption proposed by Feenstra and Hanson (1996, 1999) and used also in the construction of the OECD STAN dataset, among others. In this extended version, we assume that an industry i uses an import of a particular industry j from country c in That is, if the motor vehicle industry uses $0.15 proportion of its total use of that industry j. 13 worth of steel as an input to produce $1 worth of output ( a ijt = 0.15), and 20% of all steel 10 More specifically, they compute the ratio of imported intermediate inputs to total expenditure on non-energy intermediates and call it a broad measure of foreign outsourcing share. 11 We use offshoring and offshoring intensity interchangeably. 12 Data can be downloaded from 13 We extend the import proportionality assumption in the sense that we distinguish imports by country of precedence by using M jct in equation (4). This differs from Feenstra and Hanson (1999) who used instead, M jt, aggregated imports by industry j. This is because they seek to measure offshoring intensity with no particular interest in the source country, whereas our goal is to capture exactly the offshoring intensity mapped against source country. C. Canals and F. Şener 6

7 M jct consumed in the US is imported from Brazil ( 0.2) jt C =, then estimate that to produce $1 worth of motor vehicle industry 3 cents of a dollar of Brazilian steel are used. Moreover, we have to make some additional structural assumptions to accommodate the change in industry classification for the years, following the NIPAs revision in industry codification. 14 Beginning with 2000, the BEA has started clustering certain sets of industries, which resulted in the IO tables containing fewer manufacturing sectors than in previous years. Thus, in order to retrieve the larger set of industries for the period we combine the consumption data and the IO coefficients in the clustered industries with the consumption and IO coefficients in More specifically, we construct the new consumption and IO coefficients for using a particular weighting based on the 1999 data and assuming this to be constant for 2000 and onwards (see the Appendix for a detailed explanation). We do, nevertheless, run robustness checks not including the years, and find that this assumption does not alter our main results. Finally, we exploit the rich information set in the IO tables by distinguishing between hightech- and low-tech- intensive offshoring in each industry, specifically, we calculate: H O cit and L O cit, respectively. More O O H cit L cit M = cjt aijt (2) j H C jt M = cjt aijt (3) j H C jt where j H stands for high-tech industries and comprises the following sectors: industrial and other chemicals, electrical machinery, office machines and automatic data processing machines, motor vehicles and other transportation equipment, and scientific and professional instruments. Our industry classification follows from Branstetter et al. (2010, p.7). Similarly, the industries are also grouped according to this high-tech and low-tech classification. Hence, both high-tech and low-tech industries undertake both high-tech offshoring activities and lowtech offshoring activities. 15 Evolution of the offshoring variable Before entering into the specifics of the empirical approach, we discuss the changes in offshoring intensity before and after IPR reforms. Recall that offshoring intensity refers to 14 NIPAs revision was designed by the BEA to increase consistency between IO tables, GDP industry tables and NIPA tables. 15 For each industry we do compute a narrower offshoring variable (intra-industry offshoring) that considers offshoring that takes place within the same industry. C. Canals and F. Şener 7

8 imported intermediate inputs to produce one dollar worth of product; thus, offshoring values are below the 1$ level. The bars represent the average inputs that an industry uses from a particular country in a given year (according to our sample). 1 shows the large increase in the offshoring intensity after IPR reform, from 0.07 cents for each dollar of output in the average industry-country pair, to close to 0.20 cents for each dollar, a 166% rise. Panel B in Figure 1 shows that the increase in aggregate offshoring after IPR reform is larger for high-tech industries. On average, low-tech industries raise their aggregate offshoring by 136%, while high-tech industries increase it by 223%. Moreover, the right hand figure also illustrates that high-tech industries tend to offshore more per dollar produced, both before and after reform. Figure 1 Offshoring per dollar of output 16, 17 In particular, Panel A in Figure Panel A Panel B Before Reform After Reform % % % 0.00 Before Reform After Reform 0.00 Low Tech industries High Tech industries If we distinguish high-tech offshoring from low-tech offshoring, we observe that after the reform the growth rate in high-tech offshoring intensity (for the average industry-country pair) is more than double the growth rate in low-tech offshoring intensity (264% versus 119%) (see Figure 2). However, there is a base effect here, since the starting level of high-tech offshoring intensity is lower than the starting level of low-tech offshoring intensity. If we further differentiate types of offshoring in high-tech and low-tech industries, we observe that high-tech industries increase their high-tech offshoring by a relatively large margin, while the low-tech industries increase their low-tech offshoring by a smaller margin (see Figure 3). Moreover, for the high-tech industries, the growth rate in high-tech offshoring activities is larger than the growth rate for low-tech offshoring activities after the IPR reform. 16 In order to compute the average in the descriptive statistics, each country, industry and year is given the same weight, that is, it is considered as a data point. Industry value added weights or country imports weights are not used since they are not taken into account in the regression. Thus, these averages cannot be taken as the behavior of the average US industry. 17 Notice that each year, a particular industry uses inputs from numerous countries, more than the ones in our sample, plus from itself. C. Canals and F. Şener 8

9 The same is true for low-tech industries. Though again, the base effect here is also evident, since the starting point of high-tech offshoring for low-tech industries is very low. Figure 2 Offshoring per dollar of output High-tech Offshoring 264% Low-tech Offshoring 119% Before Reform After Reform Before Reform After Reform Figure 3 Offshoring per dollar of output 0.30 High-tech Industries 0.16 Low-tech Industries % Before Reform After Reform Before Reform After Reform 122% % % 0.00 High-tech Offshoring Low-tech Offshoring 0.00 High-tech Offshoring Low-tech Offshoring Finally, if we use more disaggregation at the industry level (see Figure 4), we observe that the top three industries that registered the largest change in offshoring intensity in response to IPR reform are Industrial and other chemicals and Office machines and automatic data processing, and Plastic and Synthetic materials. We note that two of them are in the hightech industry group, whereas one of them is in the low-tech industry group. The bottom three industries in terms offshoring response are Tobacco products, Non-ferrous metals manufacturing and Iron and steel manufacturing. And all three of them are in the low-tech industry group. C. Canals and F. Şener 9

10 Figure 4 Offshoring per dollar of output 18. Office machines and automatic data processing machines High-tech industries Percent Change (%) Motor vehicles and related Electric machinery, equipment and supplies Ophtalmologic and photographic instruments Scientific and controlling instruments Industrial and other chemicals Footwear, leather, and leather products 16. Metal containers 4. Apparel 23. Miscellaneous manufacuring 17. Machinery except electrical 14. Iron and steel manufacturing 3. Textile products 15. Non-ferrous metals manufacturing 1. Food and kindred products 6. Furniture and fixtures 10. Drugs, cleaning and toilete preparations 9. plastic and synthetic materials 5. Lumber and wood products 11. Rubber and miscellaneous plastics 13. Non-metallic mineral manufacturing 2. Tobacco products 7. Paper and similar products Low-tech industries Percent Change (%) Before Reform After Reform C. Canals and F. Şener 10

11 3. Empirical Specification To assess the impact of IPR reform on offshoring, we employ a difference-in-difference approach, following Branstetter et al. (2006, 2010). Our empirical analysis differs from Branstetter et al. in three key aspects. Two of them are largely explained in the previous section, since they refer to the construction of the offshoring variable. First, as our dependant variable, we consider a variety of offshoring measures at the industry-level, in contrast to Branstetter et al. papers where the dependant variables are mostly parent-affiliate measures at the firm level. 18 Even though our measure is at a less disaggregated level (industry level vs. firm level), it has the advantage of capturing a wider spectrum of MNF and Northern firm activity. This is because it includes both parent-affiliate activities that take place within the boundaries of the MNF and also arm s length activities such as subcontracting to outside foreign firms. Second, we have information about each industry s distribution of offshoring across all other industries. Thus, we are able to distinguish between offshoring types within each industry as high-tech offshoring and low-tech offshoring. Such a refinement makes it possible to examine the changes in the technology content of offshoring to IPR reforms. Third, we use time series data from 1973 up to 2006 and thus extend the time period of Branstetter et al. (2010), which starts in 1982 and ends in Methodologically, we track the evolution of offshoring activities performed by each US industry across countries and industries, and over time. We create a dummy variable to capture IPR reform and estimate how offshoring changes in response to reform, controlling for country and industry specific characteristics. The main equation to estimate is: O = α + α + α + β H + β R + β R Tech + ε (4) cit ci t ct 0 ct 1 ct 2 ct i cit where the subindex c identifies the country to which the US is offshoring, i is the industry index, and t is the year index. As already discussed, our measure of offshoring O cit (or offshoring intensity) represents the value of imported intermediate inputs that the US industry i purchases from country c in order to produce one dollar worth of value in year t. In estimating equation (4), we control for country-industry pair fixed effects α ci, time fixed effectsα t, and country-specific time trends α ct, and a number of time-varying country characteristics H ct, which include GDP per capita, GDP, real exchange rate with respect to the US dollar and a measure of trade openness (total trade over GDP). Following Branstetter et al (2006, 2010), we use a Post Reform Dummy variable ( R ct ) that takes the value of one after 18 We should note that Branstetter et al. papers use other dependent variables as well. More specifically, Branstetter et al. (2010) use Southern industry-level data to assess whether IPR reform has a net positive effect on aggregate production per industry. In addition, Branstetter et al. (2006) consider host-country patent data differentiated as resident and non-resident patent filings to estimate the impact of IPR reform on patenting activity. C. Canals and F. Şener 11

12 substantive changes in the patent regime were achieved in the source country c, and zero otherwise. 19 Its coefficient estimates the change in offshoring or offshoring intensity in all industries after a strengthening of the IPR regime occurs. Hence, if offshoring into a country after a patent reform occurs. Similarly, we construct a binary variable β 1 > 0, the US increases its Tech i to distinguish between more technologyintensive and less-technology intensive industries as high-tech and low-tech industries. Particularly, Techi equals one for the High-Tech industries listed in Figure 4. We then interact Techi with the Post Reform Dummy variable R ct. The coefficient β2 on the interaction variable ( Rct Techi ) captures the differential impact of IPR reform on offshoring depending on the technology intensity of the industry. Thus, if β 2 > 0, we can conclude that patent reform exerts an additional impact on offshoring in high-tech industries. 20 We also run two additional regressions for the entire sample (country-industry-time pairs) and use H O cit and L O cit as dependent variables, which measure, respectively, high-tech offshoring and the low-tech offshoring intensity within each industry. The construction of these measures follow from equations (2) and (3). We thus have the following two equations to estimate: O = α + α + α β H + β R + β R Tech + ε (5) H H H H H H H H cit ci t ct 0 ct 1 ct 2 ct i cit O = α + α + α β H + β R + β R Tech + ε (6) L L L L L L L L cit ci t ct 0 ct 1 ct 2 ct i cit The coefficient interpretation is the same as before. By distinguishing between high-tech and low-tech offshoring, we plan to capture the response of technology content of offshoring within each type of industry. Hence, when estimating equations (5) and (6) we are focusing on the effects IPR improvements have on high-tech offshoring activities, and low-tech offshoring activities, respectively. 19 The details on the construction of the IPR reform dummy are provided in Branstetter et al. (2006) and in the Appendix to their paper. Using the IPR reform dummy of Branstetter et al. in the context of a difference-in-difference approach has the following advantages. First, the IPR reform dummy is constructed by considering both the observable features of the relevant IPR laws and their enforcement. This is an advantage over the Ginarte and Park Index, developed in Ginarte and Park (1997) which is heavily based on the written laws as opposed to enforcement considerations. Second, Branstetter et al (2006, Appendix) provide detailed empirical and historical evidence making the case that the timing of IPR reform is likely to be exogenous, in particular, to MNF activity. Third, by using the reform dummy in a disaggregated panel data set with a difference-in-difference approach, one can control for many factors (country fixed effects, time effects, country specific trends, and time variant country characteristics) that affect the dependent variable in question. Using an index such as that of Ginarte and Park will be quite problematic as it is correlated with other county-specific factors that affect offshoring. Moreover, it will limit the use of time series data since the index is available for every 5 years. Nonetheless, we do run some robustness checks using the most updated version of Ginarte and Park s index provided in Park (2008) (available at: by completing the missing years with averages. The results follow the same pattern as the results with the Branstetter et al. (2006, 2010) dummy but the effects are larger. 20 Following the literature, our working hypothesis is that high-tech industries, as heavy users of intellectual property, would respond more strongly to IPR reform than low-tech industries. Most of the existing empirical work incorporates this hypothesis into their estimations. See among others, Branstetter et al (2010, 2006), Ivus (2010), Maskus and Penubarti (1995), Smith (1999). C. Canals and F. Şener 12

13 Data set We use several sources to construct the final dataset and end up with a total of 23 US manufacturing industries, 6 high-tech and 17 low-tech (see Table 2), that offshore certain tasks over the period (information for 23 out of the 34 years is available) 21 and in 16 different countries (see Table 1) that have experienced a patent reform episode during the time period considered. Table 2 Industries 1. Food and kindred products 2. Tobacco products 3. Textile products 4. Apparel 5. Lumber and wood products 6. Furniture and fixtures 7. Paper and similar products 8. Industrial and other chemicals 9. Plastic and synthetic materials 10. Drugs, cleaning and toilete preparations 11. Rubber and miscellaneous plastics 12. Footwear, leather, and leather products 13. Non-metallic mineral manufacturing 14. Iron and steel manufacturing 15. Non-ferrous metals manufacturing 16. Metal containers 17. Machinery except electrical 18. Office machines and automatic data processing machines 19. Electric machinery, equipment and supplies 20. Motor vehicles and related 21. Scientific and controlling instruments 22. Ophtalmologic and photographic instruments 23. Miscellaneous manufacuring Note: grey area indicates high-tech industries As aforementioned, to construct the offshoring intensity variable we combine IO tables (from BEA) and bilateral imports data (from Center for International Data at UC Davis). Independent variables include: GDP, GDP per capita, trade openness (exports plus imports over GDP), Real Exchange Rate (RER) with respect to the US, a Post Reform Dummy variable and a Technology-Intensive Dummy variable. The first three variables are obtained directly from the World Development Indicators (2009). 22 RER is constructed using the nominal exchange rates from the Penn World Tables 6.3 and inflation from WDI (2009) for most of the countries. 23 However, for China, inflation is taken from the IMF s World Economic Outlook (WEO) from 1980 onwards and assumed to be equal to the average between 1980 and 1985 for the seventies. For Brazil, inflation for the seventies is taken from Fundación Getulio Vargas. Moreover, to construct the RER for Brazil we 21 Final years are restricted to the availability of IO tables: 1973, 1974, 1975, 1976, 1978, 1979, 1980, 1981, 1983, 1984, 1985, 1986, 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004, 2005 and Taiwan information on GDP, GDP per capita, trade and inflation is obtained from National Sources. 23 To compute the RER we take 1970 as the base year. C. Canals and F. Şener 13

14 re-base to 1 the price index for Brazil and the US in 1994, after a new currency (Real) was adopted. 24 The Post Reform Dummy and the Technology-Intensive Dummy variable are extracted from Branstetter et al. (2006, 2010). While their Post Reform variable is based on Maskus (2000), Qian (2007) and Ginarte and Park (1997) work which identifies periods of strengthening of the IPR through patent reforms in numerous countries, their Technology Dummy is consistent with Maskus and Penubarti (1995), among others. Table 3 summarizes the changes in the independent variables over the years at the country level. In particular, it shows the heterogeneity in the country sample used. The sample includes a number of low income per-capita countries such as Indonesia and China in the 1970s, and advanced economies such as Japan and Spain. The differences between the countries are also evident regarding size (GDP). In 2006, Philippines and Colombia are the smallest economies in the sample, and their GDP does not reach 4% of Japan s GDP that same year, the largest economy by far. The sample also includes a number of countries that differ substantially in terms of their openness levels. Table 3 shows the dramatic changes realized by China, Mexico and Turkey. All of them were relatively closed in the 1970s with a trade openness value below 20% of GDP and became very open economies in the 2000 with values above 50%. Table 3 Trade Openess GDP per capita (US$) GDP (Billions US$) (X+M/GDP) Argentina Brazil Chile China Colombia Indonesia Japan Mexico Philippines Portugal S. Korea Spain Taiwan Thailand Turkey Venezuela Such a procedure follows Marc-Andreas Muendler (2003) notes on Brazilian data. C. Canals and F. Şener 14

15 Regression Results We now present our estimates of the equations (4), (5) and (6) for the complete time period, We report heteroscedasticity consistent standard errors, given that we are using a difference-in-difference approach. First, we estimate our main equation using aggregate offshoring as the dependent variable (see results in Table 4). Then we estimate our equations distinguishing between high-tech and low-tech offshoring activities (see results in Table 5). As mentioned earlier, we are interested in the Post Reform Dummy coefficient ( R ), as well as, in the interaction between the Post Reform Dummy and the Technology-Intensive Dummy variable coefficient ( Rct Techi ). We take the log of the dependent variable and most of the independent variables in order to have a semi-elasticity interpretation for the estimated coefficients. In Column 1 of Table 4 we present our results when the aggregate offshoring variable is considered as the dependent variable. The coefficient on the Post Reform Dummy ( R ) is positive but not statistically significant. Once we add the interaction variable, the coefficient on the Post Reform Dummy ( R ) is still statistically insignificant, however, the coefficient on ct the interaction term ( Rct Techi ) appears positive and highly significant (see Column 2 on Table 4). This indicates that high-tech industries are the ones increasing their offshoring intensity when a reform takes place. In particular, the regression results using all years in the sample and all countries imply that, following an improvement in the IPR regime, high-tech industries increase their level of offshoring by 27%. These results are robust to the exclusion of China and Argentina, two countries for which some concerns have been raised regarding the enforcement of new patent laws. 25 Secondly, we run two extra sets of regressions, one using low-tech offshoring as the dependent variable and another using high-tech offshoring as the dependent variable. This distinction, on top of the interaction term that differentiates between high-tech industries from low-tech sectors, provides us with some information about the change in the technology content of offshoring within each type of industry. Table 5 provides some of the results and, again, the same robustness checks have been carried for each of the two specifications. In Columns 1 and 2, we show the regression results for the low-tech offshoring variable. First we only consider the Post Reform Dummy and then add the interaction variable. In both columns, the coefficient on the Post Reform Dummy ( ) is statistically significant and ranges between 15% and 22%. Thus, when patent laws are improved, low-tech offshoring increases on average for the whole industry sample. ct ct Rct 25 Other robustness checks have been performed. In particular, following Branstetter et al (2010) specification we have also excluded GDP; and we added separate country and industry fixed effects instead of a country-industry fixed effect. Results are also robust to only considering the period, as in Branstetter et al (2010). C. Canals and F. Şener 15

16 Nonetheless, the responses of high-tech and low-tech industries are not the same. The coefficient on the interaction term ( Rct Techi ) is negative, close to 30% and statistically significant. Moreover, the restriction that β1+ β2 = 0 cannot be rejected at the standard confidence levels. Thus, we conclude that, after IPR reform, low-tech industries increase their low-tech offshoring while for high-tech industries the change in low-tech offshoring is not significant. The results imply that the rise in the intensity of sample average for low-tech offshoring activities is largely driven by the increase in such activities by low-tech industries. Again, the results are robust to the exclusion of Argentina and China (see Column 5), as well as, exclusion of other countries, later years, GDP and addition of other type of fixed effects. In Columns 3 and 4 of Table 5, we show the regression results for the high-tech offshoring variable. In both columns, the coefficient on the Post Reform Dummy ( R ct ) is not statistically significant, while in Column 4 the coefficient on the interaction between Post Reform Dummy and Technology Intensity ( Rct Techi ) is positive and significant. This suggests that after IPR reforms, US high-tech industries increase their high-tech offshoring by as much as 35%. In this case, the increase in high-tech offshoring activities by high-tech industries is not sufficient enough to raise intensity of high-tech offshoring for the sample average. These results are robust to the exclusion of Argentina and China (see Column 6). Nonetheless, in this case the coefficient on the interaction term is smaller and close to 22%. To sum up the results in Table 5, high-tech industries increase their high-tech offshoring after an IPR reform, while low-tech industries increase their low-tech offshoring. Thus, our results provide clear insights on the change in the technology content of offshoring within each type of industry. In particular, after an IPR reform, industries seem to be increasing their core offshoring intensity, that is, offshoring to industries that are in line with the classification of their own industry. 26 Lastly, Table 6 reports the regression results using our narrowed-down offshoring measure, intra-industry offshoring, which captures offshoring activities that takes place within the same industry. In Column 1 of Table 6, the coefficient on the Post Reform Dummy ( R ) is not statistically significant. In Column 2, when we include the interaction term ( Rct Techi ) the coefficient on ( R ) is negative and becomes almost statistically significant at 10%, whereas ct the coefficient on ( Rct Techi ) is positive and clearly statistically significant. These estimates suggest that low-tech industries decrease their intra-industry offshoring after IPR reforms by around 23% at marginal significance levels, while high-tech industries clearly increase their ct 26 A note on the aggregate offshoring results versus the disaggregated regressions can be useful. In the aggregate offshoring regressions of Table 4, high-tech industries increase their offshoring intensity but such an increase cannot raise the intensity for the complete sample average, the same way it occurred in high-tech offshoring activities regressions of Table 5. C. Canals and F. Şener 16

17 intra-industry offshoring by around 60% (82%-23%). The results do not change in a major way when we drop China and Argentina (See Columns 3 and 4). 27 When we combine the empirical results from Table 5 and 6, we can make the following statements. In response to IPR reform, low-tech industries increase their offshoring to lowtech industries, but reduce their intra-industry offshoring albeit at only marginally statistically significant levels. Thus, the increase in core-offshoring by low-tech industries (i.e., offshoring by low-tech industries to industries that are in the low-tech classification) is driven by lowtech-targeted inter-industry offshoring and not by intra-industry offshoring. Following IPR reform, low-tech industries do not change their offshoring to high tech-industries at statistically significant levels. In the meanwhile, high-tech industries increase their offshoring to high-tech industries and also their intra-industry offshoring. The regressions suggest that the increase in intra-industry offshoring, as expected, play a role in driving the increase in core-offshoring by high-tech industries (i.e. offshoring by high-tech industries to industries that are in the high-tech classification). Following IPR reform, high-tech industries do not change their offshoring to low-tech industries at statistically significant levels. Notice that, even though the time period for our sample is quite long and the countries in our sample are fairly heterogeneous in terms of both their stages of development and industrial structure, some prudence is required when projecting our results to other countries, especially so since most of the countries in our sample had a large amount of offshoring activity emanating from the US before patent reform. 27 When we consider only the period, as in Branstetter et al (2010), we find that the coefficient on R t is clearly not significant, and the coefficient on the interaction term (R ct Tech i) is still significant but smaller (45%). Hence, for the restricted dataset, low-tech industries do not change their intra-industry offshoring after IPR reform while high-tech industries still do increase their intra-industry offshoring but in a smaller percentage (45% versus 60%). Moreover, we note that for certain industrycountry pairs the level of intra-industry offshoring is zero. These values are thus excluded from our log- based regressions since the log of zero is undefined. Consequently, in the regressions shown in columns 1 and 2 of Table 6 there are 118 observations missing. C. Canals and F. Şener 17

18 Table 4 Dependent variable Sample Time period log Offshoring log Offshoring All countries Drop China and Argentina (1) (2) (3) (4) Post Reform Dummy (R) (0.0687) (0.0688) (0.0699) (0.0701) Post Reform Dummy * High -Tech Dummy (R Tech) *** *** (0.0294) (0.0314) log GDP per capita (0.6397) (0.6365) (0.6657) (0.6628) log GDP (0.6366) (0.6365) (0.6654) (0.6626) log Real Exchange Rate *** *** (0.0055) (0.0054) (0.0079) (0.0079) log Trade Openess *** *** *** *** (0.0619) (0.0616) (0.0708) (0.0705) Num. Obs All specifications include country-industry and year fixed-effects and country specific time trends. Heteroskedasticity-consistent standard errors are in parentheses, where ***, ** and * denote significance at the 1%, 5% and 10% levels. Table 5 Dependent variable log Low-Tech Offshoring log Low-Tech Offshoring log High- Tech log High- Tech log Low-Tech log High- Tech Offshoring Offshoring Offshoring Offshoring Sample All countries All countries All countries All countries Drop China Drop China and Argentina and Argentina Time period (1) (2) (3) (4) (5) (6) Post Reform Dummy (R) 0.147** *** *** (0.0692) (0.0692) (0.0894) (0.0895) (0.0705) (0.0904) Post Reform Dummy * High -Tech Dummy (R Tech) *** *** *** *** (0.0295) (0.0382) (0.0316) (0.0405) log GDP per capita ** *** *** *** ** *** (0.6438) (0.6400) (0.8320) (0.8278) (0.6669) (0.8552) log GDP *** *** *** *** *** *** (0.6406) (0.6368) (0.8279) (0.8237) (0.6666) (0.8549) log Real Exchange Rate *** *** ** * (0.0055) ( ) (0.0071) (0.0071) (0.0079) (0.0102) log Trade Openess ** *** ** *** *** *** ( ) (0.0619) (0.0805) (0.0801) (0.0709) (0.0910) Num. Obs All specifications include country-industry and year fixed-effects and country-specific time trends. Heteroskedasticity-consistent standard errors are in parentheses, where ***, ** and * denote significance at the 1%, 5% and 10% levels. Table 6 Dependent variable Sample Time period log Intra-industry Offshoring log Intra-industry Offshoring All countries Drop China and Argentina (1) (2) (3) (4) Post Reform Dummy (R) (0.1458) (0.1452) (0.1469) (0.1465) Post Reform Dummy * High -Tech Dummy (R Tech) *** *** (0.0626) (0.0663) log GDP per capita * ** (1.3690) (1.3550) (1.4130) (1.3995) log GDP * * (1.3628) (1.3489) (1.4131) (1.3996) log Real Exchange Rate *** *** ( ) (0.0118) (0.0175) (0.0173) log Trade Openess *** *** *** *** (0.1341) (0.1327) (0.1530) (0.1515) Num. Obs All specifications include country-industry and year fixed-effects and country specific time trends. Heteroskedasticity-consistent standard errors are in parentheses, where ***, ** and * denote significance at the 1%, 5% and 10% levels. C. Canals and F. Şener 18