Wage Rigidity and Job Creation

Christian Haefke, Marcus Sonntag and Thijs van Rens

Abstract

Standard macroeconomic models underpredict the volatility of unemployment fluctuations. A common solution is to assume wages are rigid. We explore whether this explanation is consistent with the data. We show that the wage of newly hired workers, unlike the aggregate wage, is volatile and responds one-to-one to changes in labor productivity. In order to replicate these findings in a search model, it must be that wages are rigid in ongoing jobs but flexible at the start of new jobs. This form of wage rigidity does not affect job creation and thus cannot explain the unemployment volatility puzzle.

August 2008 [download pdf] - Also available as IZA Discussion Paper 3714.
First version: April 2007
Previous version (July 2007) still available here.

Data

The dataset includes the following time series for wages in the US, 1979-2006 (quarterly):

• Mean and median wage of all 25 to 60 year old workers (nominal and real)
• Mean and median wage of newly hired workers
• All of the above, corrected for composition bias

In addition, composition bias corrected versions of these wage series are included, as well as the sample averages and regression coefficients from the earnings equation of the worker characteristics that were used to correct for composition bias:

• Education level
• Experience
• Gender
• Race and ethnicity
• Marital status

Also included in the dataset are the standard errors for all wage series and various aggregate variables.

All series from the CPS are hours-weighted means/medians and are seasonally adjusted. Wages are for non-supervisory wage and salary workers from the private non-farm sector. These data were constructed from the CPS outgoing rotation groups and basic monthly data files. See the paper for a full description of the data.

Download:

• Dataset described above (comma-separated file)
• Outgoing rotation group data from the CEPR website
• Basic monthly data from the NBER website
• Stata do-files to construct the wage series from the underlying micro-data
• Stata do-files to produce the results of the paper

Note: If you use the aggregate data to calculate business cycle statistics or at the right-hand side of a regression, you may want to correct your estimates for sampling error, see the paper for details.

Legend

WAGE SERIES:

• lwage = log wage (nominal)
• lrwage = log real wage
• se = standard error

Subgroups of workers:

• allm = all workers
  newh = newly hired workers

Measure of centrality:

• mn = mean
• md = median

Composition bias corrected series:

• sbp_educonly = corrected for fluctuations in the average education level
• sbp_all =  corrected for fluctuations in all observable worker characteristics

Worker characteristics:

• b = coefficient from a Mincer regression of the log wage on various worker characteristics
• m = sample average of various worker characteristics for various subgroups of workers
• educ = years of education
• exp = years of experience
• expX = experience to the power X
• female = fraction of female workers
• black = fraction of African-American workers
• hisp = fraction of Hispanic workers
• marrd = fraction of married workers

Other variables:

• year = year
• qrt = quarter
• lwage_agg = log aggregate wage (nominal)
• lrwage_agg = log real aggregate wage
• lrlabprod = log real labor productivity
• unempl = unemployment rate

If you have any questions about these data or are interested in series that are not in the downloadable file, please email me.

---

Thijs van Rens  | CREI  |  Department of Economics and Business  |  Universitat Pompeu Fabra