Do Group Unemployment Rates Send Warning Signs about the Broader Labor Market?

for various groups g over horizons k from 1 to 12 months, where ∆ represents the one-month change from the prior period. The coefficient ${\beta }_1$ represents the degree to which the overall unemployment rate will be higher k months from now when the unemployment rate for group g rises by one percentage point from last month to this month, controlling for this month’s change in the overall unemployment rate and the last two months of changes in both the overall unemployment rate and the unemployment rate for the group in question.¹

I first consider the groups that the research mentioned above suggests experience greater cyclicality in labor market outcomes: Black workers, Hispanic workers, workers ages 16 through 19, and workers who have completed less than a high school education (LTHS). The Bureau of Labor Statistics publishes official seasonally adjusted estimates of the unemployment rates for these groups each month, and I use these estimates to calculate changes in unemployment rates.² Figure 1 plots the ${\beta }_1$ coefficients from the equation above, estimated separately for each group at horizons from one to 12 months.

Estimates suggest that the unemployment rates for Black workers and those who did not complete high school do predict higher overall unemployment two to six months later. The LTHS unemployment rate is more immediately predictive, with a 1 percentage point increase in that measure associated with an increase in the overall unemployment rate that reaches about 0.2 percentage points two months later. This effect is statistically significant from two to four months later and fairly consistent in magnitude through 10 months later, after which point it declines modestly.

Changes in the aggregate unemployment rate following increases in the Black unemployment rate take a little bit longer to manifest fully but reach a similar magnitude. The overall unemployment rate tends to be about 0.2 percentage points higher four to seven months later for each percentage point increase in the Black unemployment rate today. This relationship is statistically significant in months five and six. In both the Black and LTHS cases, the estimated subsequent increases in the overall unemployment rate are larger than would be implied mechanically by a persistent increase in unemployment within these groups, given their shares of the labor force.³

Coefficients on increases in the Hispanic unemployment rate are consistently around 0.1 percentage points in months two to 11, but never statistically significant over the horizons considered. Estimates for the unemployment rate for ages 16 to 19 are consistently closer to zero. For all four of these initial groups, results are qualitatively similar, if occasionally less precise, if regressions control for only one prior month of changes in overall and group-specific unemployment rates rather than two. Results are also similar, and even slightly stronger, if the regression is specified in levels of overall and group unemployment rates rather than changes. In this formulation, the predicted increases in the overall unemployment rate associated with increased Black or LTHS unemployment peak at about 0.3 percentage points rather than 0.2 percentage points.⁴

Results for Other Groups

While the public discussion of leading indicators is often oriented around groups with the most cyclically sensitive unemployment rates, the monthly employment report details conditions for a wide range of groups, including one very much at the heart of the labor market: prime-age workers, those from the ages of 25 through 54. Figure 2 repeats the prediction exercise above for narrower age ranges within this group (25–34, 35–44, and 45–54) and reveals that increased unemployment rates among workers in the middle of their prime working years (35–44) is also statistically significantly associated with future increases in the overall unemployment rate.

Increases in the unemployment rate for workers aged 35–44 predict a higher overall unemployment rate with statistical significance even one month later, at which point overall unemployment is 0.1 percentage points higher for each 1 percentage point increase within this group. The increase in overall unemployment grows over time before leveling off at a little over 0.3 percentage points four months later and remaining at roughly that level through at least 12 months later.⁵

Looking to other groups defined by other characteristics does not yield the same success in identifying leading indicators. The unemployment rate for white workers is very highly correlated with the overall unemployment rate. Once changes in the overall unemployment rate are controlled for, there is little predictive power left in the white unemployment rate. Estimates for other education groups are too imprecise to rely on.

Conclusion

The intuition that changes in unemployment rates for more cyclically exposed groups of workers might tell us something about how the overall unemployment rate will change going forward turns out to be partially correct. Increases in the Black and LTHS unemployment rates tend to be followed by higher overall unemployment, on the order of a 0.2 percentage point increase in overall unemployment per 1 percentage point increase in group-specific unemployment. But the predictive power of group unemployment rates is not limited to these more vulnerable groups. The unemployment rate among workers aged 35–44, the age group that typically has the highest employment and participation rates among those officially tabulated, also predicts subsequent changes in the overall unemployment rate, and its relationship to those subsequent changes is, if anything, stronger than the other groups considered.