Monthly US Employment Analysis - August 2019

August 2019 Monthly US Employment Analysis

Edward Leamer, Professor Emeritus, UCLA Anderson School
September 6, 2019

The August payroll release of 130,000 jobs is less than the 159,000 jobs in July. It is difficult to discern the meaning of that decline because the payroll data have some extreme ups and downs. The news in these releases can be captured by two filtered series, one based on just the overall payroll data and another one that includes the components of the payroll data like nondurable manufacturing. The actual and the filtered versions are illustrated in the figure below.

The news last month made evident by the filtered series was that the decline in the rate of job growth had ended since the filtered data showed a stronger July than June. The August numbers eliminate most of those gains, which revives concerns about a trend downward in payroll growth from 220,000 per month a year ago to only about 160,000 recently.

Figure 1

One of the most important positive features of the August release is the annualized rate of growth of nondurable manufacturing equal to 0.8% which contrasts with the negatives that we had been suffering from 1995 to 2010.

Figure 2


The Bureau of Labor Statistics has released an estimate of an increase in payroll employment in August by 130,000. This is lower than the July release of 164,000 and considerably lower than the January estimate of 312,000, but a lot better than February and May, which were only 56,000 and 62,000.

The month-by-month payroll data clearly suffer from a lot of fluctuations that make it likely that especially good numbers are followed by weaker ones and especially bad ones are followed by better ones. To make wise use of these data, this noise needs to be removed from the data, leaving a series that does not support inappropriate swings back and forth from elation to despair.

The first figure above includes the actual data and two filtered versions of the actual data. One filter asks what would be the forecast for next month’s growth of payrolls using only the recent overall payroll growth numbers and the other filter uses also the payroll components including, for example, construction, durable and nondurable manufacturing and mining.

Both filters greatly smooth out the actual data. Both filtered versions make clear that there was a shift downward in the rate of growth of payrolls from above 200,000 jobs until February 2019 to about 160,000 thereafter. The filtered series that adjusts for abnormal components is still trending downward, suggesting that the structure of jobs underlying the recent total payroll numbers is not so favorable. In particular, the favorable government job gains of 34,000 are treated negatively in the component-based filters, since strong government payrolls to some extent are followed by weaker overall payrolls.

Historical Statistics

The table below reports mean annualized rates of growth of the components of overall payroll during the expansions since 1970, excluding the first 24 months of each expansion when growth is often strong because people are going back to work. Overall growth in payrolls has been 2.2%. The exceptional sectors with annualized rates of growth at 3% or more have been construction, professional and business services, education and health services, and leisure and hospitality. The weakest sector with rates of growth at 1% or less were mining, durable and nondurable manufacturing.

The last two columns in this table report the August release and a Z-stat equal to the difference between the August value and the historical mean, divided by the historical standard deviation. This column is color coded to attract attention to the strongest components of the release and the weakest. The strongest were nondurable manufacturing and government. The weakest were Trade, Transportation and Utilities, and Other Services.

Table 1

Regression Studies

The two regression models used to filter the noise from the data are based only on data from US expansions, after 1970 and after the first 24 months of the expansions, which fits today’s circumstances better than using all the data. In addition, the models include binary indictors for each expansion to capture the declining rate of growth of payrolls. These regressions have been estimated with a stepwise algorithm that omits variables with coefficients less than one in absolute value. These omissions increase the adjusted R-squared and can increase the accuracy of predictions.

The first regression uses only the lagged rates of growth of payrolls. The filtered July estimate using this model is the constant 0.17 plus the Expansion 12 effect -.009 plus the July value times 0.123 plus June times .234 plus May times 0.97 plus April times 0.88. This is not an unweighted moving average, and it is not a weighted moving average. A weighted moving average would have weights that add to 1.0 but these add only to 0.543. A weighted moving average does not have a constant, but this one has a constant equal to 0.008. This makes the filtered estimate smoother than a weighted average.

The second regression replaces the lagged rate of payroll growth overall with all the components, but with one statistically unimportant component omitted (PBS), and with the components ordered by the coefficient t-statistic, a measure of statistical importance. This means that the strong value of NDMANUEMP in the July release is good news, but the strong showing of EHS is bad news.

Table 2

Table 3

UCLA Anderson Forecast