# «Der Open-Access-Publikationsserver der ZBW – Leibniz-Informationszentrum Wirtschaft The Open Access Publication Server of the ZBW – Leibniz ...»

In recent work, globalisation took center stage as the most likely candidate among many explanatory factors in the analysis of declining labour shares all over continental Europe and beyond. Numerous studies have analysed the eﬀects of greater trade integration on factor shares. In general, it is argued that greater trade openness exerts downward pressure on the labour share either through Stolper-Samuelson-type relative factor price eﬀects or via power-shifts in the wage bargaining process. In the latter case it is assumed that in more open economies the ﬁrm’s outside option improves relative to that of employees if costs of relocating production or sourcing goods from spatially differentiated locations are less than prohibitively high. Furthermore, stiﬀer international competition can decrease mark-ups of ﬁrms, raising labour shares. An early study by Harrison (2002) shows negative eﬀects of increasing trade openness and occurences of exchange rate crises for a large sample of developing and developed countries. Jaumotte and Tytell (2007) add further measures of globalisation including immigration and also establish a negative link. However, the eﬀect of increased trade openness also features a certain degree of heterogeneity with respect to time, regional aspects and measurement.

Guscina (2006) stresses that the eﬀect is signiﬁcant in the most recent two and a half decades only; Buch et al. (2008) show that Italian regions are diﬀerently aﬀected than German ones; and Hutchinson and Persyn (2009) develop more sophisticated measures of trade openness that render the eﬀect insigniﬁcant.

Besides globalisation, labour market institutions are frequently brought up as explanatory factors in the quest for uncovering the mechanisms of factor share dynamics.

Checchi and Garcia-Peñalosa (2008) point to potentially heterogeneous eﬀects of institutions on the labour share and stress the importance of empirically determining the direction of the overall eﬀect. Blanchard and Giavazzi (2003) emphasize the intertemporal aspect of such institutions. They propose that in the short run institutions which increase the bargaining power of workers could lift the labour share, while the same institutions could set the labour share oﬀ on a declining track when employers substitute capital for relatively more expensive labour in the long-run. Most studies with a focus on globalisation also integrate into the analysis aspects of labour market institutions.

**We contribute to the inspiring research outlined above and by stressing two points:**

dynamics and cross-country heterogeneity in estimated slope coeﬃcients. The ﬁrst point has so far mostly been addressed in a rather ad-hoc manner by simply including the dependent variable as a one period-lag. Yet, as Pesaran and Smith (1995) have shown, there is considerable danger in blind trust in pooled dynamic models. In such models, severely biased estimates could be the results of data best described by heterogeneous slope-coeﬃcients across sample units, i.e. if the eﬀects of certain variables diﬀer across countries in our case. Therefore we apply estimators which allow us to directly test the homogeneity assumption of all slope-coeﬃcients inherent in most previous studies.

Assessing heterogeneity furthermore enables us to retrieve country speciﬁc insights into the driving forces of movements in the labour share.

The idea of possibly heterogeneous slope coeﬃcients is clearly valid; each of the most prominent explanatory variables in studies on the labour share gives at least some reason to doubt a uniform impact mechanism across countries. As stated above, this also implies worries about potentially biased results in dynamic estimations.

The impact of the capital-output ratio has been shown by Bentolila and Saint-Paul (2003) to be sector-dependent - crucially inﬂuenced by the sector’s elasticity of substitution between production factors. This in turn implies that diﬀerent sectoral compositions of the economies in the sample could potentially introduce heterogeneity across countries as well. However, the distribution of value added and employment across sectors is fairly similar for the countries in our sample. This might limit the scope for heterogeneous coeﬃcients in this case.

The impact of Total Factor Productivity (T F P ) developments across countries may also diﬀer. This variable is mostly included in order to capture the nature of technological change. This makes T F P a more or less suitable variable on a country-by-country basis given the true nature of technological change may be diﬀerent across countries.

Reason to doubt the cross-country homogeneity of the inﬂuence globalisation exerts on the labour share particularly comes from the complex interaction of trade openness and the production and employment structure in the respective countries. In addition, if one assumes that increased openness puts labour at a general disadvantage in the wage bargaining process, the country-speciﬁc institutional arrangements matter as well. Note that for all these cases, should heterogeneity be indeed important, ﬁxed eﬀect methods provide insuﬃcient controls, since they merely account for the time-constant elements of country speciﬁc characteristics and capture heterogeneity through diﬀering intercepts only.

However, it is not clear whether and to what degree this heterogeneity is indeed important. It might not bias the results after all. For now, we merely state the possibility and take it seriously in the estimation below. This is, we rely on technical methods to check the validity of the pooling assumption implied in most econometric treatments the literature oﬀers. We test a basic model of the labour share consisting of the main explanatory variables that surfaced in the literature. Yet, we do not restrict the inﬂuence of those factors to be homogeneous across countries. We estimate the driving forces of labour share ﬂuctuations in a dynamic heterogeneous panel framework. Particularly, we employ the pooled mean group (PMG) estimator and the mean group (MG) estimator as in Pesaran et al. (1999) and Pesaran and Smith (1995), respectively. The PMG estimator represents a dynamic pooled model with a homogeneity restriction on all long-run coeﬃcients, which are in the focus of our analysis. The MG estimator explicitly allows for slope-heterogeneity in those long-run coeﬃcients in contrast to mere intercept or short-run heterogeneity. Therefore, it provides the basis for direct tests on the validity of the pooling assumption which we carefully discuss. More generally, our estimates serve as a robustness check of the results previously brought forward in the literature, since we also employ standard ﬁxed-eﬀects estimators and compare the results to our preferred speciﬁcations.

Note that our aim is not to participate in the quest for further possible explanatory variables, but to take a structured approach at assessing the relevance and importance of those factors that can be seen as fairly established in the literature. We believe that a thorough treatment of dynamics and possible country speciﬁc impacts could shed further light on the development of the labour share.

This paper is organised as follows: The section following this introduction brieﬂy outlines the theoretical framework and clariﬁes the predicted impacts of our explanatory variables. In section 3 the theoretical model is transformed into an estimation setup and the empirical strategy is explained. The estimators are introduced and their suitability and particular use are carefully discussed. Section 4 reports sources and computations of the data while section 5 presents the results of our econometric exercises. A ﬁnal section concludes.

**2 Theoretical background**

The goal of this section is to motivate, in a way consistent with theory, the explanatory variables that are assumed to aﬀect the labour share (LS). We mostly build on Bentolila and Saint-Paul (2003). They show that movements in the labour share can in general be explained in terms of three diﬀerent channels. First they show the capital output ratio k = K/Y to, under certain assumptions, comprehensively explain movements of the labour share triggered by eﬀects such as changes in wages or factor shares in production.

Secondly, they show that certain departures from the original assumptions can shift this relationship. Thirdly, they provide guiding theory for cases in which the economy is put oﬀ the schedule deﬁned by the relationship between k and LS. We follow their theoretical insights and brieﬂy introduce each case.

The capital output ratio as a simple but comprehensive determinant of ﬂuctuations of the labour share emerges irrespective of a strict functional form. As long as ﬁrms produce under constant returns to scale, labour and capital are the sole inputs, labour markets are perfectly competitive and technological progress is not capital augmenting, the labour share can be expressed as a function of k, LS = g(k). This encompasses all changes in wages, interest rates, factor inputs or labour augmenting technological change as long as the above assumptions are maintained. The direction of the eﬀects on the labour share then depends on the elasticity of substitution. It can be shown, that a higher k only lowers the labour share if the factors are substitutes, i.e. δLS/δk 0 only if the elasticity of substitution is greater than one.

If the assumption on the nature of technological progress is lifted and capital augmenting technological change is allowed for, changes in k are no longer a suﬃcient explanation for labour share movements. Bentolila and Saint-Paul (2003) show that capital augmenting technological change shifts the curve described by g(k) in a multiplicative way. This means that the original relationship is preserved and a change in factor prices or inputs moves the labour share following the same mechanism as above, but it does so at a diﬀerent level of LS which is determined by the size of the factor bias inherent in capital augmenting technological change. At this point it is enough to note that now LS = g(k, A) with A representing capital augmenting technological change.

A second possibility for deviations from the original, purely k-based, relationship are non-competitive features in the product or labour market. If factors are not paid their marginal product, the economy moves oﬀ the schedule derived under the strict set of assumptions above. Consider for example a situation in which bargaining takes place over wages and assume that the process can be modeled in an eﬃcient bargaining context.

Then, the labour share is aﬀected by the relative bargaining power of employers and employees. Following the literature, we consider trade openness an important indicator of relative bargaining power. If trade openness is a valid approximation for an economy’s integration into world markets and its cost of access to the latter, the value of the outside option of ﬁrms in the bargaining process increases with openness. Thus, the labour share is negatively aﬀected. It is interesting to note, that trade openness can aﬀect the labour share in numerous ways. If trade triggers Stolper-Samuelson-type eﬀects, those should be captured by the g(k) schedule, since they imply simple changes in factor inputs and prices. Trade openness could also act as competition enhancing, driving down mark-ups of ﬁrms via reducing their market power. For now, we consider the impact and sign of the coeﬃcient of trade openness an empirical issue and postpone further details to later sections. At this point we simply state a general relationship for the labour share as LS = g(k, A)h(X) with X standing for all possible "shift factors" driving a wedge between the marginal product of labour and the real wage. We assume h(.) to have an exponential form.

In the estimations detailed in section 5, we allow for all the above cases by including the variables most commonly used in the literature. We directly control for the capital output ratio and allow for the possibility of capital augmenting technological change by including an index of Total Factor Productivity. An important test will be to compare the signs of the estimated coeﬃcients on k and T F P. Only with the coeﬃcients for k and T F P being equally signed one can infer that technological change is indeed capital augmenting. When assessing the impact of k, T F P and trade openness on the labour share, we also allow for diﬀerent institutional arrangements across countries.