The impact of offshoring on technical change: Evidence from Swedish manufacturing firms

This paper examines the link between offshoring and technical change measured by patent and total factor productivity in order to sort out the causality. It applies instrumental variable and matching approaches on a panel of more than 7000 Swedish manufacturing firms over the period 2001– 2014, and identify offshoring-related intermediate imports by the United Nations Broad Economic Categories system. Accounting for self-selection and reverse causality, no impact of offshoring on TFP is found and only weak effect on patenting.


| INTRODUCTION
With the advent of offshoring in the 1990s, a new phase of the contracting-out phenomenon was introduced.Explanations for this development can be sought among a handful of key factors: technological advances, institutional developments favoring trade liberalization, competitive pressures to reduce costs, and the potential for improved productivity (Olsen, 2006).The single most important factor is the digitalization of the economy, which has opened the potential for | 797   BAUM et al.   conducting business activities in entirely new ways, and in an extended spatial area in which a supply chain of local, regional and international firms produces various inputs.
International trade theory is ambiguous about the importance of offshoring.While some of the literature predicts that offshoring of business functions to locations outside of the firm's national borders stimulates innovation and productivity, other authors explore why offshoring may have only a negligible impact on renewal and growth in the focal firm, at least above some threshold level.While offshoring may improve firms' innovation capabilities by replacing laborintensive and routine tasks with cognitive and non-routine ones, offshoring production of intermediates may also reduce the feedback from production to research efforts.
Existing empirical research has not provided clear support for any of these opposing theoretical predictions.Possible explanations for difficulties in achieving consensus in empirical research relate to the difficulties in observing both the extent of the offshoring and innovation performance, as well as methodological challenges to sort out causality, as innovative and high productive firms are more likely to buy more imported inputs (Hummels et al., 2014).
The purpose of this paper is to contribute to the empirical literature assessing the importance of offshoring for corporate innovation and productivity by addressing some shortcomings in previous studies.We take advantage of the United Nations Broad Economic Categories (BEC) to identify offshoring among other imported goods.Our contributions to the literature are as follows: first, we study the universe of manufacturing companies (with at least 10 employees) in an industrialized economy.Second, we follow these companies with an unbalanced panel consisting of unique employer-employee data for a 14-year period.Third, by observing detailed business characteristics for all companies in the economy, we are able to deal with both selection and simultaneity issues, using a control group and appropriate econometric techniques.Fourth, we increase the precision of our estimates by controlling for both routine tasks at the individual level (Frey-Osborne Index) and offshorable activities (Blinder Index).
Examining the impact of offshoring on patenting, trademarks and TFP using a panel of 7000 Swedish manufacturing firms over the period 2001-2014, we find that offshoring may be positively associated with both innovation and productivity.We then estimate an empirical model with offshoring firms defined by a threshold corresponding to 10% of the firms' sales, and a control group consisting of otherwise similar companies.The results show that the link between offshoring on the one hand and innovation and productivity on the other is largely explained by self-selection and reverse causality.We find a positive impact of offshoring on innovation at the lowest acceptable significance level, and no effect on technical change measured as total factor productivity.
The rest of the paper is structured as follows.Section 2 surveys the related literature.Section 3 presents the data and the outcome variables.Section 4 details the empirical strategy, Section 5 reports the results, and Section 6 concludes.

| RELATED LITERATURE
A key feature of the global economy over the last three decades is the rapid growth of offshoring in production and service tasks that were previously produced domestically (Feenstra & Hanson, 2003;Hummels et al., 2014).Intermediate inputs now account for two-thirds of world trade (Acemoglu et al., 2015).
The question how increased input from foreign sources affects firms' innovation and technological change has been studied primarily in the theoretical literature, and to a lesser extent in various strands of the empirical literature.A major focus of this research are the indirect effects in the form of changes in the composition of the labor force: the ratio between skilled and unskilled workers and their relative wages.Offshoring generally involves unbundling and relocating labor-intensive work tasks from the focal firm to foreign firms with lower labor costs, while cognitive and non-routine activities that require specialized skills and technologies remain in house (Baldwin, 2016;Yamashita & Yamauchi, 2019).However, Blinder (2009) and Blinder and Krueger (2013) argue that low-skilled and high-skilled jobs are equally likely to be affected by offshoring.Instead of low skill intensity, the main candidates for offshorability are jobs lacking requirements of physical contact and geographic proximity (Blinder, 2006), as well as jobs associated with codifiable instructions (Leamer & Storper, 2001) and automation (Frey & Osborne, 2017).
Economic research on offshoring has theoretical roots in several different disciplines.They include, among others, the proposition that firms can increase their productivity by focusing on what they do best and outsource the rest (Coase, 1937), the related comparison of the global value-chain process with the Ricardian principle of comparative advantage (Porter, 1985), the concept of an international product cycle proposed by Vernon (1966), the endogenous theories on trade, spillovers and growth by Grossman and Helpman (1991), as well as the literature on shifting production from North (West) to South (East) aimed at raising rate of innovation and productivity in the North (West) (Branstetter & Saggi, 2011;Chung & Yeaple, 2008;Naghavi & Ottaviano, 2009).Also of note are discussions of trade-induced technical change (Bloom et al., 2016), the skill-biased technical change literature on offshoring (Acemoglu et al., 2015), concepts of offshoring driven by fractionalization of production that unbundles supply chains into finer stages across countries (Grossman & Rossi-Hansberg, 2008), and theories on the weakened feedback from offshoring sources to the focal R&D investing firms firm due to imperfect knowledge spillovers (Naghavi & Ottaviano, 2009).
Broadly, the theoretical literature on offshoring predicts two possible outcomes for firms' innovation and productivity when they relocate production overseas.The first is that offshoring can improve firm performance through within-firm resource allocation and efficiency gains.The second is that it can slow the rate of innovation and productivity by limiting the possibility of knowledge creation and transfers between R&D operations and manufacturing due to physical separation.Empirical assessment of these conflicting hypotheses has not reached a consensus on the net effect of offshoring.Below, we illustrate some of the divergent results in recent literature.Yamashita and Yamauchi (2019) study Japanese multinational firms for the period 1995-2011 and find that increased offshore production has little effects on onshore innovation performance as measured with patent statistics.Moreover, the authors report some weak evidence that increased offshore production degrades the quality of innovation, as measured by patent citations.This finding is consistent with the theoretical predictions of the negative effect of the separation of production to offshore locations and domestic innovation activities.Similar results are reported by Branstetter et al. (2017).They study the Taiwanese electronics industry and exploit a fall in the offshoring costs of Taiwanese firms through exogenous policy changes to identify the causal relationship between offshoring and innovation, as measured by patenting.The authors find that firms' propensity of innovation was reduced as a causal effect of greater offshoring of production to China.
In contrast to the main finding in these studies, Bøler et al. (2015) show that imported intermediate goods stimulate R&D among Norwegian firms.Other studies reporting a positive causal impact of offshoring on firms' innovation include Dachs et al. (2015) who study data for more than 3000 manufacturing firms from seven European countries.They present evidence that offshoring firms employ a higher share of R&D and design personnel, introduce new products more frequently to the market, and invest more frequently in advanced process technologies compared to non-offshoring firms.Positive net effect of offshoring is also reported by Fritsch and Görg (2015).They use firm-level data for over 20 emerging market economies to investigates the link between outsourcing and innovation and show that outsourcing is associated with a greater propensity to spend on research and development, to introduce new products and upgrade existing products.
Most closely related to the empirical observations in our study, Tingvall and Karpaty (2011) use data on Swedish multinational firms and find that offshoring to other European countries and North America has a negative effect on R&D intensity at home.However, offshoring to emerging economies is found to have a negligible or even a positive effect on R&D intensity.The latter finding is in line with the theoretical argument on offshoring as a strategy to specialize in knowledge-intensive activities while more routine-based production processes are offshored to exploit relatively lower labor costs.However, the results are not consistent with the assumptions that offshoring to technologically advanced countries may provide access to higher quality inputs, allowing firms to absorb knowledge spillovers on new technologies (Abramovsky & Griffith, 2006).
The empirical literature also provides support to the hypothesis on an inverted U-shape impact of offshoring.Based on a panel dataset of 2421 R&D-active firms in Germany, Steinberg et al. (2017) distinguish between R&D offshoring to foreign affiliates and external foreign parties and find that both offshoring strategies, when pursued intensively, eventually harm firms' innovation performance.Other studies that confirm the existence of an inverted U-shape pattern of offshoring on innovation include Hurtado-Torres et al. (2018).Their paper considers how geographical diversification of firms' R&D offshoring affects innovation performance among 110 MNEs in the energy industry.
In contrast to the still limited firm-level studies on internal innovation and productivity effects of offshore production, the literature has devoted substantial attention to the overall impact of offshore.This research considers consequences on at the spatial, industrial or national level in the offshoring economy.While the significance of international fragmentation of production and relocating manufacturing operations abroad is unclear at company level, there is a more coherent and positive picture at the aggregate level.For instance, Bloom et al. (2016) examine the impact of Chinese import competition on broad measures of technical change-patenting, IT, and TFP-using panel data across twelve European countries from 1996 to 2007 and suggest that the absolute volume of innovation increases within the firms most affected by Chinese imports in their output markets.Castellani and Pieri (2013) show that productivity growth of 262 regions in Europe is associated with offshoring of R&D activities by domestic multinational enterprises based in the same regions.They find a large and positive correlation between the extent of R&D offshoring and the home region productivity growth.
In summary, the empirical literature on offshoring and innovation suggests that no definite conclusions can be drawn about positive or negative causal effects of offshoring on innovation and productivity.Many studies suggest that the disadvantages outweigh the advantages, and among studies with positive results there are indications that offshoring is only an effective strategy up to a certain threshold level.
One explanation for the heterogeneous results in existing studies of the relationship between offshoring, innovation and technological development is that they capture actual differences in outcomes between products, companies, industries, and destinations, as well as the importance of the scope of the outsourced activities.
It may also be the case that the results across studies are not comparable due to differences in the quality of data, measurement of offshoring and measurement of innovation.Another key issue is how the studies have been able to correct for endogeneity.There is extensive evidence in the literature that more innovative firms are those that aggressively engage in offshoring in production (Yamashita & Yamauchi, 2019).The decisions of engaging in offshore production and innovation are therefore endogenous to individual firms.Researchers have addressed this challenge with various empirical approaches such as instrumental variables estimation.Recently, a small number of studies have exploited the occurrence of exogenous shocks (Bloom et al., 2016;Autor et al., 2016;Branstetter et al., 2017) and Bøler et al. (2015).Propensity score matching is another approach that has been used to analyze the causal effect of offshoring on innovation.For instance, Dachs et al. (2015) use a propensity score matching estimator to identify a control group of non-offshoring firms with characteristics similar to those of offshoring firms.

| DATA
The data come from several sources.The combined employer-employee dataset is obtained from Statistics Sweden, and covers the population of Swedish manufacturing firms (2-digit NACE Rev.2 codes 10-37) and their employees for 2001-2014.Similar to most other studies using Swedish trade data, we only consider firms with 10 or more employees, as the information provided for smaller firms is likely to be less reliable.
The employer dataset contains information on sales, value added, exports, imports, capital stock, corporate ownership structure and number of employees at the firm level.Continuous variables are deflated using deflators for exports, imports and producer prices provided by Statistics Sweden.Firm-level data are matched with patent data retrieved from the European Patent Office (EPO).By merging this data with the employee dataset, we can access information on employees' level of education, occupation and income levels.Beginning with Feenstra and Hanson (1999), researchers have defined offshoring as imports of intermediate inputs.More recent research has advanced the identification by measure offshoring as imports of the same four-digit industries (Hummels et al., 2014), or same six-digit industries that importers produce domestically (Bernard et al., 2020).
In this paper, we apply a different approach to identify reallocated production of inputs.As offshoring production leads to firms' imports, it is possible take advantage of the United Nations Broad Economic Categories (BEC), which is a three-digit classification system grouping transportable goods according to their main end use: capital goods, consumer goods and intermediate goods.The latter has been applied as a proxy for offshoring.A main challenge for offshoring research based on the BEC system is that revisions imply that unique products might be classified differently over time.To account for the reclassification, we apply the algorithm suggested by Pierce and Schott (2012) and further developed by Van Beveren et al. (2012) for concording trade and production data over time, and consider an imported product as offshored if it is classified as an intermediate good.
To mitigate possible bias due to spurious correlation, we control for the potential trends that may make jobs more likely to be offshored, 1 using the Blinder index of offshorability. 2Applying the classification method proposed by Blinder and Krueger (2013), we first consider 430 job titles in the Swedish labor market and estimate their offshorability.Each occupation is then classified according to whether it has a high risk of being moved abroad.We then calculate a firm-specific offshorability measure, defined as the ratio of offshorable jobs to total employment.
We also include the Osborne-Frey index (Frey & Osborne, 2017) in our analyses.This index is designed to capture the likelihood for each occupation to be replaced by computers or robots in the near future.The computed Osborne-Frey index is also firm-specific.
The main challenge in estimating total factor productivity (TFP) is that due to positive productivity shocks, firms tend to respond by expanding their level of output and by demanding more inputs, and vice versa for a negative shock.The positive correlation between the observable input levels and the unobservable productivity shocks is a source of bias in TFP.
Recent years have seen a number of methodological developments of TFP computation addressing this bias: Olley and Pakes (1996), Levinsohn and Petrin (2003), Ackerberg, Caves, andFrazer (2006, 2015), and Manjón and Mañez (2016) contributed to the literature proposing twostep estimation procedures, while Wooldridge (2009) showed how to perform consistent estimation within a one-step GMM framework.Most recently Mollisi and Rovigatti (2017) proposed a new estimator, based on the Wooldridge approach, using dynamic panel instruments as used in the Blundell and Bond (1998) methodology.In this paper, we apply the Wooldridge TFP approach.
To control for heterogeneous levels of ability, we estimate residuals from Mincer equation, defined over traditional individual-level variables such as age, age squared, education and gender.We take this measure as our proxy for ability and calculate the average ability of the firm's workforce.While the Mincer earnings function explains wage income as a linear function of schooling and experience, this approach allows for non-linearities in the relationship between earnings and schooling (Bhuller et al., 2017).
A growing number of studies shows the importance of corporate ownership structures on productivity and managerial practices.There are not only potential differences between foreign and domestic multinational firms, but also among the various categories of domestic firms.Our study separates firms in four ownership categories: domestic non-affiliated firms, domestic affiliated firm (UNE), and domestic and foreign multinational firms (MNEs).Other controls included in our regressions are measures of firm size, industry-specific effects for 18 two-digit industries and time-specific effects.
Table 1 lists all variables used in the analyses and provides detailed definitions for each of them.

| Descriptive statistics
As shown in Table 2, the average annual number of firms observed is about 7500, which yields a total of 73,722 firm-year observations.There is substantial attrition, approximately 20%, in the sample, from 8219 firms in 2001 to 6569 firms in 2014.Most firms in our sample are domestic non-affiliated or independent companies (78%) located outside metropolitan or large cities (45%), have fewer than 50 employees (75%) and are categorized as low or medium-low technology companies (56%).Only 18% of the firms are multinationals, have fifty or more employees and are located in metropolitan areas.More interestingly, only about 5% are high-technology firms.See Table 3 for more detailed descriptive statistics.
About two-thirds of manufacturing firms in our sample carry out offshoring (see panel Figure 1).Most firms offshore to OECD countries, (excluding the G7), other Nordic countries and the BRICS at the beginning of the sample.There is, however, a substantial shift towards "East Europe", defined as all countries in the former Soviet union except Russia, and other lessdeveloped countries which jointly account for 50% of all offshoring in 2014, in comparison to the 20% recorded at the beginning of the sample period.When we consider the relative size of offshoring rather than the fraction of offshoring firms, Table 2 shows similarities between domestic and foreign multinationals, as well as between that the OECD countries have a dominant role.
Figure 2 provides a snapshot of offshoring patterns and intensity across regions for the four different ownership categories.The relative importance of regions appears to be heterogeneous overall, but homogeneous across groups (MNEs vs. non-MNEs).Most noticeable is the relative growth of offshoring to Eastern Europe for all firms, although other less-developed countries have also benefited, as captured by "rest of the world".Offshoring intensity varies from 93% in foreign MNEs to 42% in domestic non-affiliated companies.
Related to the potential impact of offshoring on labor market outcomes such as income inequality, visual inspection of Figure 3 suggests that as offshoring has increased, there has an increase in the Swedish skill premium.Another relevant observation is that about 50% of all jobs in Swedish manufacturing are potentially offshorable and only a small fraction of firmsapproximately 5%-innovate.However, it should be noted that we use both patent and TFP as indictors of technical change and the latter is calculated for all firms.

| EMPIRICAL STRATEGY
In order to estimate how the offshoring destination affects the likelihood of innovation, proxied by making a patent application, we specify the following model: where µ i is a firm-specific error term and λ t is a year effect.This model is estimated as a random effects probit model.Next, we estimate the impact of offshoring on the firm's productivity, expressed as log TFP, in a dynamic specification.This model is specified as: To estimate this dynamic panel model specification, we employ the first-difference GMM estimator developed by Arellano and Bond (1991).This framework is convenient because it is (1) Pr(patent it = 1) = f (offshoring destination it , potential offshorability it , workers � ability it , automation potential it , controls it , i , t ) (2) relatively easy to allow for endogeneity of offshoring, which is instrumented with both its own lagged level values and external instruments together with other covariates.

| Propensity score matching
In this empirical analysis we focus on estimating the causal effect from offshoring on those firms that persistently offshore.As we cannot observe the counterfactual for those firms that offshore, which is what would have been their outcomes in case they did not choose to offshore, we establish a quasi-experimental research design by defining a control group of non-offshoring firms which are most similar to the offshoring firms.To identify those firms we use propensity score matching PSM (Rosenbaum & Rubin, 1984, Rubin 1997). 3The first step in PSM is to estimate the likelihood that a firm is a persistent offshoring firm, for which we use a probit model.This model is estimated based on variables for year 2001, while persistent offshoring is determined for the entire sample period. 4

| EMPIRICAL RESULTS
In this section, we present summary statistics and estimates for the models specified in Equations ( 1) and ( 2).We employ different estimation techniques in order to gauge the importance of offshoring on different aspects of firms' innovation strategies.Table 3 reports summary statistics for key variables in the analysis.Five percent of the firms are protection their innovations with formal intellectual property rights (patent). 5Total factor  productivity is expressed in logarithms.We define employees with three years of university education as skilled employees, and those with a lower level of education as unskilled.On average, the firms have 7% of employees are classified as skilled, and the mean skill premium is 0.75.
Our main offshoring measure is reported for all six destinations in the study.The most prevalent destinations are the G7 countries and other OECD countries when offshoring is normalized by sales.We estimate a Mincer residual for each employee, assuming that it can be used as a proxy for ability as a complement to human capital.
Approximately 50% of jobs are potentially offshorable as expressed by the Blinder index, and the Osborne-Frey index suggests that 60% of the jobs in Swedish manufacturing can potentially be replaced by machines or robots.We denote this measure as Automation potential, assuming that a high value of this index reflects unexploited efficiency potential.

| Innovation and technical change
Table 4 reports average marginal effects of the propensity to apply for patents.A priori, we assume that offshoring allows firms to switch resources from production to research.The working hypothesis is that this should be manifested through increased innovation capabilities.The hypothesis is confirmed for offshoring to low wage destinations, and partly for offshoring to other OECD countries (MTL and LT).
Our second analysis considers total factor productivity as a measure of technical change.Some of the prior empirical literature reports a positive impact of offshoring on TFP.However, what distinguishes our analysis from most of the existing literature is that we observe mainly small firms over a long period.
Table 5 presents results from four different dynamic models: pooled OLS, fixed effects, difference GMM and system GMM.The two latter are estimated by the Arellano-Bond approach.In this analysis, we measure offshoring by the logarithm of its nominal value.
The approaches presented in columns ( 1) and ( 2) show a positive and highly significant association between offshoring and TFP.However, both the pooled OLS and fixed effects estimates are potentially biased in a dynamic setting.Columns ( 3) and ( 4) presents results from the F I G U R E 3 Offshoring and the skill premium in Sweden.The left panel depicts the fraction of firms in Swedish manufacturing that offshore production.The right panel plots the skill premium, defined as the wage ratio between university-educated and non-university educated workers, for the 75th percentile.Both of these figures are based on firms with at least 10 employees  Arellano-Bond instrumental variable estimator for the dynamic panel setting which allows for a causal interpretation of the estimates.Both columns show positive and highly significant coefficients on the offshoring variable.The size of the coefficient estimate is 0.011 in the difference GMM model and 0.022 in the system GMM model.The test statistics in the foot of the table show that the instruments are valid in both Arellano-Bond estimators and that there is no second-order serial correlation in the differenced error terms.
The overall results in the first step of the analysis provide a positive link between global value chains, as reflected by an increased reliance on offshoring, and innovation and technical change.
It should be noted that we may only interpret this relationship in terms of causality with regard to the effect on total factor productivity.Further, as we do not use any external instruments in the Arellano-Bond model, the results should be interpreted with some caution.
In the next step of the analysis, we test the sensitivity of the parameter estimates above in an approach that accounts for self-selectivity and reverse causality between innovation and productivity (Table 6).

| Robustness test
It is plausible that firms that are more productive and have higher innovation capabilities are more likely to engage in offshoring activities.In this case high productivity and high innovation capability jointly determine the likelihood and intensity of offshoring.In fact, the results from the PSM shown in Table 7 imply that persistent offshoring firms, in our case about 1000 of the 7000 firms in the sample, have more patents and significantly higher productivity compared to non-offshoring firms.However, it is an open question whether this is a result of offshoring or itself a determinant of the likelihood to engage in offshoring.
With the help of PSM we can define a control group of non-offshoring firms which are most similar to the offshoring firms in their productivity and innovation outcomes in 2001.The results Note: Cluster-robust standard errors in brackets.Estimation method is reported underneath the column number.Dependent variable is total factor productivity (TFP), calculated à la Wooldridge (2009).The difference between columns (3) and ( 4) is that offshoring is treated as endogenous in the latter.7 indicate that there are no remaining significant differences between the treatment group of offshoring firms and those in the control group after matching.
We then study the outcome variables patent and TFP for 2002-2014 for these two groups of firms.Table 8 presents estimates for the matched sample.Probit estimates for the innovation model are reported in columns 1 and 2. While we found highly significant and positive point estimates for the category all firms in Table 4, the pooled probit estimate results in column 1 is still positive but is no longer significant.Column 2 considers a panel probit model using the random effect estimator.The effect of offshoring on innovation is positive, but only at the 90% level of significance.Thus, after accounting for self-selectivity, the treatment effects from offshoring become much weaker in the preferred random effects model.
Columns 3 and 4 reveal the causal impact of offshoring on TFP using the matched sample of offshoring and non-offshoring firms.Not accounting for self-selection and not properly controlling for endogeneity, the dynamic Arellano-Bond estimates in Table 5 supported the literature that suggests that firms that replace production of intermediate inputs with insourcing from foreign destinations increases their productivity.Although this effect remains in the pooled OLS model reported in column 3, the effect disappears completely when we include firm fixed effects in the TFP model presented in column 4.
Taken together, the matching results imply that offshoring is clearly endogenous, with selection as an important factor that should be addressed in the empirical approach.Much of the previous research has neglected this issue.

| CONCLUSION
The rapid growth of offshoring in production and service tasks that were previously produced domestically is a key feature of the global economy over the last three decades.However, understanding how offshoring affects firms' innovation and technical change remains an important question in economics.Recent theoretical development and empirical studies yield ambiguous conclusions.While one group of studies provides evidence supporting important efficiency and specialization gains from this internal resource allocation, other authors question the positive effects of offshoring and argue that separating the production and development functions of a firm can undermine its innovation capacity and hinder its productivity growth.
( This paper contributes to the literature by performing an empirical investigation that addresses some of the shortcomings in previous studies.We study the universe of manufacturing companies in an industrialized economy.The data allow us to observe the firms and their employees over a 14-year period.By observing detailed business characteristics for all companies in the economy, we are able deal with both selection and simultaneity issues, using a control group and appropriate econometric techniques.
There are two main results from our analysis of Swedish manufacturing firms.First, without accounting for self-selection into offshoring and properly controlling for endogeneity, the estimates suggest that innovation and productivity are increasing functions of offshoring.Second, applying a matching approach with a control group of similar firms, our estimates show that the positive link between offshoring, innovation and total factor productivity largely is explained by self-selection and reverse causality.The positive impact of offshoring on innovation remains, but significantly different from zero only at the 10% level.With control for the two sources of potential bias, we find no effect from offshored intermediate production on firms' total factor productivity as an indicator of technical change.
The results show that the link between offshoring on the one hand and innovation and productivity on the other is largely explained by self-selection and reverse causality.We find a positive impact of offshoring on innovation at the lowest acceptable significance level, and no effect on technical change measured as total factor productivity.Innovative and productive firms are more likely than other firms to be engaged in offshoring activities, but they are not more innovative or productive due to their import of intermediate goods.Our results are consistent with the trade literature that finds that exporters are on average more productive than other firms, and that self-selected exporters are more innovative.A growing number of studies fail to find strong evidence for positive effects of exporting on firm performance. 6In line with these findings, we do not find evidence of a strong causal relationship between offshoring and firms' innovation and productivity once the endogeneity of the offshoring decision is considered in the empirical approach.Areas for future studies may include assessing the relevance of this conclusion for companies in different size classes, different industries, different offshoring destinations and different ownership connections to foreign suppliers of intermediate inputs.
Offshoring patterns by destination and ownership [Colour figure can be viewed at wileyonlinelibrary.com] Cluster-robust standard errors in brackets.Estimation is for panel-data, random-effects probit models.Dependent variable is a dichotomous variable for patenting.Non-offshoring firms, firms with 10-29 employees, foreign MNEs, firms located in Metropolitan areas and High-tech firms (HT) constitute the reference groups.The measure for potential offshorability is the firmspecific computed Blinder index.The measure for ability is the firm-specific, fully-saturated Mincer residual.The measure for automation potential is the firm-specific Frey-Osborne index.Measures of offshoring are winsorized to exclude the 1% extreme values of the upper tail of the distribution.Regressions include ownership, firm and time fixed effects.*p< .10;**p < .05;***p < .01.
−1 , log offshoring it , potential offshorability it , automation potential it , workers � ability it , controls, i , t ) Descriptive statistics for Swedish manufacturingfirms, 2001-2014 T A B L E 2Note: MNE stands for multinational enterprise and UNE domestic affiliated firm.Metropolitan areas are Stockholm, Gothenburg and Mälmo, and large cities are those with more than 100,000 residents.Technology groups were defined according to the OECD classification by taking information on R&D and human capital intensity.
Offshoring prevalence and destination.BRICS are Brazil, Russia, India, China and South Africa.G-7 includes Canada, France, Germany, Italy, Japan, the United Kingdom, and the United States.Nordic Countries are Denmark, Finland, Iceland and Norway.Other OECD countries are all OECD countries except those in G-7.East considers all former members of the Soviet Union with the exception of Russia [Colour figure can be viewed at wileyonlinelibrary.com] Note: Innovation is an indicator of patent application activity.Human capital is defined as the share of university-educated workers in total employment.The skill premium is the ratio of wages of university-educated to non-university educated workers.Offshoring to destination r is proxied by the value of imported intermediate goods relative to sales.Workers' ability is the fully-saturated Mincer residual.Potential offshorability and automation potential are the computed Blinder and Frey-Osborne indexes, respectively.F I G U R E 1 Firms with 10-29 employees, foreign MNEs, high-tech firms and firms located in Metropolitan areas are the reference groups.Important to note is that we estimate with an absolute measure of offshoring (this avoids having productivity shocks artificially caused by sales movements).The measure for potential offshorability is the firm-specific computed Blinder index.The measure for ability is the firm-specific, fully-saturated Mincer residual.The measure for potential offshorability is the firm-specific computed Blinder index.The measure for automation potential is the firmspecific Frey-Osborne index.All regressions include firm-and time-specific effects, with the sole exception of OLS which does not include firm-specific effects.Standard errors in brackets.Offshoring refers to firms that have at least 10% offshoring relative to sales in at least 80% of observation years 2001-2014.Tests of balancing assumption after propensity score matching, year 2001 Note:*p < .10;**p< .05;***p< .01.T A B L E 7 Robust standard errors in brackets.Abbreviations: DiD, difference-in-difference estimator; FE, fixed effects; RE, random effects.*p < .10;**p < .05;***p < .01. Notes: