U.S. job growth revised down by the most since 2009. Why this time is different

There’s been a lot of debate about how much of a signal we should take from the downward revision of U.S. payroll numbers by 818,000, the biggest since 2009. Is it a sign of a recession?

A few facts worth considering:

• By the time the 2009 revision was released (which overstated 824,000 jobs), the National Bureau of Economic Research had already declared a recession six months earlier.
• Jobless claims, the data source for the same period, surged to more than 650,000, and the unemployment rate among insured people peaked at 5% that month.
• At the time, GDP figures had already recorded four consecutive quarters of contraction (two of which have since been revised upwards, and one of which has since been revised upwards to show growth rather than contraction, but the economic weakness was broadly evident in the GDP numbers, the ISM index, and many other data points).

The current revision covers the April-March 2023 period, so we don’t know if the current numbers are higher or lower. The model the Bureau of Labor Statistics uses may be overstating the strength of the economy at a time when weakness is building. There are signs of a softening of the labor market and economy, and this could be further evidence of that, but here’s how the same indicators in 2009 are performing now:

• No recession has been declared.
• The four-week moving average for jobless claims is at 235,000, unchanged from a year ago, and the insured unemployment rate is at 1.2%, unchanged since March 2023. Both are a fraction of the numbers from the 2009 recession.
• This marks the eighth consecutive quarter of reported GDP growth, and would have continued for longer had it not been for anomalies in the first two quarters of 2022.

As a signal of serious weakness in the economy, this large revision is currently an outlier compared to the data for the same period, but it is more or less accurate as a signal that job gains during the revision period were overstated by an average of 68,000 jobs per month.

But this would only reduce the average employment gain from 242,000 to 174,000. How the BLS distributes this weakness over the 12-month period will help determine whether the revisions were concentrated toward the end of the period and therefore more relevant to current conditions.

If that were the case, the Fed might not have raised rates as much. If the weakness persists beyond the correction period, the Fed’s policy might be more accommodative now, especially if, as some economists expect, productivity figures rise further because the same level of GDP appears to be achieved with less work.

But the inflation numbers remained the same, and the Fed was reacting more to inflation numbers than to employment data during the period in question (and still is).

Therefore, this revision may slightly increase the likelihood of a 50 basis point cut in September for the Fed, which is already leaning towards a rate cut in September. From a risk management perspective, this data may intensify concerns that the labor market is weakening more rapidly than previously thought. In the process of cutting rates, the Fed will likely monitor growth and employment data more closely, just as it monitored inflation data more closely in the process of raising rates. However, the Fed is likely to place more weight on current jobless claims, business surveys, and GDP data than on retrospective revisions. It is worth noting that in the past 21 years, revisions have been in the same direction only 43% of the time; that is, 57% of the time, a negative revision is followed the year after by a positive revision, and vice versa.

Data agencies make mistakes, sometimes big mistakes, and they frequently go back and correct their mistakes, even three months before an election.

In fact, Goldman Sachs economists said late Wednesday that the Bureau of Labor Statistics may have overstated its revised estimate by as much as 500,000. The Wall Street firm said that undocumented immigrants who were initially registered as workers but are no longer registered with the unemployment insurance system account for some of the discrepancy, in addition to the tendency for initial revisions to be inflated.

Employment data is subject to noise from immigrant hiring and can be volatile, but if the economy was struggling like it did in 2009, there would be plenty of macroeconomic data to signal that. Right now, that’s not the case.