Offshoring and the Skill Structure of Labour Demand in Belgium

Abstract

A major concern regarding the consequences of offshoring is the worsening of the labour market position of low-skilled workers. This paper addresses this issue by providing evidence on the impact of offshoring on the skill structure of manufacturing employment in Belgium between 1995 and 2007. Offshoring is found to significantly lower the employment share of low-skilled workers. Its contribution to the fall in the employment share of low-skilled workers amounts to 35 %. This is mainly driven by offshoring to Central and Eastern European countries. While most of the previous papers on this subject focus on materials offshoring, we show that offshoring of business services also contributes significantly to the fall in the low-skilled employment share. As a complement to the existing literature, we compare the widely used current price measure of offshoring with a constant price measure that is based on a deflation with separate price indices for domestic output and imports. This reveals that the former underestimate the extent of offshoring and its impact on low-skilled employment. Finally, we also find that the impact of offshoring on low-skilled employment is significantly smaller in industries with a higher ICT capital intensity.

Appendix

Translog production and cost functions were introduced in the first half of the seventies (Christensen et al. 1971) and have been frequently used in empirical work since then. They allow substitution elasticities to be unrestricted and they are nonhomothetic, i.e. cost-minimizing relative input demands may depend on the level of output (Tables 9, 10).³⁴

Table 9 Data sources

Table 10 Descriptive statistics

Denoting total variable costs $C$, the prices of $N$ variable input factors $P_{j}$ and output $Y$, the general formulation of the translog cost function is as follows:³⁵

$$\begin{aligned} \ln C&= \beta _0 +\mathop \sum \limits _{j=1}^N \beta _j \ln P_j +\frac{1}{2}\mathop \sum \limits _{j=1}^N \mathop \sum \limits _{k=1}^N \delta _{jk} \ln P_j \ln P_k +\beta _Y \ln Y\nonumber \\&\quad +\frac{1}{2}\delta _{YY} \left( {\ln Y} \right) ^{2}+\mathop \sum \limits _{j=1}^N \delta _{jY} \ln P_j \ln Y \end{aligned}$$

(6)

In a classic KLEMS framework, Eq. (6) represents a five-factor model ($N=5$) with capital (K), labour (L) and energy (E), materials (M) and services (S) inputs as variable factors. Labour can further be divided into different skill levels (Tables 11, 12 and 13).

Table 11 GMM estimation results with total offshoring intensities

Table 12 Own-price elasticities for low-skilled and high-skilled workers for estimations with total offshoring intensities

Table 13 Descriptive statistics for high-tech and low-tech industries (unweighted averages)

In Eq. (6), $N(N-1)/2$ symmetry conditions ($\delta _{jk} =\delta _{kj} )$ can be imposed without loss of generality. Moreover, a ‘well-behaved’ cost function should be homogeneous of degree 1 in prices, i.e. a proportional increase in all variable input prices shifts total variable costs by the same proportion. This implies the following restrictions:

$$\begin{aligned} \mathop \sum \limits _{j=1}^N \beta _j =1;\quad \mathop \sum \limits _{j=1}^N \delta _{jk} =\mathop \sum \limits _{k=1}^N \delta _{jk} =\mathop \sum \limits _{j=1}^N \delta _{jY} =0. \end{aligned}$$

(7)

According to Shephard’s lemma, the cost-minimizing input quantities $X_j$ can be derived by differentiating total costs with respect to the prices of the input factors:

$$\begin{aligned} \frac{\partial C}{\partial P_j }=X_j. \end{aligned}$$

(8)

From this, one obtains a set of $N$ cost share equations, which add up to 1.

$$\begin{aligned} S_j&= \beta _j +\mathop \sum \limits _{k=1}^N \delta _{jk} \ln P_k +\delta _{jY} \ln Y\end{aligned}$$

(9)

$$\begin{aligned} \mathop \sum \limits _{j=1}^N S_j&= \mathop \sum \limits _{j=1}^N \frac{P_j X_j }{C}=1 \end{aligned}$$

(10)

The own price elasticities $\varepsilon _{jj}$ and cross price elasticities $\varepsilon _{jk}$ are given by:

$$\begin{aligned} \varepsilon _{jj} =\frac{\delta _{jj} }{S_j }-\left( {1-S_j } \right) \end{aligned}$$

(11)

$$\begin{aligned} \varepsilon _{jk} =\frac{\delta _{jk} }{S_j }+S_k \quad j\ne k \end{aligned}$$

(12)

These elasticities are not constant, but differ at every data point. It is common practice to compute them either for the mean or the first, central or last year of the sample. Estimates of these elasticities should use fitted rather than observed cost shares.³⁶