Foreign ownership and its effects on employment and wages: the case of Sweden

In a recent study, Dumont et al. (2012) show that the bargaining position of low-skilled labour appears to fall with increases in imports and offshoring, while the bargaining power of high-skilled workers is affected positively by research and development activities. They found that trade unions seemed to focus on wages rather than employment. Moreover, Modén (1998) revealed that foreign ownership led to an increase in labour productivity in Swedish manufacturing, implying a reduction in employment (at least in the short run); however, Bandick and Karpaty (2011) found no negative impact of foreign ownership on employment. Navaretti et al. (2003) found that, for any given wage increase, multinational firms reduced total employment less than national firms did. These authors (ibid.) also established that labour market regulations were largely irrelevant to the labour market behaviour exhibited by multinational firms and concluded that concerns about labour market regulation scaring off foreign direct investment (FDI) seemed unfounded.

There are by now several theoretical studies trying to explain the effect of the flow of FDI on wages and employment. Most of these studies employ a Nash bargaining approach to model the strategic bargaining relationship between a union and a firm. Bughin and Vannini (1995), using such a model, established that trade unions were indifferent when it came to choosing between exporting or investing abroad. Naylor and Santoni (2003) argued that FDI in a potential host country was more likely if the union wage-bargaining power in that country was weaker and if there was a lower degree of substitutability between potential rivals’ products. Zhao (1995) and Zhao (1998) found that, with symmetric FDI flows in both directions, the negotiated wage decreased for every level of employment, while each firm’s profit increased; such FDI flows increased employment and national welfare if the union was wage-oriented but decreased employment and national welfare if the union was employment-oriented.

A number of empirical studies have examined the effects of foreign ownership on wages. A frequent finding is that foreign-owned firms are more productive than domestically owned firms, although this finding tends to be sensitive to the exact specification used. This higher productivity in turn tends to translate into higher wages, although this effect frequently increases the equilibrium wage in the entire sector and thus spills over into domestically owned firms in the same sector as well (see e.g. Lipsey 1994; Feliciano and Lipsey 1999). Concerning ownership per se, Hatzius (2000) presented evidence to the effect that the bargaining power of employees was reduced when domestic (British) firms were acquired by foreign owners. Aitken et al. (1996) investigated the wages paid by domestic and foreign-owned firms in Mexico, the USA, and Venezuela. They presented evidence that foreign investment in a sector led to higher productivity and higher wages, but that (at least for the USA) wages also increased in domestically owned firms in the same sector. Girma et al. (2001) found, for the UK, that a large share of the difference in productivity was linked to larger size and larger capital intensity of foreign-owned firms. They found little indication of wage spillovers from foreign-owned firms to domestically owned firms, although they noted that the econometric specification that they used may have masked some of the spillover effects. Conyon et al. (2002), using a short panel of UK firms that were acquired either by foreign or by new domestic owners, found that wages increased by 3.4 % when foreign investors acquired the companies compared to when the companies were acquired by new domestic owners; they attributed the entire wage difference to productivity differences. Griffith and Simpson (2003) also established a positive premium for foreign-owned companies in the UK but attributed the entire wage and productivity difference to differences in investment levels. For Portugal, Almeida (2007) presented evidence that the workforce in foreign-owned companies was more highly skilled and was paid a higher wage premium than in domestically owned companies but also found that these differences were entirely attributable to cherry picking of acquisition objects by foreign investors. Also studying Portugal, Martins (2004) revealed a positive average-wage premium for the years 1991–1999 but found successively smaller premia when controls were added for firm type and worker type. Using Finnish data, Huttunen (2007) established that, when Finnish companies passed into foreign ownership, the wage dispersion within the companies increased. However, in a related study of Brazil, Martins and Esteves (2008) found no evidence that foreign acquisitions resulted in higher wage premiums; instead, such takeovers tended to decrease the demand for labour.

To sum up, one could say that, in general, foreign ownership tends to increase the wage rate, largely because of improved labour productivity, and that at least some studies have found that these wage increases spill over into other firms in the same sector. Nonetheless, it needs to be borne in mind that most of the previous studies have looked at labour markets considerably less unionised than Sweden’s. It also needs to be borne in mind that the type of firm acquired by foreign investors may be different from an average firm, in the country and even in the same sector.

We use a bargaining model first developed by Manning (1987). We derive the structural equations for determining employment and wages on the assumption that both are sequentially negotiated between an employer and a trade union and that these parties are completely informed. The resulting model will be used as the basis for an empirical analysis in Section 4. The basic idea here is that the employer seeks to maximise profit, given a production technology, and that the union seeks to maximise its members’ utility. How well each party succeeds depends on its bargaining strength. Bargaining strength may, in turn, depend on a number of different things, one of which is the institutional setting. In this paper, we ultimately set out to test whether foreign ownership affects employment and wage-setting, and our hypothesis is that if there is an effect, this effect may be transferred through the trade union’s bargaining power.

In most collective bargaining, there is a two-stage negotiation process where wages are determined prior to employment. In Sweden, such employment contracts often last for a year or two. Partly because of this, trade unions may act differently—and may have a different bargaining power—when negotiating about wages, compared to when they are negotiating about employment. Manning (1987) offers a theoretical explanation for this phenomenon. Firstly, owing to uncertainty and the impossibility of concluding complete contracts that will cover all the possible scenarios and supplements, it may be best to reach agreement on wages first and then determine employment. Secondly, wage negotiations are mostly carried out at a higher level, e.g. nationally, while employment levels are usually negotiated at plant level.

where L is the share of the labour force that is employed and (1 − L) the unemployed share.

where p is the price of output.

where Φ ₂ ∈ [0, 1] is the union’s bargaining power over employment. If Φ ₂ = 0, the firm sets the employment level. U ₀ and Π ₀ are the fallback utility and fallback profit, assumed to be given by U ₀ = u(R) and Π ₀ = 0, respectively.

From Eq. (5), we see that the employment level is chosen such that a linear combination of the value of the average productivity of labour and the value of the marginal productivity of labour is equal to the wage. If Φ ₂ = 1, the firm has no influence on determining employment, and the wage will be set equal to the value of the average product. If Φ ₂ = 0, then employment is chosen such that the marginal product of labour equals the wage.

where Φ ₁ ∈ [0, 1] is the trade union’s bargaining power in the wage-setting stage. If Φ ₁ = 1, the trade union sets the wage, which in combination with Φ ₂ = 0 gives us the monopoly model (see e.g. Dunlop 1944).¹ If Φ ₁ and Φ ₂ are equal, we have the efficient bargaining model of McDonald and Solow (1981); Manning (1987) notes that in practice, Φ ₁ > Φ ₂ in many real-world unionised labour markets and that different bargaining strengths on different issues are a key source of economic inefficiencies associated with unionisation.

Perhaps surprisingly, we cannot say how the wage is affected by a change in bargaining power over employment without having a more explicit model specification that lets us determine the wage-setting rule (e.g. if the production function is of a Cobb-Douglas type, it can be shown that ∂w/∂Φ ₂ = 0).

Equations (8) and (9) constitute a system of equations describing equilibrium employment and equilibrium wage determination that can, in principle, be estimated. A problem, however, is that the bargaining parameters, Φ ₁ and Φ ₂, are not directly observable. To get around this problem, we will follow the empirical approach developed by Alogoskoufis and Manning (1991), Doiron (1992) and Vannetelbosch (1996), in which it is assumed that bargaining power can be expressed as a function of a vector of exogenous variables, z.

In particular, we are interested in investigating how differences in ownership affect bargaining power and, ultimately, employment and wage-setting. If the belief that foreign ownership reduces unions’ bargaining power is correct, we would expect the bargaining parameters to be lower for foreign-owned firms. However, even if there is an impact on unions’ bargaining power, it is not clear which of the two bargaining parameters will be affected the most. Thus, the actual impact of foreign ownership on the two bargaining parameters is a matter for empirical analysis.

The data we used was a panel data set for manufacturing firms in Sweden compiled by one of the authors. The panel covers the years 1980 to 2005 and consists of all Swedish manufacturing firms with at least 50 employees. A classification of the industries in question is given in Table 3 in the Appendix. Since the classification of industries changed during the period studied, only firms that belonged to the same industry in both classification systems (SNI69 and SNI92) were included, and in order to permit analysis of effects over time, only firms with at least five consecutive observations were retained in the dataset. In Sweden, 242 firms met all these requirements and were thus included in the dataset. To be classified as foreign-owned, more than 50 % of the votes in the company had to be held by foreigners.

The variables used in the study are presented in Table 4 of the Appendix, with their definitions. The data for wages, employment and capital for each firm were collected from each respective firm’s annual report, which by law have to be submitted to the Swedish Companies Registration Office. The number of employees in the data set covers both white- and blue-collar workers, while the trade union data (discussed below) cover blue-collar workers (see e.g. Huttunen 2007). This could give us a bias in the measure of wages, since salaries for white-collar workers are significantly higher than those for their blue-collar counterparts (see e.g. Huttunen 2007). However, in our case, this is a minor problem since all the firms in the data set are quite large plants and blue-collar workers dominate their labour forces. The variables that are assumed to affect the bargaining power through the Φ ₁ and Φ ₂ functions are, in addition to country-specific ownership, trade union density in the industry and openness; these reveal how open the respective industries are to international trade. To calculate the measure of openness for each industry, defined as (export + import)/total sales (see e.g. Leamer 1987), data on exports, imports and total sales in each industry were collected from Statistics Sweden’s industry statistics. The average unemployment level for each trade union’s membership was obtained from Swedish Public Employment Service statistics. Data on union density were collected from the respective trade union’s annual reports, and the matching between trade unions and companies was based on trade union representation in companies, as reported in such unions’ annual reports.

From Fig. 2, we can see that the share of foreign-owned firms in the sample increased sharply—from a level of about 5 % in 1980 to approximately 13 % in 2005. It is also clear from Fig. 1 that most foreign owners of the sample companies originated from European countries.

In addition to the country-specific ownership of the firm, the bargaining power Φ ₁(z ₁) in the wage equation is assumed to be a function of the industries’ union density, the industries’ openness to international trade and the fraction of trade unionists living on unemployment benefits. The bargaining power Φ ₂(z ₂) in the employment equation includes the same set of variables as Φ ₁(z ₁), with the exception of the fraction of trade unionists living on unemployment benefits. We have chosen to exclude this variable due to the short-run dependence between the number of workers in a firm and unemployment in the industry.

The lowest mean wage per employee is consistently found among workers in the pulp, paper and paperboard industry. The reservation wage at time t has, therefore, been specified as the lowest observed wage per employee among these workers at time t.

where μ _i is an unobservable individual-specific effect, v _it is the remainder disturbance in respective equation, x _1it  = R _it , x _2it  = [p _it , K _it ] and \( {\boldsymbol{\upgamma}}_2^{\hbox{'}}=\left[{\gamma}_p^{\hbox{'}},{\gamma}_K^{\hbox{'}}\right] \). The Φ functions are specified as logistic distribution functions, Φ ₁(z ₁) = 1/[1 + exp(−z ₁ ρ ₁)] and Φ ₂(z ₂) = 1/[1 + exp(−z ₂ ρ ₂)], so that Φ ₁ ∈ [0, 1] and Φ ₂ ∈ [0, 1].

Before estimation, the original data are transformed into deviations from their time mean, such that the typical element of the wage series, for example, becomes \( {w}_{it}^{\circ }={w}_{it}-{\overline{w}}_i \), with \( {\overline{w}}_i={T}^{-1}{\displaystyle {\sum}_{s=1}^T{w}_{is}} \). Although this within-group transformation eliminates time-invariant regressors, a constant is still needed in each equation since E[Φ _k (z _k )] ≠ 0. All variables, except the ones included in z _k [k = 1, 2], were log-transformed (this is in line with previous empirical studies, see e.g. Alogoskoufis and Manning (1991) or Vannetelbosch (1996)). To both equations, we also added a linear time trend.

The within-group estimation of the non-linear simultaneous equation systems was carried out by the generalised method of moments (GMM), using \( {Z}^{\circ }=\left[{x}_1^{\circ },{\mathbf{x}}_2^{\circ },\mathbf{z}\right] \) as instruments.³ The variables that are used as instruments are the output price (producer price index), trade union density, openness, unemployment level, time and each dummy for owner’s domicile. A general description of simultaneous equation estimation is given in Hausman (1983), for example, while a description of non-linear GMM estimation of simultaneous equations based on panel data can be found in Arellano (2003), among others. To account for heteroscedasticity and autocorrelation, we use a Newey and West estimator of the covariance matrix. The covariance matrix was adjusted to account for the non-estimated fixed effects.