1
Research Master Thesis
Technology, Offshoring and The Functional Structure of
Labor Demand in Advanced Economies
Aobo Jiang
Supervisor: dr. G.J. (Gaaitzen) de Vries Co-assessor: prof. dr. M.P. (Marcel) Timmer
2
Technology, Offshoring and The Functional Structure of Labor
Demand in Advanced Economies
Abstract
This paper examines the link between international outsourcing, investments in informational and communication technologies (ICT) and changes in the functional structure of labor demand in the U.S., Japan, and seventeen European countries for the period from 1995 to 2008. We apply a business function framework by classifying workers into three groups, namely those who conduct R&D, production and other activities. We include 19 advanced countries and distinguish 35 industries in our analysis. We apply translog cost function framework, and estimate a system of variable demands for workers classified by business functions. Our result show that offshoring and technological progress benefit workers who conduct R&D activities, but harm workers who do production activities, and this result is stronger for workers in services than in manufacturing industries. Our results suggest that these changes in functional labor demand are tempered due to the complementary nature of R&D and production activities.
3
Contents
1. Introduction 2. Literature review
2.1 Technology and offshoring
2.2 Relationship among business functions
3. Econometric methodology 4. Data 4.1 Data Construction 4.2 Descriptive Statistics 5. Results 5.1 Key Results
5.1.1 Technological change and international offshoring 5.1.2 ICT and non-ICT related capital
5.1.3 Relationships between groups of activities 5.2 Robustness and Extensions
5.2.1 Manufacturing industries only
5.2.2 Different groupings of business functions
4
1. Introduction
The income shares of production factors have changed over time. Recent empirical evidence clearly shows that among all kinds of labor inputs, there has been a general job polarization trend in many developed countries (Michaels et al, 2013; Goos, Manning and Salomons, 2014). In other words, during the past few decades, the demand for medium skilled workers has decreased a lot, while the demand for high skilled workers, to the opposite, has increased by a considerable amount. However, the demand for low skilled workers has increased or remained unchanged (Acemoglu and Autor, 2010). Researchers have suggested several possible ways to explain the job polarization phenomenon, among which offshoring and technological progress are the most debated driving forces and have attracted the most attention from researchers. According to Autor, Levy and Murnane (2003), modern technology has greatly affected labor demand globally, especially in advanced countries, so that demand is shifting in favor of higher educated workers where technology complements their activities but against those workers with medium-skills doing routine activities like administration where technology substitutes their activities. In terms of tasks, some tasks that can be easily codified and programed by computers have been taken over by machines, and other tasks that need information as input would benefit from the technological complementary effect, and this is known as the ‘routine-biased technological change’. Furthermore, owing to the increasing opportunities of international fragmentation, there are more and more offshoring activities to emerging countries undertaken by companies located in advanced countries. By offshoring some production activities to other countries and locations, companies in advanced countries benefit from lower labor cost in emerging countries. Apparently, tasks and production activities that have been offshored are mainly less skill intensive compared to those tasks and production activities that have remained domestically. As a result, part of the decline in lower skill income shares domestically can be explained by offshoring.
One of the most important novelties of our research is that, different from the existing research, which focuses their analysis on industry, tasks and especially the skill content of labor, we will analyze labor cost shares classified by business functions. Business functions mainly classify the tasks that conducted by a company. The tasks belong to a specific business function group and form a set of activities that fulfill a certain kind of service or produce certain goods. Business functions are similar in definition to occupations, however, the emphasis of business function is on certain business activities, rather than the activities that are carried out by individual workers (Brown, Sturgeon and Lane, 2014).
5
each business function typically includes several occupations. Sturgeon et al. (2012) also provide arguments to support the use of a business function framework. They argue that the global value chain nowadays is different from that in the past, in a way such that production tends to be more and more bundled by services. Value added is not fully captured by just looking at the production stage and specifically the inputs that being used to produce the final goods. On the other hand, offshoring activities are operating beyond the level of main products to the level of vertical business functions. In other words, companies nowadays not only offshore intermediate or the assembly of goods, but also increasingly the intangible and support business functions like R&D, sales, marketing, etc. Besides, a considerable amount of existing research that examines the effect of technological progress and offshoring on cost shares of labor classifies labor according to their skill level. However, According to Maurin and Thesmar (2004), every activity involves both high-skilled and low-skilled workers, thus technology and offshoring do not affect labor by different skill levels, but instead they affect labor skill distributions within a certain activity. We thus find that classifying labor according to business functions that they conduct is a more direct way to study the effect of technological progress and offshoring than classifying them according to skill levels. Furthermore, the business function framework can provide us with more details and insights beyond the findings we already got from the existing research. More discussions will follow in the next section. Taken all the above mentioned reasoning into consideration, we believe that the business function framework can be thought as a more reasonable and efficient framework to study labor demand change than other frameworks that focus on industry, skill or task level.
In this paper, we classify business functions into three general groups, namely R&D, production and other activities. This is a new way of classifying labor and it is very convenient and flexible for different research aims. In our study, we are mainly interested in the relationship between R&D and production activities, so that we classify labor into the above mentioned three groups. For other research interests, business functions can be classified differently by aggregation or dis-aggregation.
We study labor demand change across 19 advanced countries and 35 industries for the period from 1995 to 2008. Both manufacturing and service industries are included in our analysis. Even though some existing research mainly focus their attention on manufacturing industries when study the effect of offshoring, and the reason is mainly because that offshoring was always believed to influence only, or at least mainly manufacturing and production industries (Brown, 2008). However, according to Brown, Sturgeon and Lane (2014), since early 2000s, offshoring gradually extended to service related industries as well. Especially job losses in information and communication related activities have arisen as another very important composition of ongoing effect of offshoring besides production industries.
6
flexible translog cost function framework makes it possible to estimate substitution and complementary elasticities among business functions. It also provides a fairly precise estimate of the relationship between demand for workers classified by business functions and other variables like economy wide technology trend, ICT/ non-ICT capital compensation and offshoring. Instead of estimating single equations of relative demand as in Michaels et al. (2013), we simultaneously estimate a system of variable functional labor demands using panel data techniques as in Hijzen et al. (2005), which generates more efficient results than single equation estimations whenever disturbances are correlated across equations.
The main results of this study show that, first, both economy wide technological progress and investment in ICT positively correlate with the demand for workers who are involved in R&D activities, but negatively correlate with the demand for workers who are involved in production activities. This finding can probably be interpreted as economy wide technological progress in general, ICT related technology in particular benefit R&D activities, but harm production activities; second, offshoring positively correlates with labor cost share of R&D activities, but negatively correlates with labor cost share of production activities. This is in line with many existing research (see Michaels et al. 2013; Hijzen et al. 2005; etc.). It can be interpreted as offshoring benefits R&D activities, but harms production activities. These findings are generally in line with the argument of the ‘routinization hypothesis’.
As we state in the previous paragraph, translog cost function framework enables us to estimate the substitution and complementary elasticities among labor cost shares of different business functions. Results show that R&D and production activities are complementary to each other. This basically suggests that R&D and production activities tend to be conducted in the same location thus to benefit from the complementarity effect. Considering the results regarding the effect of technological progress and offshoring on labor demand change classified by business functions we mentioned before, we propose the following possibility: as technology develops and the increase in offshoring activities these years, advanced countries tend to increasingly demand more for R&D related activities, but increasingly offshore more production activities abroad. Thus, the share of payments to R&D activities will increase, but the share of payments to production activities will decrease. However, owing to the complementary effect between R&D and production activities, R&D and production activities tend to be located in the same place. As a result, the complementary effect also serves as a moderating effect, such that production activities will not be offshored as much as it supposed to be without the complementary effect.
7
2. Literature review
In our study, we mainly build on two strands of literature. Our main focus is on the effect of technological change and offshoring on the functional structure of labor demand, and we will also pay attention to the existing theory and evidence on the relationship among business functions, especially between R&D and production activities.
2.1 Technology and offshoring
Technology and offshoring have long been proposed as the two main driving forces of labor demand change in advanced countries. The existing research, however, mainly classifies labor groups according to their skill levels, which is mainly approximated by educational attainment level (Michaels et al, 2013). However, according to Maurin and Thesmar (2004), the impact of technological progress on labor demand is indirectly through the channel of skills, but directly through the channel of the activities that labor do, namely business functions. Specifically, they claim that the key effect of modern technology is not that they are biased towards an increased demand for high-skilled labor compared to low-skilled labor, but that they greatly change the distribution of workers over activities. According to Maurin and Thesmar (2004), the effect of technological progress on labor demand is mainly determined by how easy it is to program certain activities, and every activity involves both high-skilled and low-skilled workers. They find that, for example, firms have declined the labor demand for both high-skilled and low-skilled workers who are involved in production activities, but expanded labor demand for the conception and marketing of products. We thus find that grouping labor according to the activities they conduct is a more direct way than according to their skill levels. Furthermore, even though on average, we can still believe that technology affects demand for different skilled level of workers, a new classification of labor according to business functions they are involved in would certainly bring us more insights and details on this research topic.
8
This is possible a reason why Michaels et al. (2013) does not find a significant effect of trade on labor demand change.
On the other hand, however, Autor, Dorn and Hanson (2013) do a study for the US, and they mainly try to investigate the influence of Chinese imports on the US labor demand change. They find that in the period 1990-2007, import competition from China nearly explains 25% of the decline in US domestic manufacturing employment. This indicates a strong impact of international trade on demand for labor who perform certain tasks like production work. Similarly, Hijzen et al. (2005), also find a strong effect of offshoring on the skill structure of labor demand in the UK. They use a narrow definition of international outsourcing as a proxy of trade, which only takes into account the imported intermediates in a given industry from the same industry. They claim that narrow outsourcing is preferred to broad outsourcing, which considers inputs imported from all industries, since narrow outsourcing is more relevant to the nature of fragmentation which certainly occurs within industry. Paying attention to only manufacturing industries for the period from 1982 to 1996, they find that international outsourcing has a significantly negative effect on low-skilled labor. Technological change in terms of R&D activities, on the other hand, is found to benefit the demand for high-skilled labor, but the effect on the demand for medium and low-skilled labor is insignificant. As a result, Hijzen et al. (2005) find evidence to support only international outsourcing as the driver of labor demand change in the UK, rather than technology. The use of R&D intensity as the indicator of technological change in the industry might be a reason why they do not find a significant correlation between technological progress and labor demand change since R&D intensity is not necessarily the best indicator of technological progress. The ‘routinization hypothesis’ raised by Autor, Levy and Murnane (2003) states that it is ICT that substitutes for routine tasks but complements non-routine analytical tasks. Furthermore, Michaels et al. (2013) use ICT capital intensity as the indicator of technological progress and they find a significant effect of ICT capital intensity on labor demand change, which suggests that technological progress is a driving force of labor demand change.
Furthermore, Feenstra and Hanson (1999) also find significant effects of both technology and trade on labor demand change in the US. In their study, trade is approximated by the imported intermediate goods that are caused by outsourcing, and technological progress is approximated as the expenditures on high technology related capital, e.g. computers. Different specifications are applied to quantitatively trace the explanation power of the effect of trade and technology on labor demand change. Even though the relative explanation power of trade and technology varies with different empirical settings, all the results are in line with the idea that trade and technological progress increase the wage share of nonproduction labor, and decrease the wage share of production labor.
9
The above mentioned literature about the effect of trade on labor demand change focus their analysis on industry level. Other research also provide firm level analysis. Muendler and Becker (2010) examine whether multinational enterprises (MNEs) react to wage differentials by substituting domestic jobs for foreign employment using German MNEs data. They find that multinational labor demand responds to cross location wage differentials both at the extensive margin (when an MNE extends to foreign locations) and the intensive margin (reallocate jobs to existing subsidiaries). This finding basically provides evidence that trade in intermediate goods from abroad would affect labor demand domestically. Disdier and Mayer (2004) and Becker et al. (2005) find similar results for French and Swedish MNEs respectively.
To sum up, existing research provides evidence in support of both trade and technological progress as potential drivers of labor demand change in advanced countries. Considering the limitations of the existing research as we mentioned above, in this study, we will apply the best indicators for both trade and technological progress, namely narrow offshoring and ICT capital intensity, respectively. Furthermore, most of the research classifies labor according to their educational attainment level. However, in order to provide more details and study the most direct effect of technological progress and offshoring on labor demand change, we will contribute to the existing research by studying a business function framework, which focuses on the specific activities that workers perform.
2.2 Relationship among business functions
As the existing research we have mentioned suggests , trade and technological progress could have an impact on demand for workers classified by business functions, so that demand for workers who are involved in different business functions are affected. However, relationship among business functions in terms of whether they are substitutable or complementaryto each other also has important effect on domestic shifts in the functional structure of labor demand, which is too relevant to be ignored (Defever, 2006). A complementary effect between two business functions will undoubtedly lead to a colocation decision of the two business functions by firms.
10
by the relationship among business functions. We will provide more discussions on this in the analysis of the results we get.
Recently, Devever (2012) finds that there are complementary effects between R&D and production activities, which leads to a decision of colocation of R&D and production activities by firms. Furthermore, according to He and Xiao (2011), production activities benefit from specialized labor and technology and knowledge spillovers, which are abundant in R&D activities. Thus, firms would benefit from collocating production and R&D activities. Similarly, Enright (2009) states that both R&D and production are scale sensitive and knowledge seeking, and thus they tend to seek a bigger market with abundant knowledge spillovers. As a result, R&D activities apparently benefit from agglomeration effects, and at the same time, production activities benefit from locating together with R&D activities. The above mentioned literature all elaborate location choice of business activities. However, limitations still remain. Some of the existing research focus their analysis on different skilled type of labor instead of the activities that have been done by those labor, and this is an indirect approach to investigate the relationship among business functions. Other research are mainly on firm level, which is difficult to reach a generalized conclusion on the relationship among business functions on industry level. We will thus do a macro study on cross country and industry level, and we will directly investigate the relationship among different business functions, especially between R&D and production activities.
3. Econometric methodology
In order to investigate the effect of offshoring and technological progress on the functional structure of labor demand, we use a standard translog cost function framework to conduct the econometric analysis. The econometric model is mainly based on Christensen et al (1973). The translog function represents a group of flexible functional forms for the cost functions. The flexible translog production function framework provides a broad range of diversified elasticities of substitution. One of the most attractive advantages of the translog production function is that, different from the Cobb-Douglas production function, the translog function does not assume perfect substitution between business functions or perfect competition on the production factors’ market. Much research that study the nature of technological progress and the effect of offshoring on skill-demand have used translog function framework, such as Hijzen et al. (2005), Foster-McGregor et al. (2013) and Baltagi and Rich (2005).
11
provide us with more detailed information of the structural change on industry level labor demand. The variable factors are labor demand for three groups of business functions, namely R&D, production and other activities. To keep it in line with Berman et al. (1994),we assume capital and output to be quasi-fixed, which means that both capital and output can be treated as exogenous in the short run. In addition, according to Timmer and de Vries (2015), owing to data limitations, it is still impossible to allocate capital income to specific business functions. Capital is thus better treated as exogenous accordingly.
To further distinguish the possible different effects of ICT-capital and non-ICT capital compensation on labor cost shares classified by business functions, we separate capital compensation into ICT related and non-ICT related capital compensation. To investigate the effect of offshoring on labor cost shares of different business functions, we include an offshoring indicator. To keep it in line with Hijzen et al. (2005), we apply a narrow definition of offshoring, which only takes imported intermediates in a given industry from itself into account. According to Hijzen et al. (2005), narrow offshoring is preferred to broad offshoring because it is more closely connected to the nature of fragmentation, which mainly takes place within industries.
According to Christensen et al. (1973), a translog function is twice differentiable, linear homogenous and concave in factor prices. For a specific product, the translog function is given by: 𝑙𝑛𝐶(𝑝𝑡, 𝑦𝑡, 𝑡) = 𝛼 + ∑𝑖𝜖𝐹𝛽𝑖𝑙𝑛𝑝𝑖𝑡+12∑𝑗𝜖𝐹∑𝑖𝜖𝐹𝛾𝑖𝑗𝑙𝑛𝑝𝑖𝑡𝑙𝑛𝑝𝑗𝑡+ 𝛽𝑌𝑙𝑛𝑦𝑡+12∑𝑖𝜖𝐹𝛾𝑖𝑌𝑙𝑛𝑝𝑖𝑡𝑙𝑛𝑦𝑡+ 1 2𝛾𝑌𝑌(𝑙𝑛𝑦𝑡)2+ 𝛽𝑇𝑡 + 1 2∑𝑖𝜖𝐹𝛾𝑖𝑇𝑡𝑙𝑛𝑝𝑖𝑡+ 1 2𝛾𝑇𝑇𝑡2, (1)
Where 𝐶, 𝑝𝑖, 𝑦 𝑎𝑛𝑑 𝑡 respectively represent the total variable cost, prices for business functions 𝑖(iϵF, F refers to the set of business functions), output and the time trend. Labor demand elasticities ban be inferred from the parameters 𝛽𝑖 and 𝛾𝑖𝑗. 𝛽𝑌 and 𝛾𝑖𝑌 will provide us
with the information on possible scale-bias in production. 𝛽𝑇 indicates the speed of Hicks-neutral technological progress, which serves as an indication of economy-wide technological progress. Accordingly, 𝛾𝑖𝑇 shows a trend of the relationship between technological progress and production factor 𝑖: a positive 𝛾𝑖𝑇 indicates a complementarity effect and a negative
𝛾𝑖𝑇indicated a substitution effect.
We assume cost minimization, and the envelope theorem indicates that 𝜕𝐶 𝜕𝑝⁄ 𝑖 = 𝑥𝑖, and in logarithmic form 𝜕𝑙𝑛𝑝𝜕𝑙𝑛𝐶
𝑖 =
𝜕𝐶 𝜕𝑝𝑖∗
𝑝𝑖
𝐶. Substituting the equation from the envelope theorem we get: 𝜕𝑙𝑛𝐶
𝜕𝑙𝑛𝑝𝑖 = 𝑥𝑖 ∗
𝑝𝑖
𝐶 = 𝑆𝑖, where 𝑆𝑖 is the cost share of the 𝑖th production factor. Thus, by using
Shephard’s lemma we can derive the labor cost share equation for business function 𝑖: 𝑆𝑖𝑡 = 𝛽𝑖+ ∑𝑗𝜖𝐹𝛾𝑖𝑗𝑙𝑛𝑝𝑗𝑡+ 𝛾𝑖𝑌𝑙𝑛𝑦𝑡+ 𝛾𝑖𝑇𝑡, (2)
Where 𝑆𝑖𝑡 = 𝑝𝑖𝑡𝑥𝑖𝑡, and 𝑥𝑖𝑡 is the quantity of labor demand for business function 𝑖. We should
12
a proportional increase in input prices must increase total cost by the same proportion, holding output constant. By taking the total differentiation of the log term of total cost while keeping output constant, we get the following equation:
𝑑𝑙𝑛𝐶 = ∑ [𝛾𝑖 𝑖𝑌𝑙𝑛𝑦𝑡]𝑑𝑙𝑛𝑝𝑖 + ∑ 𝛽𝑖 𝑖𝑑𝑙𝑛𝑝𝑖 +12∑ ∑ 𝛾𝑖 𝑗 𝑖𝑗𝑑𝑙𝑛𝑝𝑖𝑑𝑙𝑛𝑝𝑗, (3)
According to our assumption, 𝑑𝑙𝑛𝑝𝑖 is the same across all inputs. This simplifies equation (3)
to:
𝑑𝑙𝑛𝐶 = 𝑑𝑙𝑛𝑝̅ ∑ [𝛾𝑖 𝑖𝑌𝑙𝑛𝑦𝑡]+ 𝑑𝑙𝑛𝑝̅ ∑ 𝛽𝑖 𝑖+ 𝑑𝑙𝑛𝑝̅2 1
2∑ ∑ 𝛾𝑖 𝑗 𝑖𝑗, (4)
In order to make sure that 𝑑𝑙𝑛𝐶
∗
𝑑𝑙𝑛𝑝̅ = 1, we need to ensure the following constraints to hold:
∑ 𝛽𝑖 𝑖 = 1; (5)
∑ 𝛾𝑖 𝑖𝑌 = 0; (6) ∑ ∑ 𝛾𝑖 𝑗 𝑖𝑗 =0. (7)
In order to analyze the relationship between labor cost shares for groups of business functions, capital and international offshoring, we decompose value added by labor into detailed components, namely value added by R&D, by production and by other activities (further discussed in the next section). We include ICT and non-ICT capital compensation, as well as offshoring as control variables in the system of equations; we then estimate a simultaneous equation system to improve the efficiency of parameter estimates. According to Berndt (1991), the restricted equation system can be estimated by first dropping one equation and then confirming the other equations one by one. We thus drop the cost-share equation for other activities (this choice is arbitrary). We will apply the method of Zeller by using Iterative Seemingly Unrelated Regressions (ISUR) where the choice of which equation to drop is innocuous. Accordingly, the system of unrestricted cost-share equations are listed as below: 𝑆𝑟𝑑𝑡 = 𝛽𝑟𝑑+ 𝛾𝑟𝑑𝑟𝑑𝑙𝑛(𝑝𝑟𝑑𝑡⁄𝑝𝑜𝑡) + 𝛾𝑟𝑑𝑝𝑙𝑛(𝑝𝑝𝑡⁄𝑝𝑜𝑡) + 𝛾𝑟𝑑𝑖𝑐𝑡𝑙𝑛𝑖𝑐𝑡𝑡 + 𝛾𝑟𝑑𝑛𝑖𝑐𝑡𝑙𝑛𝑛𝑖𝑐𝑡𝑡+ 𝛾𝑟𝑑𝑜𝑓𝑓𝑡𝑙𝑛𝑜𝑓𝑓𝑡+ 𝛾𝑟𝑑𝑦𝑙𝑛𝑦𝑡+ 𝛽𝑟𝑑𝑇𝑡, (8) 𝑆𝑝𝑡 = 𝛽𝑝+ 𝛾𝑝𝑟𝑑𝑙𝑛(𝑝𝑝𝑡⁄𝑝𝑜𝑡) + 𝛾𝑝𝑝𝑙𝑛(𝑝𝑝𝑡⁄𝑝𝑜𝑡)+ 𝛾𝑝𝑖𝑐𝑡𝑙𝑛𝑖𝑐𝑡𝑡 + 𝛾𝑝𝑛𝑖𝑐𝑡𝑙𝑛𝑛𝑖𝑐𝑡𝑡+ 𝛾𝑝𝑜𝑓𝑓𝑡𝑙𝑛𝑜𝑓𝑓𝑡+ 𝛾𝑝𝑦𝑙𝑛𝑦𝑡+ 𝛽𝑝𝑇𝑡 , (9)
Where rd, p, o, ict, nict represent R&D activities, production activities, other activities, ICT-capital compensation and non-ICT ICT-capital compensation, respectively. It is easy to see that the bias in technical change is modelled as linear trends in this system of equations. Baltagi and Griffin (1988) propose a model in which the time trend 𝑡 is replaced by year dummies, with the first year as the base year. Thus, for a business function 𝑖, 𝛾𝑖𝑡𝑡 is replaced by ∑13 𝜆𝑖𝑡
13
where 𝐷𝑡 represents year dummies and ∑𝑖𝜖𝐹𝜆𝑖𝑡 = 0 is for all t. By also estimating this specification, we take into account the possibility that the time trend is non-linear.
Besides reporting parameter estimates of the cost share functions, we will also calculate the elasticities of substitution among labor inputs for various business functions. In equation (1), a positive coefficient of 𝛾𝑖𝑗 represents a net substitution between business function 𝑖 and 𝑗. According to the so-called Allen-Uzawa partial elasticities of substitution:
𝜎𝑖𝑗 = 𝛾𝑖𝑗
𝑆𝑖𝑆𝑗+ 1 (for 𝑖 ≠ 𝑗) (10)
Where 𝜎𝑖𝑗 is the substitution elasticity between business function 𝑖 and 𝑗. If 𝜎𝑖𝑗 > 1, there is a net substitution between business function 𝑖 and 𝑗; if 𝜎𝑖𝑗 < 1, there is a net complementarity
between business function 𝑖 and 𝑗.
Similarly, the price elasticity of demand for business function 𝑖 with respect to price of business function 𝑗 (𝜀𝑖𝑗) is given as follows:
𝜀𝑖𝑗 = 𝜎𝑖𝑗𝑠𝑗, (for 𝑖 ≠ 𝑗) (11)
𝜀𝑖𝑗 =𝛾𝑆𝑖𝑖
𝑖 + 𝑠𝑖− 1, (12)
Where 𝑠 is the average cost shares of a certain business function.
4. Data
4.1 Data Construction
In this study we combine three databases, namely the World Input-Output Database (WIOD), data on the occupational employment structures and the EUKLEMS database.
4.1.1 The World Input-Output Database (WIOD)
WIOD provides annual world input-output tables for a large set of countries and detailed industries from 1995 onwards (Timmer et al. 2015). These tables track where intermediate inputs are being sourced from.
Our measure of offshoring is from WIOD. The calculation is based on the definition of offshoring from Feenstra and Hanson (1999), which measures offshoring as the share of imported intermediate inputs in the total purchase of intermediates. Thus, our offshoring data mainly measures how much of the intermediate inputs are imported from abroad rather than produced domestically. With this offshoring data, we aim to capture the effect of trade on labor demand for different groups of business functions.
14
persons engaged by country and industry. We split the labor share for each industry in each country into the share from workers in R&D, production and other activities. For that, we make use of labor force surveys, which is discussed next.
4.1.2 Occupation data
Our study uses a business function framework, which provides information and evidence on cross-country distributions of business activities. For this purpose, we need data on occupations. The sources and approaches for the set of occupation by industry data differ across countries. For EU 27 countries, the yearly occupation by industry data is from the European Labor Force Survey (EU LFS), the EU LFS is a broad household survey, which offers information on the labor force participation rate for individuals who are 15 years or older. The EU LFS is carried out by the national statistical institutes across all 27 EU countries and the results are coordinated and processed by Eurostat. In addition to employment data, wage data is also needed to calculate shares in labor compensation. We obtain wage data from the Structure of Earnings Surveys (SES), waves 2002 and 2006. By applying the harmonized data on earnings in EU countries from the SES, relative wages by 2-digit occupation for each European country are computed.
Employment and wage data for the US are generated from the Occupational Employment Statistics (OES). The OES contains information on around 800 occupations on industry level. The data is classified by SIC from 1997 to 2001 and by NAICS from 2002 onwards. Timmer and de Vries (2015) further convert all industry codes into the ISIC rev. 3 classification. For Japan, employment data is from the very detailed five yearly occupational employment data by industry from the Japan Population Census for years 1995, 2000, 2005 and 2010. Wage data is from the wage structure surveys for the same years with the employment data. Occupations are then matched to the business functions on our list (see Table 14 for several examples), and accordingly, we measure the share of each business function in total labor compensation by country and industry from WIOD.
The most important process in our study is to match occupations and their wages to the business functions on our list. We use the groups of business functions according to Sturgeon and Gereffi (2009). Particularly, we have eleven business functions. We classify them into three big groups, namely research and development activities (R&D, technology and process development); production activities (operations and primary activity of the business; assemblers); other activities (sales and marketing; transportation, logistics and distribution; customer and after-sales service; facilities maintenance; administration, and back office functions; general and strategic management; other). However, we will consider different groupings of business functions in an extension analysis later on.
4.1.3 EUKLEMS database
15
version. The specific included industries are somewhat different between these two versions, hence, we make some changes in data aggregation (see Appendix).
Even though the WIOD includes 40 countries, we still end up with a smaller sample owing to missing data for some variables of some countries. Specifically, there are no data of business function shares, labor cost shares, capital, ICT-capital share and non-ICT capital share for some countries. As a result, we end up with 19 countries, namely Austria, Belgium, Czech Republic, Denmark, Spain, Finland, France, Germany, Hungary, Ireland, Italy, Japan, Luxembourg, the Netherlands, Portugal, Slovenia, Sweden, the UK and the US. To avoid the disturbances of the economic crisis after 2008, our analysis is restricted to the period from 1995 to 2008.
Besides, our main analysis includes all the 35 industries in the WIOD, among which are agriculture, various manufacturing and various service industries. Existing research mainly focuses on manufacturing industries since offshoring is always thought to have effect only or at least mainly on manufacturing jobs and production activities (Brown, 2008). However, According to Brown et al. (2014), in the early 2000s, offshoring activities began to spread to service industries as well. Especially job losses in information technology related industries came into being an important ongoing change in industries. We thus find it important to also include service industries in our sample so that the influence of service industries on job losses can be incorporated and investigated as well. Furthermore, agricultural industries have also increasingly involve modern technology, thus it is also relevant to our study.
4.2 Descriptive Statistics
Table 1 reports the average cost shares of labor classified by three business functions between 1995 and 2008. We see that during this period, the average cost share of labor related to R&D activities is 10.9% of all labor, which is the lowest among the three; the average cost share of labor related to production activities is 30% of all labor, which is much higher than that of R&D activities.
Table 2 displays details on percentage points change in labor cost shares and percentage change in labor prices and quantities from 1995 to 2008. On average, labor that are related to R&D activities see the biggest increase in cost shares (2.95%), then are labor who are related to other activities (0.69%), but labor that are related to production activities decreases in cost shares (-3.63%). All labor experiences relatively similar level of increase in their price regardless of which business functions they are related to, which is around 50%. However, labor who are related to R&D activities (203.15%) far exceed those who are related to production (66.97%) and other activities (60.26%) in its increase in quantity. These numbers mainly indicate that for the past 20 years, advanced countries experienced an increase in cost shares of labor who conduct R&D related activities compared to those who conduct production and other activities. The main reason is because of the very high increase in employment in R&D activities than production and other activities.
16
investment and offshoring. In 2008, offshoring is 5 times higher than that in 1995, and for ICT capital compensation, it is around 1.23 times higher. It indicates that during this period, both investment in ICT capital and offshoring activities play an important role in the development of advanced nations and their industries.
5. Results
The main results of the Iterated seemingly unrelated regression (ISUR) are reported in table 4. We estimate three different regression specifications, namely pooled ISUR, fixed-effect ISUR with country and product dummies and fixed-effect ISUR with country, product and year dummies. We include a fixed effect with year dummies estimation in order to account for non-linear economy wide technological change. After comparing the time trend of fixed effect ISUR with fixed effect ISUR with year dummies (Figure 2), we find that the time trend is almost linear. It basically indicates that fixed effect ISUR is good enough to account for the economy wide technological change. Furthermore, a Hausman test clearly rejects the use of pooled ISUR and suggests the use of fixed effect ISUR. As a result, we will choose to report the results of fixed effect ISUR specification only.
It is crucial to check whether the estimated cost function is consistent with economic theory, since we assume cost minimization. It basically means that the cost functions are reasonable and well-behaved only if they are quasi-concave in factor prices. This requires that the Hessian matrix of the second-order derivatives regarding certain factor prices must be negative semi-definite. To test this, we check whether the Hessian matrix of (𝐻 − 𝐷𝑖𝑎𝑔(𝑣) + 𝑣𝑣′) is negative semi-definite (Diewert and Wales, 1987). 𝐻 is a symmetric matrix that
includes all factors of elasticities of substitution 𝜎𝑖𝑗, and 𝑣 is a column vector that includes all
labor cost shares for certain business functions. To check whether the Hessian matrix is negative semi-definite, we need to evaluate eigenvalues of this matrix for all observations. After checking the quasi-concavity for all observations for the fixed effect ISUR specification, we find that only 22.5% observations (1737 out of 7715) are problematic, so that these observations have positive eigenvalues. This suggests that in most of the situations, the Hessian matrix associated with the cost functions that we estimate is negative semi-definite. According to Hijzen et al. (2005), we need to make sure that curvature conditions are satisfied on average. Apparently our case satisfies this requirement. We can thus believe that the estimated translog cost function is consistent with the economic theory and cost minimization assumption.
17
equation with one cost share of labor input as dependent variable, in our case, it is the cost share of labor related to other activities. Since the restriction on cost shares for all labor inputs is that ∑𝑖𝛾𝑖𝑗 = 0, then 𝛾𝑖𝑜= −(𝛾𝑖𝑟𝑑+ 𝛾𝑖𝑝), where rd, p and o represent R&D, production and other activities respectively. Hence, we can also calculate and report the implicit estimates for the other activities’ labor cost share.
5.1 Key Results
5.1.1 Technological change and international offshoring
Table 4 suggests that economy wide technological change is significantly positive related with labor cost shares of R&D (𝛽𝑟𝑑𝑇 > 0)and other activities(𝛾𝑜𝑇 > 0), but significantly negative related with labor cost share of production activities(𝛽𝑝𝑇< 0). It thus suggests that,
as technology develops, the demand for R&D related workers increases. To the contrary, however, the demand for production activities related workers decreases, which indicates that the demand for production related workers decreases as technology develops. It is highly likely that technology complements those activities that require information as the main input, but substitutes those activities that can be easily codified and programmed by computers or done by machines (Autor, Levy and Murnane, 2003). Since R&D activities relatively need more information as input and production activities are generally easily programmed by computers or done by machines, the demand for workers who are involved in R&D activities increases, but the demand for workers who are involved in production activities decreases as technology develops. Furthermore, the demand for other activities related workers increases, since there are many headquarter and management related activities in other activities, the reason for this can be that there is a higher demand for management and organization workers in order to manage the increasing R&D activities and offshoring activities.
As for international offshoring, we use the measure of narrow offshoring. Results show that offshoring is significantly positive related with labor cost share of R&D activities (𝛾𝑟𝑑𝑜𝑓𝑓>
0), significantly negative related with labor cost share of production activities (𝛾𝑝𝑜𝑓𝑓 < 0)
but insignificantly negative related with other activities (𝛾𝑜𝑜𝑓𝑓< 0) . It is very likely that as
technology develops, advanced countries increasingly specialize in R&D activities, which needs more information as input, but offshore some production activities that are easily codified, programmed and done by machines to countries with lower labor cost, such as China and India, etc. As a result, the labor demand for R&D activities increases, but the labor demand for production activities decreases.
5.1.2 ICT and non-ICT related capital
Both ICT and non-ICT related capital compensation are included in the regression as control variables. Results indicate that ICT related capital compensation is significantly positive correlated with labor cost share of R&D activities(𝛾𝑟𝑑𝑖𝑐𝑡 > 0), but significantly negative correlated with labor cost share of production activities(𝛾𝑝𝑖𝑐𝑡< 0); non-ICT related capital
18
correlated with labor cost share of production activities(𝛾𝑝𝑖𝑐𝑡> 0). The results basically suggest that ICT capital benefits R&D activities, but harms production activities. This reasoning is consistent with the “routinization hypothesis” which states that technological change (ICT) complements activities that need more information as input, but substitutes activities that are easily programmed and codified by computers or done by machines. The similar line of reasoning applies to the relationship between non-ICT related capital compensation and labor cost shares of certain groups of activities.
5.1.3 Relationships between groups of activities
Table 5 reports implied price elasticities and implied elasticities of substitution among labor inputs classified by business functions assessed at the average of the cost shares. For price elasticity, it is clear that it is negative for labor cost share of all the activities and their own prices, which indicates that an increase in the labor price of a business function would decrease the demand for the labor related to this business function. It is consistent with the assumption of concavity of the cost functions. This effect is strongest for R&D related activities but weakest for other activities. Furthermore, implied price elasticities are positive between any group of labor and the other groups’ labor price.
For implied elasticity of substitution between labor inputs, the results show that complementary effects exist among all business functions, since all elasticities are smaller than one. Specifically, we are mostly interested in the complementary relationship between R&D and production functions. This is in line with some existing research like Baldwin et al (2001), which assumes that R&D is located in the same place as production stage, and that firms’ private knowledge is globally immobile. Markusen (2005) also argues that firms would consider the complementarity effects between different stages of the value chain, which could also benefit the firms by reducing coordination costs between stages. This complementarity effect could lead to co-location of value chain stages in the same country due to vertical linkages between stages like R&D and production.
5.2 Robustness and Extensions 5.2.1 Manufacturing industries only
19
Column 2 of table 6 shows that similar to the full sample analysis, economy wide technological change is significantly positive related with labor demand for R&D activities, but significantly negative related with labor demand for production activities. However, offshoring is not significantly correlated with labor demand for any business functions. This is quite to the opposite of the existing research, which argues that manufacturing industries used to be the ones that influenced the most by offshoring activities and offshoring benefits high-skilled intensive activities and harms low-high-skilled intensive activities. It is likely because that our study classifies labor by business functions, which is different from existing research that classifies labor by skill levels, so that the system of our analysis is different in a way that each business function includes different skilled type of labor. As a result, the results we get and their line of reasoning are not necessarily the same with the existing research since we are studying changes in labor demand on different levels. Furthermore, ICT capital compensation is significantly positive associated with labor demand for R&D activities, but not for production and other activities.
We also include an estimation for service industries only. Results are displayed in column 3 of table 6, which show that offshoring is significantly positive related with labor cost share of R&D activities, but significantly negative related with labor cost share of production activities. This is in line with the evidence that since early 2000s, offshoring gradually extended and affected service industries as well, and especially information and communication related industries, thus service industries become another very crucial part of ongoing effect of offshoring besides manufacturing industries.
Our results thus indicate that during the past two decades, offshoring tends to have a more significant effect on service industries instead of manufacturing industries.
For implied elasticity of substitution for manufacturing industries, the results are reported in table 7. They are very similar with that of the main study, and for our main focus, R&D and production activities are still complementary to each other.
5.2.2 Different groupings of business functions
20
For all industries’ analysis, table 8 indicates that economy wide technological progress is significantly positive related with labor cost share of R&D, logistics and sales functions, but significantly negative related with labor cost share of production function. This suggests that economy wide technological progress has a complementary effect on R&D, logistics sand sales functions, but a substitution effect on production function. Offshoring is significantly positive related with labor cost share of R&D activities at 10% level, but insignificantly related with labor cost share of production functions. ICT capital compensation is still significantly positively associated with labor cost share of R&D activities, but significantly negative associated with labor cost share of production activities. The results regarding the relationship between offshoring and labor cost shares of production activities are quite different between the main study and this extension study with new classification of business functions, since the result from the extension study loses significance. The translog cost function framework we apply has all the advantages we mentioned in the previous paragraph, however, it also has its own constraints. When the number of equation increases, the parameters that we need to estimate simply explode. On the one hand, this could lead to a problem of insufficient observations; on the other hand, this may also lead to a harmful collinearity problem. These potential problems can certainly affect the feasibility of the results in a negative way. Take this into account when analyzing our results, we find that the extension study adds two more equations to the main study, thus it potentially creates more burden and problems for the translog cost function framework to achieve a more feasible result. We thus tend to trust the results from the main study more than that from the extension study.
For implied elasticity of substitution, results are displayed in table 10, which suggests that only R&D and sales activities are substitutes of each other, but the relationship between any other two business functions are complementary. This basically suggests a possibility that business functions except for sales activities can benefit from locating together, however, the situation for sales activities are somewhat uncertain. Defever (2012) has a similar finding with us, which shows that all business functions of firms can benefit from co-location except for sales activities, which tend to spread across different locations. It is thus very likely that R&D activities tend to locate in a few main R&D centers together with some other business functions, but on the other hand, sales activities are better located to many different places that are near to as many consumers as possible. For our main research interest, same with the main study, R&D and production activities are complementary to each other.
21
R&D activities. Since advanced countries specialize in R&D activities, many production activities come back to domestic countries, and as the blooming of R&D activities, the demand for production activities also increases because of the complementary effect. As a result, offshoring is positively related to labor cost share of production activities.
For implied elasticity of substitution, results from table 13 suggest that same with that for all the industries, only R&D and sales functions are substitutes of each other. Specifically for our interest, R&D and production activities are complementary to each other. The line of reasoning for the results are the same with that for all industries’ case.
To sum up, from the results of implied elasticity of substitution, we see that R&D, production and logistics functions are all complementary to each other. This indicates that these functions can benefit from locating together. Furthermore, sales and R&D functions are substitutes of each other, hence they are not necessarily located together. Existing research like Defever (2012) suggests that sales activities tend to spread across locations, thus it is likely that R&D and other business functions co-locate in some main centers, but as an exception, sales activities spread across locations to better reach more consumers.
6 Concluding remarks
22
23
Reference
Acemoglu, D. and Autor, D (2010), “Skills, Tasks and Technologies: Implications for Employment and Earnings”. NBER Working Paper, 16082.
Autor, D., Dorn, D., and Hanson, G. (2013), “The China Syndrome: Local Labor Market Effects of Import Competition in the United States”, American Economic Review, 103(6), 2121-2168.
Autor, D.H., Levy.F and Murnane, R.J. (2003), “The Skill Content of Recent Technological Change: An Empirical Exploration”, The Quarterly Journal of Economics, 1279-1333.
Baldwin, R., Martin, P and Ottaviano, G. (2001), “Global Income Divergence, Trade and Industrialization: the Geography of Growth Take-off”, Journal of Economic Growth, 6, 5-37. Baltagi, B.H., and Griffin, J.M. (1988), “A General Index of Technical Change”, Journal of Political Economy, 96(1), 20-41.
Baltagi, B. H., and Rich, D. P. (2005). “Skill-biased Technical Change in US Manufacturing: A General Index Approach”, Journal of Econometrics, 126(2), 549-570.
Becker, S. O., Ekholm, K., Jackle, R., and Muendler, M. (2005), “Location Choice and Employment Decisions: A Comparison of German and Swedish Multinationals.” Weltwirtschaftliches Archiv (Review of World Economics), 141(4), 693–731.
Berman, E., Bound, J. and Griliches, Z. (1994). “Changes in the Demand for Skilled Labor within US Manufacturing: Evidence From the Annual Survey of Manufacturers”, Quarterly Journal of Economics, 109, 367-397.
Berndt, E. (1991), The Practice of Econometrics: Classic and Contemporary.
Brown, S. P. (2008), “Business Processes and Business Functions: a new way of looking at employment”, Monthly Labor Review, 51-70.
Brown, C., Sturgeon, T. and Lane, J. (2014). “Using a business function framework to examine outsourcing and offshoring by US organizations”, IRLE Working Paper, 1214, 1-15.
Christensen, L. R., Jorgenson, D. W., and Lau, L. J. (1973). “Transcendental Logarithmic Production Frontiers”, Review of Economics and Statistics, 55(1), 28-45.
Defever, F. (2006), “Functional Fragmentation and the Location of Multinational Firms in the Enlarged Europe”, Regional Science and Urban Economics, 36(5), 658-677.
24
Diewert, W., and Wales, T.J. (1987), “Flexible Functional Forms and Global Curvature Conditions”, Econometrica, 55(1), 43-68.
Disdier, A., and Thierry, M. (2004), “How Different is Eastern Europe? Structure and Determinants of Location Choices by French Firms in Eastern and Western Europe.” Journal of Comparative Economics, 32(2), 280–96.
Enright, M.J. (2009), “The location of activities of manufacturing multinationals in the Asia-Pacific”, Journal of International Business Studies, 40, 818-839.
Feenstra, R.C. and Hanson, G.H. (1999), “The impact of Outsourcing and High-Technology Capital on Wages: Estimates for the United States, 1979-1990”, The Quarterly Journal of Economics, 114(3), 907-940.
Foster-McGregor, N., Stehrer, R., and de Vries, G. J. (2013). “Offshoring and the Skill Structure of Labor Demand”, Review of World Economics, 149(4), 631-662.
Gereffi, G., and Fernandez-Stark, K. (2010). “The Offshore Services Value Chain: Developing Countries and the Crisis”, World Bank Policy Research Working Paper No. 5262. Goos, M., Manning, A., and Salomons, A. (2014). “Explaining Job Polarization: Routine-Biased Technological Change and Offshoring”, American Economic Review, 104(8), 2509-2526.
He, C., and Xiao, X. (2011). “Geography of Multinational Corporations and Functional Specialization in Chinese Cities”, Emerging Issues in Management, 1, 29-40.
Hijzen, A., Gorg, H., and Hine, R. C. (2005). “International Outsourcing and the Skill Structure of Labor Demand in the United Kingdom”, Economic Journal, 115(506), 860-878. Markusen, J.R. (2005), “Modelling the Offshoring of White-Collar Services: from Comparative Advantage to the New Theories of Trade and FDI”, NBER Working Paper, vol.11827.
Maurin, E., and Thesmar, D. (2004), “Changes in the Functional Structure of Firms and the Demand for Skill”, Journal of Labor Economics, 22(3), 639-664.
Michaels, G., Natraj, A and van Reenen, J. (2013), “Has ICT Polarized Skill Demand? Evidence from Eleven Countries over 25 years”, The Review of Economics and Statistics, 96, 60-77.
Muendler, M., and Becker, S.O. (2010), “Margins of Multinational Labor Substitution”, American Economic Review, 100(5), 1999-2030.
25
Sturgeon, T., and Gereffi, G. (2009), “Measuring Success in the Global Economy: International Trade, Industrial Upgrading, and Business Function Outsourcing in Global Value Chains”, Transnational Corporations, 18(2), 1-35.
Sturgeon, T.J., Nielsen, P.B., Linden, G., Gereffi, G. and Brown, C. (2012), “Direct Measurement of Global Value Chains: Collecting Product- and Firm-Level Statistics on Value Added and Business Function Outsourcing and Offshoring ”, Forthcoming as Chapter 9 in the World Bank volumn: The Fragmentation of Global Production and Trade in Value-Added-Developing New Measures of Cross Border Trade, based on a Trade Workshop held at World Bank Headquarters.
Timmer, M.P., and de Vries, G.J. (2015), “Functional Specialization in International Production Chains: An Exploration based on Occupational Data”.
Timmer, M.P., and Ye, X. (2014), “Offshoring, Biased Technical Change and the Increasing Capital Share: an Analysis of Global Manufacturing Production”, Paper to be presented at UNU-MERIT conference 26-28 November 2014.
26
Table 1. Average labor cost shares for various business functions, 1995-2008
Variable Obs Mean Std. Dev. Min Max
rls 7715 10.9 9.9 0.0 63.7
pls 7715 30.0 23.1 0.0 96.4
ols 7715 59.1 25.0 0.0 99.5
Note: Average labor cost shares for three business functions in total labor cost. Rls, pls and ols denote labor cost share of R&D, production and other activities respectively. Observations are averaged across 19 countries, 35 industries and 14 years.
Table 2. Changes in labor cost shares of various business functions (% points change between 1995 and 2008), prices and quantities of labor classified by business functions (% change between 1995 and 2008)
Factor Mean Std.Dev 5% pct 25% pct Median 75% pct 95% pct Cost shares R&D 2.95 7.50 -6.55 -0.46 1.97 5.33 17.06
Production -3.63 11.71 -20.51 -8.73 -1.97 1.67 13.23 Other Activities 0.69 11.50 -17.32 -4.47 0.57 5.93 17.28 Price R&D 50.24 42.16 -9.23 24.43 50.92 69.65 112.79 Production 62.80 77.49 -6.29 23.18 45.48 83.56 185.91 Other Activities 49.26 41.60 -7.02 28.04 45.32 66.42 118.74 Quantity R&D 203.15 774.21 -74.16 -13.73 39.48 147.48 613.93 Production 66.97 384.71 -69.87 -36.39 -7.04 36.32 341.64 Other Activities 60.26 302.45 -65.64 -16.29 11.45 46.61 216.40 Note: Mean, standard deviation and percentile distribution of changes in business function labor cost shares, prices and quantities. Unweighted average across 391 observations. (19 countries times 35 industries, missing values automatically dropped.)
Table 3. Changes in ICT capital compensation and narrow offshoring (% change between 1995 and 2008)
Variable Mean Std.Dev 5% pct 25% pct Median 75% pct 95% pct ICT capital compensation 123.44 569.88 -91.91 12.48 102.56 196.68 591.92 Narrow offshoring 569.41 7039.45 -57.25 -14.67 29.01 103.28 621.57 Note: Mean, standard deviation and percentile distribution of changes in ICT capital
compensation and narrow offshoring. Unweighted average across 712 observations for narrow offshoring and 510 observations for ICT capital compensation. (19 countries times 35
27
Table 4. Determinants of labor cost shares for various business functions, all products
Fixed Effect ISUR Pooled ISUR Fixed Effect with year dummies
Variable Coef Std.E Coef std.E Coef std.E
𝛽𝑟𝑑 0.0275 0.0102 ** 0.2390 0.0053 *** 0.0276 0.0105 ** 𝛽𝑝 0.4047 0.0157 *** 0.3047 0.0108 *** 0.3999 0.0162 *** 𝛽𝑟𝑑𝑇 0.0017 0.0002 *** 0.0015 0.0003 *** - 𝛽𝑝𝑇 -0.0034 0.0002 *** -0.0032 0.0005 *** - 𝛾𝑟𝑑𝑟𝑑 0.0202 0.0035 *** 0.0313 0.0052 *** 0.0200 0.0035 *** 𝛾𝑟𝑑𝑝 -0.0154 0.0024 *** -0.0223 0.0034 -0.0151 0.0024 *** 𝛾𝑝𝑝 0.1084 0.0038 *** -0.1397 0.0067 *** 0.1079 0.0038 *** 𝛾𝑟𝑑𝑖𝑐𝑡 0.0053 0.0008 *** 0.0255 0.0010 *** 0.0053 0.0008 *** 𝛾𝑟𝑑𝑛𝑖𝑐𝑡 -0.0021 0.0008 ** 0.0010 0.0009 -0.0021 0.0008 ** 𝛾𝑟𝑑𝑜𝑓𝑓 0.0010 0.0005 * 0.0116 0.0005 *** 0.0011 0.0005 * 𝛾𝑝𝑖𝑐𝑡 -0.0057 0.0013 *** -0.0572 0.0020 *** -0.0060 0.0013 *** 𝛾𝑝𝑛𝑖𝑐𝑡 0.0052 0.0013 *** 0.0212 0.0020 *** 0.0052 0.0013 *** 𝛾𝑝𝑜𝑓𝑓 -0.0010 0.0008 *** 0.0445 0.0009 *** -0.0010 0.0008 Implied 𝜸 associated with other activities
𝛾𝑟𝑑𝑜 -0.0048 0.0032 -0.0090 0.0047 * -0.0049 0.0032 𝛾𝑝𝑜 -0.0930 0.0035 *** 0.1621 0.0066 *** -0.0928 0.0035 *** 𝛾𝑜𝑜 0.0978 0.0046 *** -0.1531 0.0080 *** 0.0978 0.0046 *** 𝛾𝑜𝑦 -0.0153 0.0021 *** -0.0135 0.0024 *** -0.0154 0.0021 *** 𝛾𝑜𝑇 0.0017 0.0002 *** 0.0017 0.0006 *** 𝛾𝑜𝑜𝑓𝑓 -0.0001 0.0009 -0.0562 0.0010 *** -0.0001 0.0008 𝛾𝑜𝑖𝑐𝑡 0.0005 0.0013 0.0316 0.0021 *** 0.0007 0.0013 𝛾𝑜𝑛𝑖𝑐𝑡 -0.0031 0.0013 ** -0.0222 0.0020 *** -0.0031 0.0013 **
Country Dummies YES NO
YES
Product Dummies YES NO YES
Year Dummies NO NO YES
Number of Observations 7715 7715 7715
R2− rdS 0.7726 0.2197 0.7730
R2− pS 0.8997 0.3785 0.8999
Note: Estimation of parameters that determines labor cost share for business functions are according to equation (8) and (9). Standard errors are reported in columns next to parameter estimates. ***,** and * refer to 0.1%, 1% and 5% significance levels, respectively. Subscripts refer to R&D function (rd), production function (p), other functions (o), ICT capital (ict), non-ICT capital (nict), offshoring (off) and output (y). Parameters related to other functions are implicitly derived by applying the parameter restrictions discussed in the main text. R2 are
28
Table 5 business function demand elasticities
Implied Price Elasticity Implied Elasticity of Substitution
wrd wp wo rd p o
rd -0.706 0.155 0.551 - 0.523 0.926
p 0.057 -0.338 0.281 0.523 - 0.472
o 0.101 0.140 -0.241 0.926 0.472 -
29
Table 6 Determinants of labor cost shares for various business functions, manufacturing and service industries
Manufacturing industries
Service industries
Variable Coef Coef
30
Table 7 Business function demand elasticities, manufacturing industries only
Implied Price Elasticity Implied Elasticity of Substitution
wrd wp wo rd p o
rd -0.560 0.112 0.449 - 0.224 1.216
p 0.030 -0.134 0.104 0.224 - 0.283
31
Table 8 Determinants of labor cost shares for various business functions, all products (classification of business functions according to Defever (2012))
Fixed effect ISUR
Variable Coef Std.E
𝛽𝑟𝑑 -0.0026 0.0098 𝛽𝑝 0.4094 0.0161 *** 𝛽𝑙 0.1314 0.0090 *** 𝛽𝑠 0.0833 0.0099 *** 𝛽𝑟𝑑𝑇 0.0014 0.0001 *** 𝛽𝑝𝑇 -0.0035 0.0002 *** 𝛽𝑙𝑇 0.0008 0.0001 *** 𝛽𝑠𝑇 0.0013 0.0001 *** 𝛾𝑟𝑑𝑟𝑑 0.0198 0.0033 *** 𝛾𝑟𝑑𝑝 -0.0124 0.0024 *** 𝛾𝑟𝑑𝑙 -0.0101 0.0026 *** 𝛾𝑟𝑑𝑠 0.0214 0.0018 *** 𝛾𝑝𝑝 0.1093 0.0039 *** 𝛾𝑝𝑙 -0.0172 0.0024 *** 𝛾𝑝𝑠 -0.0215 0.0021 *** 𝛾𝑙𝑙 0.0716 0.0044 *** 𝛾𝑙𝑠 -0.0112 0.0019 *** 𝛾𝑠𝑠 0.0425 0.0022 *** 𝛾𝑟𝑑𝑖𝑐𝑡 0.0051 0.0008 *** 𝛾𝑟𝑑𝑛𝑖𝑐𝑡 -0.0019 0.0008 ** 𝛾𝑟𝑑𝑜𝑓𝑓 0.0010 0.0005 10% 𝛾𝑝𝑖𝑐𝑡 -0.0038 0.0013 *** 𝛾𝑝𝑛𝑖𝑐𝑡 0.0050 0.0013 *** 𝛾𝑝𝑜𝑓𝑓 -0.0002 0.0009
Implied 𝛾 associated with other activities
32
Note: Subscripts refer to R&D function (rd), production function (p), logistics function (l), sales function (s) and other functions (o) respectively.
Table 9 Implied price elasticities of business functions, all products (classification of business functions according to Defever (2012))
Implied Price Elasticity
wrd wp wl ws wo rd -0.708 0.185 -0.003 0.284 0.242 p 0.066 -0.336 0.033 0.014 0.222 l -0.004 0.110 -0.122 -0.037 0.052 s 0.357 0.049 -0.039 -0.419 0.052 o 0.063 0.160 0.011 0.011 -0.245
Table 10 Implied elasticity of substitution of business functions, all products (classification of business functions according to Defever (2012))
Implied Elasticity of Substitution
33
Table 11 Determinants of labor cost shares for various business functions, manufacturing industries only (classification of business functions according to Defever (2012))
Fixed effect ISUR
Variable Coef Std.E
34
Table 12 Implied price elasticities of business functions, manufacturing industries only (classification of business functions according to Defever (2012))
Implied Price Elasticity
wrd wp wl ws wo rd -0.566 0.206 -0.036 0.215 0.181 p 0.053 -0.163 0.049 -0.012 0.073 l -0.075 0.401 -0.371 0.020 0.025 s 0.391 -0.087 0.018 -0.533 0.211 o 0.098 0.153 0.006 0.063 -0.321
Table 13 Implied elasticity of substitution of business functions, manufacturing industries only (classification of business functions according to Defever (2012))
Implied Elasticity of Substitution
