Skill-Biased Technological Change and the Business Cycle

Almut Balleer and Thijs van Rens



Over the past two decades, technological progress in the United States has been biased towards skilled labor. What does this imply for business cycles? We construct a quarterly skill premium from the CPS and use it to identify skill-biased technology shocks in a VAR with long-run restrictions. Hours fall in response to skill-biased technology shocks, indicating that at least part of the technology-induced fall in total hours is due to a compositional shift in labor demand. Investment-specific technology shocks reduce the skill premium, indicating that capital and skill are not complementary in aggregate production.

Published in the Review of Economics and Statistics, 95(4), pp.1222-1237.

May 2012 [download pdf] - Also available as Kiel Working Paper 1775.

Technical appendix

Previous versions: May 2011 (also available as CEPR Discussion Paper 8410) and June 2009 (titled “Cyclical Skill-Biased Technological Change”, also available as IZA Discussion Paper 4258). First draft: February 15, 2008


Codes and data

All codes and data in a single file (includes readme file) [download zip]

The data we used in this paper are available for download here. For the construction of the underlying micro-data from the CPS outgoing rotation groups, please see Haefke, Sonntag and van Rens.

·         Dataset (comma-separated format, readable in Excel or in Stata using the insheet command)

·         Stata do-files that were used to construct these data

The dataset contains quarterly time series (US, 1979-2006) for the skill premium, the relative employment of skill and the wage of low skilled workers. The variable naming convention is VAR_WW_SMPL_METHOD.

VAR is the name of the variable

·         sp = skill premium (log ratio of the wage of high and low skilled workers)

·         rele = relative employment of skill (as a fraction, ln(rele/(1-rele)) are the series used in the paper)

·         wL = log wage low skilled workers

·         Series named f_* and w_* contain fractions and average log wages for various subgroups of workers. The naming convention for these variables is different, please see the do-files for details.

WW denotes the sample weights and averaging procedure

·         ew = simple average, weighting only by CPS earnings weights (ORG weights)

·         hw = simple average, weighting by CPS earnings weights and hours worked

·         hwp = predicted values from a Mincer regression, weighted by CPS earnings weights and hours worked (baseline)

SMPL denotes the subsample

·         popu = the entire population (only available for some rele measures, which should be interpreted as the relative supply rather than the relative employment of skill)

·         labf = the entire labor force (only available for some rele measures, see above)

·         empl = all employed workers in the CPS

·         priv = employed workers in the private, non-farm business sectors in the CPS (baseline)

METHOD refers to the method used to calculate the skill premium and relative employment, see the paper and the do-files for more details

·         minc = coefficient on years of schooling in a Mincer regression of log wages on schooling, a quartic in potential labor market experience and some demographic controls, for rele these series contain the average years of schooling

·         hlcm = high school degree or less versus a college degree of more education

·         edu5 = using 5 education categories (less than high school, high school degree, some college, college degree, more than college)

·         Gxedu5 = using 2 gender and 5 education categories

·         Gxedu5xexp4 = using 2 gender, 5 education and 4 experience categories (baseline)


Thijs van Rens  |  IDEAS/RePEc  |  Google Scholar  |  ResearchGate