Federal Reserve officials have said countless times they take a “data-dependent approach” to their policy decisions, including their current conundrum of when to slash interest rates. But what if the data isn’t as dependable as it once was?
That’s what appears to be happening — and it’s making central bankers’ jobs a lot harder.
“We have to make decisions in real time,” Fed Governor Christopher Waller said late last year. “Whatever data is released, that’s the data I have to use. The problem with data is it gets revised.”
That wouldn’t necessarily be so much of an issue if the revisions, which can come months after initial reports are released, were relatively small. However, many revisions over the past few years have been game-changers.
For instance, Waller pointed out that initial monthly headline employment numbers for 2021 led him to believe that the job market was “okay, but it’s not really great.” Even though inflation was at a 40-year high, he and other Fed officials were under the impression that they’d need to proceed very carefully with raising interest rates, fearing it could lead to job losses, Waller said.
But along came the revisions.
For instance, in the initial jobs report for August 2021, the Bureau of Labor Statistics estimated 235,000 new jobs were added.
As is the case with any month’s jobs report, the change in employment is subject to three revisions. The first two come in the next two month’s reports. For August 2021’s headline employment number those first two revisions meant the number of jobs added was closer to 483,000, more than double what the BLS first estimated.
The third and final revision comes in the next calendar year from when the initial estimate was released as the BLS reconciles the survey data it collected to arrive at the initial estimates with the actual data it subsequently receives from state labor departments. For August 2021, that final revision brought the monthly change in employment up to 663,000, a 428,000 increase from the very first estimate.
That’s well above the average month-over-month revised change in job totals from 1973 to the latest available revision data, according to the BLS.
“When you look back, you’re like, “Oh my God the labor market was a lot stronger than it was indicated by the release of the data at the time,” Waller said in a lecture he gave, titled “Using Economic Data to Understand the Economy,” at the November conference.
In hindsight, it’s easy to say the central bank was too slow to raise interest rates, he said. “But in real time, maybe we weren’t so late.”
Last year, however, the opposite effect played out. Revised employment gains for all but one month, July, meant the labor market was much weaker than initial reports made it out to be. (Subsequent revisions for November and December haven’t all been released).
With that in mind, Waller said last month “there’s a good chance that December will be revised down as well.”
Waller isn’t the only Fed official to find data revisions problematic
“The frequency and extent of data revisions make the task of predicting how the economy will evolve even more challenging,” Fed Governor Michelle Bowman said in remarks she delivered in November at an event hosted by the Ohio Bankers League. She highlighted what were then recent revisions to monthly job gains as well as average hourly earnings.
The official summary of what Fed officials said and discussed during their September meeting — also known as the Fed minutes — stated: “A few participants observed that there were challenges in assessing the state of the economy because some data continued to be volatile and subject to large revisions.”
Spokespeople from the Federal Reserve declined to answer which data Fed officials were referring to. It could have been any number of economic indicator reports, though.
For instance, the Commerce Department initially estimated last April that the US economy grew at an annualized rate of 1.1% in the first quarter of 2023. That was far below the 2% rate economists had been forecasting at the time. But to Fed officials seeking to rein in inflation by raising interest rates to slow economic growth, the report may have suggested they were on the right track.
Then a month later the Commerce Department revised its initial estimate up slightly to 1.3%. A month after that, in June, the Department released a subsequent final revision. This time it said the economy grew at the 2% annualized rate economists originally forecast.
It’s completely normal for GDP data to change as it undergoes the standard revision process to incorporate more comprehensive data that becomes available or is also revised. However, the revisions to the first quarter GDP data in 2023 were higher than usual.
On average, the third revision is 0.6 of a percentage point higher or lower than the first estimate of a given quarter, according to a Commerce Department analysis of GDP data published from 1996 through 2022.
Consumer Price Index inflation data is also subject to revisions. But, unlike other pieces of data, there’s only a one-time revision that comes in February of the next calendar year as a result of updates to how seasonal adjustments are made.
Last year’s revision to CPI data from 2022 erased some of the progress that was thought to be made in bringing inflation down. Waller said last month he’s worried the CPI revisions for 2023 due later this month could also change “the picture on inflation.”
“My hope is that the revisions confirm the progress we have seen, but good policy is based on data and not hope,” he said.
Why all the revisions?
It’s nearly impossible to say with certainty, for instance, how much prices rose or how many people were hired at a given point in time across an entire country’s economy. Finding out how many new hires there were in a given month would involve asking every employer how many people were on their payrolls. That’s why the government and other economic data providers often rely on surveys to make sophisticated estimates.
The BLS, Census Bureau and other government agencies that conduct surveys that inform economic reports do rigorous work to make the best possible estimates with the information they gather. And more often than not they do a tremendous job at it.
But surveys, by nature, are imperfect.
In the same way that election polls don’t always predict the candidate who ends up winning, surveys don’t capture the exact true picture. However, they can get pretty close to the truth.
In election polls and government surveys, there’s a sample size of respondents designed to be representative of the overarching group studied. The larger and more diverse a sample, the closer an estimate will be to the true value.
Getting a large and diverse sample requires a lot of outreach to recruit people to be part of a group that BLS and other agencies enlist to regularly respond to a given survey. The rate at which people are getting recruited for surveys that are used in many of BLS’ monthly reports including employment, Consumer Price Index and Job Openings and Labor Turnover are down sharply from before the pandemic.
That makes it more likely that the survey sample could be more biased, said Erica Groshen, a former commissioner of the BLS. The agency, she said, “tests for all the biases they know about, but there’s more bias that could creep in the fewer people respond.”
“You do have to be concerned that the employers who were never agreeing to participate in the survey in the first place are somehow characteristically different from the ones who do agree,” said David Wilcox, a longtime Fed staff member who is now an economist at the Peterson Institute for International Economics and Bloomberg Economics.
For instance, it could be that firms that agree to participate in the BLS’ survey are more financially stable than ones that decline, he said.
“That suggests that there’s the potential possibility that we may be mismeasuring what’s going on in the economy,” he said, adding though that it would mainly apply to the preliminary estimates and not the final one.
The declines in recruitment rates, which help explain declining response rates, aren’t causing larger revisions between initial and final employment estimates, Laura Kelter, national estimates branch chief within the division of Current Employment Statistics at the BLS, told CNN. “However, we continue to monitor the issue,” she said.
Kelter said the declines can be attributed to a variety of challenges including the voluntary nature of participating in the surveys as well as survey fatigue — that is, people getting bombarded with too many surveys.
Groshen agreed, adding that other factors such as less civic responsibility, changes in technology like caller ID and spam filtering, and heightened concerns about confidentiality and data security are also at play.
Rolling with the punches
“Even in normal times, the Fed’s ability to perceive what’s going on in the economy is like an individual who has vision out of both eyes but in neither case is that vision 20/20,” said Wilcox.
New York Fed President John Williams told CNN last year he accepts that the preliminary government data he has available at monetary policy meetings is likely to change by the time officials meet again.
That’s why he takes into account as much outside data as possible, often from private sector data providers, to give him a better sense of what’s happening in the economy.
“We want to be data dependent, but not data point dependent,” Williams said. “I try not to get too excited about every data point that we see.”
