The employment dynamics of different population groups over the business cycle

ABSTRACT

We examine differences in employment dynamics across population groups using Bayesian vector autoregressions. We document that groups who are particularly strongly affected by business-cycle fluctuations (males, young people, non-whites, the less educated, and workers in blue-collar occupations) also tend to be affected early in the build-up of a boom or bust. We further identify the drivers of the different cyclicalities across population groups. Supply shocks seem to be most important for the heterogeneous employment fluctuations and particularly for the early effects of recessions and booms on the most affected groups. Dynamics in sectoral activity and in hiring rates can help to understand our findings.

KEYWORDS:

• Employment
• demography
• heterogeneous labour-market outcomes
• business cycles

JEL CLASSIFICATION:

• J10
• J21
• E24
• E32

Acknowledgements

Parts of this research have been produced while Bredemeier was at TU Dortmund University. Financial support from Deutsche Forschungsgemeinschaft through SFB 823 is gratefully acknowledged.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

¹ See, for example, Jefferson (2008), Elsby, Hobijn, and Şahin (2010), and Hoynes, Miller, and Schaller (2012).

² For example, the statement of purpose of the recent U.S. fiscal stimulus (ARRA) states that is was the stimulus’ goal to ‘preserve and create jobs’ (point 1) and, in particular, ‘assist those most impacted by the recession’ (point 2).

³ Where available, we use employment-to-population ratios. Where this is not available in the data, we work with employment headcounts by group. The cyclical behaviour of employment ratios between such groups is similar to the ratio in the employment-to-population ratios as long as there is no substantial cyclical component in population size of the different groups considered.

⁴ As in the original Minnesota prior (e.g. Doan, Litterman, and Sims 1984), it is assumed that there is no prior uncertainty regarding $\mathrm{\Sigma }$.

⁵ Every time we are considering an alternative employment ratio, we re-estimate our five-variable VAR and obtain different impulse responses for all variables. However, it turns out that the estimated responses of output, aggregate employment, the real interest rate, and labour productivity are barely affected by rotating-in a different employment ratio. In Figure 2, we consider results from the VAR with the gender employment ratio. Results from the other VARs are available upon request.

⁶ For this and the following employment ratios, we proceed as follows. Consider two groups which sum up to the total population with population shares $w_1$ and $w_2$. Let, in any given period, ${\hat{n}}_1$, ${\hat{n}}_2$, $\hat{n}=w_1{\hat{n}}_1+w_2{\hat{n}}_2$, and $\hat{r}={\hat{n}}_1-{\hat{n}}_2$ denote the percentage trend deviations of employment in group 1, employment in group 2, aggregate employment, and of the employment ratio. From our estimations, we take the medians of $\hat{n}$ and $\hat{r}$ and calculate ${\hat{n}}_2=\hat{n}-w_1\hat{r}$ and ${\hat{n}}_1={\hat{n}}_2+\hat{r}$. For dimensions with more than two groups, we sum up groups to two larger groups (e.g. production and construction versus other occupations). We then calculate the corresponding employment ratio $\hat{r}$ between the larger groups as the population weighted averages of the ratios between the smaller groups and then proceed as above. To obtain the change in the total number of jobs by group, we multiply percentage trend deviations by the long-run means of employment in this group.

⁷ Though we are aware of the limits of interpreting shocks as a policy reaction which might – to a certain degree – be systematic, we find this thought experiment insightful to explore what demand-side policies can or can not achieve.

⁸ As measures of activity in the construction sector, we consider total private residential investment (Source: BEA, Series ID: PRFI) and total construction (Source: OECD, Series ID: PRCNTO01USQ661S). Series are divided by Civilian noninstitutional population, logged and detrended (HP 1600). We find that they lead the cycle by six and two quarters, respectively.

⁹ Using the series ‘All Employees: Government’ (Source: BLS, Series ID: USGOVT, divided by the civilian noninstitutional population, logged and detrended (HP 1600)), we find that it has the strongest correlation with aggregate employment at lag $i=5$.

¹⁰ We find that total private hires (Source: BLS, Series ID: JTS1000HIL, divided by civilian noninstitutional population, logged and detrended (HP 1600)) lead aggregate employment by two quarters.

Additional information

Funding

This work was supported by the Deutsche Forschungsgemeinschaft [Grant Number SFB 823].