Inter-industry Wage Differentials and Allocative Inefficiency

ABSTRACT

Do inter-industry wage differentials reveal information on allocative inefficiency, implying that reallocation of labor input from low- to high-wage sectors improves aggregate productivity? This question is addressed by introducing wage-weighted averages of sector employment growth in growth regressions using panel data for selected OECD countries over the period 1971-93. For the 1970s, it is found that reallocation of labor between sectors with different wages did not contribute to explaining economic growth, whereas it did for the 1980s. Hence, observable wage differentials only reflect allocative inefficiency after 1980.

Keywords:

• Inter-industry wage differentials
• growth

Acknowledgements

The author wishes to thank George J. Borjas, Svend E. Hougaard Jensen, Tryggvi Thor Herbertsson, Sunwoong Kim, Hans Christian Kongsted, Morten I. Lau, Dani Rodrik, Jonathan Skinner, Torsten Sløk, Jan Rose Skaksen, Peter Birch Sørensen, two anonymous referees, and seminar participants at the Johns Hopkins University, University of Copenhagen, Copenhagen Business School, and CEBR for helpful discussions. The usual disclaimer applies

Notes

²Slaughter (2001) finds that the own-price elasticity of demand for US labor has increased in manufacturing and five out of eight industries within manufacturing between 1961 and 1991. International trade, however, cannot be linked to the increase.

³Another approach is to assume that the Hick's aggregation theorem applies, see Diewert (1978). In this application the theorem states that the price of gross output and the prices of intermediate inputs vary in strict proportions.

⁴ln the empirical section, I use an industry structure with broadly defined sectors, such as mining, manufacturing, construction, whole and retail trade, transportation, and finance. This implies that the degree of substitutability or complementarity among products across sectors is relatively low.

⁵ρ K/ucV and ω L/ucV are estimated because the prices that enter in the ratios, i.e. ρ, ω, and uc, are unobservable.

⁶The OECD STAN database uses a more comprehensive industrial classification, covers a longer time period (approximately 1970–99), and includes a larger group of countries than the OECD ISDB. However, I do not use this database because it is based on OECD ISDB for the 1970s for the majority of countries, implying that the number of industries is not increased by using this data source. Moreover, data for the 1970s do not exist for the extra group of countries. For these reasons and because the 1970s are of great importance to the hypothesis of the paper, I have not developed new measures based on the OECD STAN database. Another reason for not using OECD STAN is that it is based on a different industrial classification than OECD ISDB. The latter is consistent with the OECD Input–Output database that is used to construct an instrument variable. Input–output tables for the countries and years included in the analysis based on industrial classification used in OECD STAN do not exist.

⁷Belgium and the Netherlands are excluded from the analysis due to incomplete data.

⁸The empirical results presented in this section are also established when measures of labor input are based on persons engaged. In order to improve the clarity of the analysis, these results are not reported.

⁹The empirical results presented in this section are also established for the total economy and private sector. I do not present these results in the paper.

¹⁰The driving mechanisms behind convergence are of course important. Two potential candidates are decreasing returns to the set of reproducible factors of production, which implies high relative returns to investments for countries on low stages of development, and technological spillovers from leader countries. Mankiw et al. (1992) find support for the former explanation such that the accumulation of human and physical capital is important for (conditional) convergence. Benhabib & Spiegel (1994) on the other hand investigate diffusion of technology between countries and find that the stock of human capital of a country affects the speed of adoption of technology from abroad. Based on the empirical analysis of the present paper, however, it is not possible to distinguish the impact of the different mechanisms.

¹¹Islam (1995) also finds that the elasticity of output with respect to capital is much lower when fixed country effects are applied. However, the estimates are still significant for the group of countries in his study.

¹²In Table 4, the later sub-period is 1981–90. Extending this period to 1981–93 does not change the results.

¹³One concern related to the results presented in Table 4 is that the level of aggregation of sectors is too crude and should be based on a higher number of industries. To deal with this concern, I have constructed an alternative measure of Ĥ including manufacturing industries; not just the manufacturing sector. Instead of using 10 broadly defined sectors with manufacturing being one aggregate sector, the alternative measure uses between 9 and 13 manufacturing industries for each country, implying a total number of industries between 19 and 23. It is possible to construct the measure with 19 industries for 6 countries and 23 industries for 5 countries. Australia is excluded from the analysis owing to incomplete data. The main result presented in this paper is also valid using this alternative measure. In order to keep the exposition as clearly as possible, these results are not presented.

¹⁴An alternative group of countries with strong union environment contains Denmark, Finland, Germany, Italy, Norway, and Sweden. The presented results in Table 4, Regressions 5–8 are insensitive to changing country groups.

¹⁵The only country that has strong positive correlation between sector productivity growth and sector wages is Norway. This correlation is driven by the mining sector.

¹⁶p. 31 footnote 8.

¹⁷Part-time frequencies for women and men are observed for 1973, 1979, 1983, 1990, 1991, 1992, and 1993. For other years, part-time frequencies are determined by linear interpolation.