Outsourcing, labour productivity and wage inequality in the US: a primal approach

Abstract

We investigate the linkages among outsourcing activities, labour productivity and wage inequality for skilled and unskilled labour by employing a primal approach that involves estimating a nested constant elasticity of substitution production function, using six-digit North American Industry Classification System US manufacturing industries from 2002 to 2005. First, we find that general outsourcing and international outsourcing have a skill-biased impact on labour productivity. However, the skill-biased impact of general outsourcing on labour productivity is larger than that of international outsourcing. Second, we find that the wage gap between skilled and unskilled labour, which is defined as their marginal productivity gap, can be better explained by general outsourcing than by international outsourcing. These two results imply that the wage inequality of US manufacturing industries during 2002–2005 was mainly due to the skill-biased labour productivity effect of general outsourcing rather than that of international outsourcing.

Notes

¹ In this view, outsourcing may also be termed as a ‘make-or-buy’ decision (Grossman and Helpman, 2002), ‘vertical disintegration’ (Holmes, 1999), ‘fragmentation’ (Arndt and Kierzkowski, 2001), ‘vertical specialization’ (Hummels et al., 2001).

² Girma and Görg (2004) argue that since the subsequent productivity effects are of their interests, it should not matter whether outsourcing takes place internationally and domestically.

³ For a theoretical treatment of international outsourcing, see Feenstra and Hanson (1996), Deardorff (2001), Jones (2000) and Jones and Kierzkowski (2001).

⁴ In an approach similar to ours, Egger and Egger (2006) construct an equation of the unskilled labour average productivity, using the CES production function and estimate it to see the impact of international outsourcing on unskilled labour productivity in the case of the EU.

⁵ For example, Díaz-Mora (2007) shows that outsourcing decisions are positively related to unit labour costs, skill requirements, national ownership and orientation to international markets.

⁶ As elaborated in the next section, neither of the literature strands distinguish between the skilled and unskilled labour productivity impacts of general and international outsourcing.

⁷ See Olsen (2006) for complete survey on impacts of outsourcing on productivities.

⁸ In contrast with Feenstra and Hanson (1996), Anderton and Brenton (1999) do distinguish between international outsourcing in low- and high-wage countries. The idea is that low-skilled activities are typically outsourced to low-wage countries.

⁹ They estimated a derived average labour productivity equation, but we estimate the production function itself.

¹⁰ For the sake of computational simplicity, we implicitly assume that the contributions of each factor of production to value added are equally weighted. Though this assumption is rather strong, allowing for different weights for value-added contributions does not change our main results qualitatively.

¹¹ Since we centre on the impacts of outsourcing on labour productivity, we need to assume that the effects of other factors on technology level, such as innovation and product development, is comprehensively taken into account by A_i.

¹² In contrast with existing studies on the impact of international outsourcing, such as that of Feenstra and Hanson (1996), we assume that substitutions between skilled and unskilled workers, skilled workers and capital, and unskilled workers and capital, are equal. Nevertheless, we also tried the case where unskilled and skilled workers are perfect substitutes. We find that our results are qualitatively unchanged.

¹³ To us, the marginal value added of workers may better reflect their productivity and thus be economically more appealing, compared with value added per worker, in that the impacts of outsourcing on skilled and unskilled labour are allowed to differently affect their productivities. Moreover, by looking at marginal impacts, we are able to capture some links between productivity and relative demand for labour.

¹⁴ See Olsen (2006) for a survey of literature using Cobb–Douglas production function for empirical analysis of the relationship between outsourcing and productivity.

¹⁵ A number of studies have examined the roles of international outsourcing on explaining the evidence of rising relative skilled wage during 1980s in most Organization for Economic Co-operation and Development (OECD) and newly industrialized economies, such as Feenstra and Hanson (1996, 1999) (US), Feenstra and Hanson (1997) (Mexico), Anderton and Brenton (1999) (UK), Geishecker (2002) (Germany), Hsieh and Woo (2005) (Hong Kong) and so forth.

¹⁶ Elaborated in Section II, these empirical findings are confirmed by a number of studies in various economies. Nevertheless, the results make use of the dual approach in the sense that a relative increase in relative demand for skilled workers is derived from either cost or profit functions. In contrast to these studies, our methodology is to directly estimate production functions to see whether the same results are confirmed.

¹⁷ The stochastic error term, ε_it, can be interpreted as neutral technological shocks.

¹⁸ The two-step NLS estimator, as first shown by Murphy and Topel (1985), has an important and desirable asymptotic property. That is, under the standard conditions assumed for the NLS estimators, the second-step estimators are consistent and asymptotically normally distributed with an asymptotic covariance matrix.

¹⁹ Robust SEs are in parentheses. The * represents statistical significance at 1%.

²⁰ The Hansen–Sargan test is calculated by nR², where n is the number of observations and R² is obtained from a regression of residuals from the second-stage regression on the instrumental variables. The Hansen–Sargan statistic is χ² distributed with the degree of freedom equal to the degree of overidentification.

²¹ With these starting values, the exceptional convergence property is obtained. Still, the results are robust to a variation of starting values.

²² The elasticities of substitution are calculated from the results of two-step IV estimations in Models 1 and 2. If the results from Fixed Effects (FE) estimations are employed, they will be equal to 1.094 and 1.12 for general and international outsourcing, respectively.

²³ The well-behaved production function requires that the parameter ρ is less than unity.

²⁴ Since it can be shown that the elasticities of substitution under the Cobb–Douglas value-added function must be equal to unity, in doing so the above mentioned null hypothesis is equivalent to specifying a negligible value of ρ(=0.0001).

²⁵ Nevertheless, for FE estimation, the results are in favour of decreasing RTS and the hypothesis that technology is characterized by Constant Returns to Scale (CRTS) (LR Test 2) is rejected with a 90% level of confidence.

²⁶ One may observe that the parameter estimates from the IV model are much larger than those of the FE model. This is to compensate the negative parameters in the IV model, which are also much larger than those in the FE model. In this sense, our further analyses regarding the productivity and wage inequality impacts of general and international outsourcing are qualitatively unchanged and therefore robust with respect to the choices of econometric estimations.

²⁷ The notion of outsourcing referring to imported intermediate inputs is first explored by Feenstra and Hanson (1996). In contrast, the aggregated definition including both domestic and international outsourcing is according to Abraham and Taylor (1996).

²⁸ In the first-step regression, all the abovementioned instruments are statistically significant at a 95% level of confidence.

²⁹ LR statistics (LR Test 2), which are distributed as a χ² distribution with 1 degree of freedom, are statistically insignificant across all specifications.

³⁰ Parameter estimates under NLS are statistically significant at minimum of a 90% level of confidence. Under two-step IV estimation, though parameter estimates in Model 3 of general outsourcing are statistically significant at a 95% level of confidence, except for β_L, they seem statistically insignificant in Model 4 of international outsourcing.

³¹ We choose to report results corresponding to two-step IV estimation as it takes into account potential endogeneity problem and therefore may convey more consistent parameter estimates. Nonetheless, the main implications do not change when the results of NLS are calculated.

³² Log-linearized specification of empirical models derived from Cobb–Douglas technology is widely used by a number of studies (see, e.g. Amiti and Wei, 2006).

³³ As shown in Table 4, LR Test 1 rejects the null hypothesis that β_H = β_L with an at least 95% level of confidence except for the NLS result in Model 4.

³⁴ Therefore, our results are consistent with the well-established results that an increasing relative wage of skilled workers within industries can be explained by the notion of outsourcing. The long-run interpretations have been explored by Feenstra and Hanson (1996, 1999) and Hsieh and Woo (2005) and the short-run results are confirmed by Anderton and Brenton (1999), Geishecker (2002) and Amiti and Wei (2006).

³⁵ The natural interpretation of the elasticities of marginal value added of unskilled and skilled workers with respect to outsourcing indexes evaluated at mean variable values is the marginal effects of outsourcing on a representative firm.