Wage setting at the firm level - insider versus outsider forces.

1. Introduction

IN TWO polar cases sector- or firm-specific performance does not matter for wage setting at the local level: the competitive labour market and the completely centralised labour market. In a competitive labour market, wages will be equalised, except for compensating wage differentials due to differences in workers' skills or in working conditions. In the completely centralised labour market, on the other hand, wages are fully determined by nationwide bargaining, and wage differentials reflect the preferences and bargaining power of central labour market organisations. Thus, in neither of these cases, is firm performance likely to influence the wage outcome.

For Sweden, often considered to possess the most highly centralised labour market institutions in the OECD, studies by Edin and Zetterberg (1992) and Holmlund and Zetterberg (1991) indicate that workers' sectoral affiliation is an unimportant determinant of wages when controlling for other factors such as individual characteristics or outside wages. Holmlund and Zetterberg conclude that 'centralisation in the Nordic countries does not seem to be just an institutional facade'.

In this paper, I address the issue of the influence of firm-level performance on wages using a data base with information on individual firms in Swedish manufacturing. More specifically, I investigate the strength of insider effects in wage setting employing the bargaining framework of Nickell and Wadhwani (1990). A bargaining model seems the obvious choice, since Swedish manufacturing is characterised by further firm-level bargaining, a situation where non-competitve wage differentials may arise naturally.

Besides facilitating an assessment of the link between firm performance and wage setting, this theoretical framework is also well suited to testing an implication of recent 'insider--outsider' theories of unemployment. In these theories, unions primarily act in the interest of currently employed insiders. Subsequent to an adverse shock to labour demand that reduces employment, the remaining insiders can push up wages, as the risk of losing the job for an insider at the going wage rate has fallen. The teories therefore predict an inverse relation between the number of insiders and the wage rate. This prediction will be evaluated in the present study.

A third objective of the paper is to throw some light on the controversial issue of the effects of public labour market programmes on wage setting. If labour market programmes are regarded by those concerned as close substitutes to regular employment, the incentives for wage restraint by a union are severely weakened by the practice of keeping laid-off workers out of open unemployment by means of e.g. relief jobs (Calmfors and Forslund 1990, 1991; Calmfors and Nymoen 1990). The strength of this effect is a crucial question in an evaluation of the effects of labour market programmes on unemployment. If the wage-raising effect is strong enough, labour market programmes may well crowd out so much regular employment that the net effect on unemployment becomes negligible. Thus, the question of the strength of this effect is addressed in the empirical analysis below.

To anticipate, my empirical results basically confirm the conclusion reached in previous Swedish studies that local performance has a minor influence on wages and that insider--outsider effects are absent. I do, however, not find any support for the hypothesis that aggregate labour market policies exert an upward pressure on firm wages. Finally, I compare standard OLS fixed effects estimates to estimates using a recently developed IV estimator due to Arellano and Bond (1991). The finding in this respect is that the empirical results are indeed sensitive to the estimation method chosen.

The rest of the paper is organised as follows: In Section 2, the data set is briefly discussed. The theoretical background is presented in Section 3. In Section 4, the empirical specification is derived, and in the concluding Section 5 the estimations and the empirical results are presented and compared to the results in earlier studies.

2. The data

The data used in this study derive mainly from an annual survey undertaken by the Federation of Swedish Industries, in which a sample of member firms respond to a questionnaire containing some 50 questions. The questions cover standard variables such as employment, hours worked, sales, purchases of intermediate goods, and raw materials, wage bills, and investment in buildings and machinery. A special feature of the questionnaire is that the firms are asked both about realised and planned/expected values of these standard variables.

The present data base covers 434 firms during the years 1980 to 1989. In practice, however, many of the firms either change identity (through mergers, etc.) during the period, or simply do not respond all years. In addition, the figures for 1989 pertain to planned or expected magnitudes, so looking only at realised values, the maximum number of observations would be about 3,900. As the dependent variable in the regressions will be a measure of firm wage rates, all observaions with missing values for that variable have had to be dropped. This leaves about 2,000 observations. Then, depending on the exact combination of right-hand-side variables required, additional observations must be discarded due to missing values. This typically reduces the number of observations by about 100. The data set used in the regressions is finally reduced by the imposition of a requirement of at least four consecutive years of observation for a firm that has not changed identity, an event recorded in the data set in the form of a dummy variable. The final data set consists of 153 firms and 842 observations (regressions with actual productivity as regressor) or 128 firms and 712 observations (expected productivity as regressor).

The sample of firms is unbalanced in the sense that the number of observations per firm ranges from four to seven (with a fairly uniform distribution of firms across the different numbers of observations). The representativeness of the sample is impossible to test, as, due to the requirement of …