Hourly wages and working time in the Dutch market sector 1962–1995

Abstract

This article analyses the joint behaviour of hourly wages and standard hours in the Netherlands. With respect to the development of full-time hours to different hypotheses are suggested: work-sharing or productivity-sharing. Under the work-sharing hypothesis, high unemployment would lead to reduced hours, whereas under productivity-sharing, increased productivity leads to higher wages or reduced hours. The evidence is in favour of the productivity hypothesis. There is no direct impact of unemployment on the evolution of hours. Moreover, although reduced hours tend to increase hourly wages in the short run, this is not the case in the long run.

Acknowledgement

I am grateful to Joan Muysken and Bertrand Candelon for useful comments on earlier versions of this article.

Notes

¹Sometimes a distinction is made between actual and standard hours. However, the empirical evidence surveyed in Kapteyn et al . (2004) suggests that actual working hours move (almost) proportionally with standard hours.

²Over the last decades part-time employment increased strongly, especially among female employees. The focus here is on full-time hours, however. For more extensive descriptions of the recent history of the Dutch labour market, see e.g. Broersma et al . (2000) or Hartog (1999). An account of the development of Dutch working time is given in De Neubourg (1991) and Van Doorne-Huiskes and De Lange (1994).

³Except for the unemployment rate u, small letters denote natural logarithms of the corresponding variables.

⁴Similar results were obtained from a reduced form specification, omitting the contemporaneous exogenous variables as explanatory variables under the general specification.

⁵To do so, the model should be augmented with a labour demand or price setting equation. This is beyond the scope of thisarticle. Jacobson and Ohlsson (2000) and Kapteyn et al . (2004) include an employment equation in their analysis.

⁶It sometimes is suggested that there is a difference between employer contributions and employee taxes, see Muysken et al . (1999). This difference is ignored in the presentarticle.

⁷The only series that seems to be stationary is the difference between the replacement rate ρ and the labour share φ_N . As both series individually are I(1), this suggests that both variables are cointegrated. The economic logic might be that higher benefits lead to higher taxes, leading in turn to an increase in the labour share, but this is speculative. Alternatively, a high labour share might increase the demands for relatively high benefits.

⁸There is some evidence in the literature that nominal wages and prices could be I(2), but relative prices, and thus real wages, are typically found to be I(1), see e.g. Banerjee et al . (2001).

⁹With annual data two lags appear to be reasonable, but because of a lack of degrees of freedom I did not test for the number of significant lags. But, as the URF of the conditional model does not seem to be misspecified, see Table 5 in the next section, the number of lags included appear to be appropriate.

¹⁰See Urbain (1995) for a discussion of the relative merits and pitfalls of modelling cointegrated systems in conditional or in full system models.

¹¹Moreover, the negative correlation between the residuals of the two equations of the model almost vanishes when w_c is replaced by w_c  + h.

¹²The marginal system is modelled as an unrestricted VAR of the four conditioning variables with 2 lags, a constant and a trend.

¹³Boswijk (1994) calls WNC an instability test, but to avoid confusion I prefer to call it a no-cointegration test. The -statistic was used to account for the trend in the model. With four exogenous variables, the 10, 5 and 1% critical values are 20.76, 23.33 and 28.51, respectively.

¹⁴A more or less similar ambiguity of the no-cointegration test statistics can be found in the URF of Table 5.

¹⁵When the replacement rate instead of benefit level is included, as in Broer et al . (2000), wages are proportional to productivity.

¹⁶As the system of six variables contains three cointegrating relations, the two cointegrating relations of the constrained structural model are not uniquely identified.

¹⁷The p-value of the likelihood-ratio test of the restriction that both coefficients in the cointegrating wage vector sum to one is 0.896.

¹⁸To test the possible U-shape between hourly wages and hours worked, lagged hours squared was added as an additional explanatory variable. This did not improve the fit, however.

¹⁹Moreover, when included in z_h , unemployment has the wrong sign.

²⁰The p-value for the imposed restriction that z_{ht − 1} does not affect the wage equation is 0.114.

²¹The alternative hypothesis that the coefficient of Δh equals 0, instead of −1, is strongly rejected, with a p-value of 0.0003 against a free estimate of that coefficient.

²²Alternatively, the growth rate of hourly productivity could have been included, instead of the growth rate of annual productivity. This gives rather similar results, the log-likelihood being slightly higher (363.848). On the other hand, the correlation between the residuals of equations increases to 0.347 in this alternative specification.

²³The characteristic roots of the parsimonious system are 0.682, −0.427, 0.271 and 0. The system is thus stable and oscillates.

²⁴The p-value for the imposed restriction on Δs_t is 0.654. Accordingly, the dynamic adjustment can indeed be specified in terms of the growth of net real wages.