Policy Hub: Macroblog provides concise commentary and analysis on economic topics including monetary policy, macroeconomic developments, inflation, labor economics, and financial issues for a broad audience.
Comments are moderated and will not appear until the moderator has approved them.
Please submit appropriate comments. Inappropriate comments include content that is abusive, harassing, or threatening; obscene, vulgar, or profane; an attack of a personal nature; or overtly political.
In addition, no off-topic remarks or spam is permitted.
May 19, 2022
An Evaluation of GDP Nowcasts during the Pandemic
On April 28, the US Bureau of Economic Analysis reported that real gross domestic product (GDP) contracted an annualized rate of 1.4 percent last quarter. This decline "surprised" GDPNow, the Atlanta Fed's GDP tracking model, which had projected a 0.4 percentage point growth rate the day before the official release. Professional forecasters, who generally expected a rate around 1 percent based on economist surveys from Reuters and the Wall Street Journal (WSJ), also turned out to be overly optimistic.
The lines in chart 1 represent errors for final forecasts of first-release estimates of real GDP growth from the WSJ Economic Forecasting Survey and GDPNow. In the five-and-a-half years before the pandemic, the WSJ survey and GDPNow both had average absolute forecast errors—that is, without regard to sign, or mean absolute errors (MAEs)—of about 0.5 percentage points. Since then, the MAEs have been 1.7 percent and 1.9 percent, respectively. The stacked bars in the chart represent a decomposition of the GDPNow line into forecast errors of subcomponent contributions to GDP growth. The chart makes it evident that the size of the bars has fallen since 2020, but continues to be larger than before the pandemic. Even GDPNow's relatively accurate forecasts for the second and fourth quarters of 2021 were largely the result of the positive and negative subcomponent errors fortuitously offsetting each other. GDPNow's error last quarter was largely concentrated in net exports. The model's final forecast of the growth rate of real final sales to private domestic purchasers—which excludes inventories and government spending in addition to net exports and has been shown to be a better leading indicator of one-quarter-ahead GDP growth than GDP growth itself is—was only 0.3 percentage points below the initial estimate of 3.7 percent.
The deterioration in GDP forecasting accuracy during the pandemic has not been isolated to the shortest horizon projections. Chart 2 shows MAEs for roughly 75-day-ahead forecasts of growth rates of real GDP and its subcomponents (as well as net exports and inventory investment contributions to growth) from both GDPNow and the Philadelphia Fed's Survey of Professional Forecasters (SPF). The decline in forecast accuracy for the SPF has been similar to the decline for GDPNow across subcomponents, with the exception of the state and local (S&L) government spending subcomponent, where the deterioration for GDPNow has been much starker. A Macroblog post from a year ago discussed the reasons for this.
The left-hand panel of chart 3 shows the MAEs for subcomponent contributions to GDP growth prior to the pandemic, and the right-hand panel shows contribution MAEs during the pandemic. Because the errors have been so much larger during the pandemic, the vertical axis is scaled to be six times larger in the right-hand panel than in the left-hand panel.
To show how forecast accuracy has evolved over a typical quarter, the figures in both panels of the charts begin with the MAE after the initial (roughly 90-day ahead) GDPNow forecast, end with the final GDPNow forecast, and use up to five other forecasts following particular data releases in between. Three of these releases are from the Institute of Supply Management's Manufacturing ISM Report On Business (its manufacturing report, specifically) for each of the three months of the quarter being forecasted, and the other two correspond to the personal income and outlays releases from the US Bureau of Economic Analysis for the first two months of the quarter.
Notice that although both panels show the subcomponent projections generally becoming more accurate over time, the rank ordering of the subcomponent accuracy has changed in some nontrivial ways during the pandemic. In particular, even though personal consumption expenditures (PCE) on services account for 45 percent of nominal GDP, it was consistently one of the smaller sources of error prior to the pandemic. But during the pandemic, it has been one of the largest. Government spending was also one of the smaller sources of error prior to the pandemic and has remained that way during the early part of pandemic quarters. But this decade, the government spending forecasts have not tended to become more accurate as quarters have progressed, so that recently, just prior to the GDP release, government spending has been nearly as large a source of error as services PCE.
In the GDPNow model, the forecasts of government spending, particularly S&L government sales to other sectors, were distorted by the large swings in the second and third quarters of 2020. This documentation on recent changes made to the GDPNow model describes the changes made to keep these types of distortions to a minimum in the future, which is relevant to the data releases used in chart 3. Generally, a few days after the release of personal income and outlays data, the ISM releases its manufacturing data for the subsequent month, providing one of the first data snapshots for that month. The ISM data for this month are used to estimate the model's factor, that then feeds through to the forecasting equations for much of the yet-to-be-released monthly GDP source data.
Chart 4 shows the MAE from the five quarters beginning in the first quarter of 2021 for real GDP growth and the S&L government contribution to GDP growth. The dashed lines are for the version of GDPNow that had been in use at the time, and the solid lines are for the "COVID-adjusted" model in use since the first quarter of 2022. (The same data releases used in chart 3 are used here.) The modified model shows a clear improvement in the forecasts of S&L government spending. Moreover, forecast accuracy no longer deteriorates following the release of the first-month and second-month ISM manufacturing releases. After the second personal income and outlays report of the quarter—and the final one before the initial GDP report—the accuracy and forecast values of the two models became more similar.
Although this is good news for improving the forecast accuracy of GDPNow, it is not clear whether near-term forecasts of GDP growth will continue to be less accurate than they were in the decade leading up to the pandemic. Ongoing geopolitical and economic uncertainties remain elevated, which could continue to have a negative impact on forecast accuracy for some time to come.
March 1, 2022
Assessing Recent Labor Market Improvement
The US Bureau of Labor Statistics' (BLS) labor report for January 2022 showed that the overall labor force participation (LFP) rate increased 0.3 percentage points from December's published level. This increase put the LFP rate at its highest level since the pandemic began and, taken at face value, might make you think that the labor supply problems that have plagued the recovery from the COVID-19 pandemic were easing.
However, it turns out that this jump in the LFP rate was entirely an artifact of the BLS incorporating population control adjustments into the January labor force data. These are independent estimates of the civilian noninstitutionalized population ages 16 and older used to make sure that labor force statistics computed from the Current Population Survey (CPS) accurately reflect the population and are incorporated into the CPS data each January. The latest adjustments are the first to use information from the 2020 decennial census , and showed that the US population was almost 1 million larger than the published estimate for December 2021. By itself, a jump in the size of the population isn't an issue for comparing LFP rates over time. But the new population adjustments also showed that the population was considerably younger than previously estimated (in particular, the share of the population aged 70 and older was smaller). This shift in the age distribution is important because a younger population generally means a higher rate of participation in the labor force.
The BLS does not revise the historical data when new population control adjustments are incorporated into labor statistics. But it did report that the population control adjustment would have lifted the December 2021 LFP rate for the population ages 16 and older by 0.3 percentage points if it had revised the December data. This increase is the same as the increase in the published LFP rate from December to January. In other words, the December-to-January increase in the published LFP rate didn't indicate an improvement in labor force participation at all.
Clearly the latest population control adjustments complicate comparison of 2022 LFP rates to earlier periods. To construct historical LFP rate series that are more comparable over time, we implemented a simple smoothing method the BLS used previously to account for annual population control adjustments (described here ). This method essentially distributes the level shifts that result from the population control adjustments back over the relevant historical period for each series. To account for the effects of adjustments (made between the decennial census) to the census 2010 population base that were made in January 2013–January 2021, we first smoothed data for January 2012 to December 2020. Then we smoothed the data for January 2012 to December 2021 to account for the effects of the 2020 census population control adjustment introduced in January 2022. We applied the method separately labor force participation rates for the population ages 16 and older, as well as populations ages 16–24, 25–54, and 55 and older. You can see these series in this spreadsheet.
Chart 1 plots the published and smoothed seasonally adjusted LFP rate series for the population aged 16 and older. Notice that the smoothing method results in a gradually increasing upward shift to the LFP rate over the 10-year period, culminating with the December 2021 smoothed LFP rate 0.3 percentage points higher than the published LFP rate.
The upward shift is even greater for the population aged 55 and older shown in chart 2. Recall that a significant part of the population adjustment was a reduction in the size of the population aged 70 and older. Given that this age group has a lower LFP rate than those aged 55 to 69, the composition shift pushed the LFP rate higher for the 55 and older population overall. The BLS estimated that the population adjustment impact on the December 2021 LFP rate for this population group was 0.7 percentage points.
For the population aged 16–24, the smoothed series shown in chart 3 is lower than the published series through December 2021. This is because the population adjustments revealed that the population aged 16–19 was larger than previously estimated. Because the 16–19 age group has a lower LFP rate than those aged 20–24, the smoothed LFP rate series is lower than the published series. The BLS estimated that the population adjustment impact on the December 2021 LFP rate for this population group was −0.3 percentage points.
Finally, for the prime-age population (25–54), the smoothed series shown in chart 4 is identical to the published series. The population adjustments had no effect on the LFP rate for this age group.
It is important to consider how the difference between the published and smoothed series may alter one's assessment of labor market dynamics surrounding the pandemic-induced recession and subsequent recovery. Chart 5 shows that for the initial year of the pandemic, measured here as the change in the LFP rates from January 2020 to January 2021, the published LFP rate data (blue bars) and the smoothed estimates (green bars) tell a very similar story for all age groups: LFP rates declined between 1.5 percentage points and 2.0 percentage points for all age groups, with no material difference in the size of the change between the smoothed and published estimates within each age group.
However, the story is much different for the period between January 2021 and January 2022. Chart 6 depicts this difference, showing that the published LFP rate for the population aged 16 and over increased by nearly 0.8 percentage points over that 12-month period. In contrast, the corresponding smoothed estimate increased by only 0.5 percentage points. The discrepancy is even greater for the population aged 55 and older. For that age group, the published LFP rate increased by 0.8 percentage points, whereas the smoothed LFP rate shows an increase of less than 0.2 percentage points during the past year. That is, the smoothed data suggest the increase was less than one quarter as large as the increase in the published data. For the population aged 16–24, the smoothed LFP rate increased by more than the published series (0.7 percentage points versus 0.5 percentage points). Finally, for the population aged 25–54, there is essentially no difference between the change in the published and smoothed estimates. Both increased by close to 0.9 percentage points from January 2021 to January 2022.
To sum up, smoothing the labor force data to account for the annual population adjustments like that described here provides a way to allow LFP rates in 2022 to be compared to prior years. These estimates show that the recovery from January 2021 to January 2022 in the overall LFP rate and the rate for the population aged 55 and older is more modest than the published data would imply, while the recovery for the population aged 16-24 is better than implied by the published estimates.
Going forward, the published labor force data for January 2022 will be comparable with data for other months in 2022 since they are all based on the same population control adjustments. In January 2023, the BLS will likely incorporate new population adjustments that use additional 2020 census information. Hopefully those adjustments will be less eventful. Stay tuned as we discuss future data.
February 3, 2022
Rounded Wage Data and the Wage Growth Tracker
The US Census Bureau recently announced some changes it plans to make this year to the Current Population Survey Public Use File (CPS PUF). Here at the Atlanta Fed, we use data from the CPS PUF to construct the Wage Growth Tracker, and one of the planned changes will significantly affect the tracker. Specifically, a person's usual weekly or usual hourly earnings, which are unrounded currently, will be rounded.
The Wage Growth Tracker bases its results on the median, or middle, observation in the distribution of percent wage changes for a sample of individuals linked between the current month and the same month a year earlier. The Wage Growth Tracker time series has yielded useful insight into the rapidly shifting dynamics of the labor market in the wake of COVID-19, especially as compositional effects have distorted wage data. It's also helped economists and policymakers understand which income levels were seeing the greatest growth and that job switchers were finding the most wage gains.
How will the rounding of wage data affect the Wage Growth Tracker? The announced CPS PUF rounding rules vary by wage level and are different if the earnings are reported on an hourly or weekly basis. (You can see more details here .) Most people in the CPS report earnings on an hourly basis, and most wage observations range from $10 to $99.99 an hour. Under the rounding rules those earnings will be rounded to the nearest dollar. So, for example, if someone reports making $14.50 an hour, that wage will be rounded to $15, while workers reporting a wage of $15.40 an hour will also have that wage rounded to $15.
One implication of the rounding rules is that it will make no wage change appear more common than it currently is. To illustrate, suppose someone's pay went from $14.50 to $15.40 an hour. The rounding rules would show no change in the person's wage (both would be recoded as $15) even though that person's actual wage increased by 6.2 percent. Chart 1 shows what happens to the proportion of zero wage changes if the rounding rules were applied to the CPS PUF earnings data used to construct the Wage Growth Tracker from 1998 to 2021.
As you can see, during the Great Recession, when labor demand was especially weak, about 17 percent of wage growth observations based on unrounded earnings data were zero. But if the rounding rules had been applied back then, more than 25 percent of wage growth observations would have been zero.
Obviously, this change is a big deal for the Wage Growth Tracker. When more than a quarter of the observations are zero near the middle of the wage change distribution, it is very likely that the median observation will also be zero. This effect is evident in chart 2, which compares the median Wage Growth Tracker series using unrounded earnings data with what it would have been if the rounded data had been used.
Clearly, if the rounding rules had been in place in the past, the Wage Growth Tracker time-series would be a much less useful indicator of wage growth or labor market trends.
So, what to do? It turns out that the rounding rules don't affect all summary measures of wage growth as much as they affect the median measure. For example, as chart 3 shows, the mean—or average—wage growth comes out of the rounding changes essentially unaffected.
Unfortunately, not only is the average higher than the median, because wage growth varies greatly across individuals (the monthly sample standard deviation is typically around 25 percent), you can also see that both of the average wage growth series are much more variable month to month than the median series using unrounded data. Indeed, the robustness to variability in the underlying wage change data is a primary reason why the Atlanta Fed's Wage Growth Tracker is based on median rather than average wage growth.
But there is potential solution. Borrowing from the research on using trimmed means of price change data to construct measures of inflation that are robust to extreme price changes, I was able to construct a trimmed-mean wage growth series using the rounded data that has broadly similar properties to the (median) Wage Growth Tracker series constructed from unrounded data. Specifically, for each month's sample, I excluded the bottom 20 percent of wage growth observations (that is, the largest percent wage declines) and the top 25 percent (the largest percent increases) and computed the average of the remaining data. (Note that the trimming is asymmetric because more of the large wage changes tend to be increases than decreases, which is also why the average is higher than the median in the previous chart.)
Chart 4 shows the trimmed-mean series constructed using rounded earnings data, along with the (median) Wage Growth Tracker series that uses unrounded data. I would describe this trimmed-mean series as a reasonable (though not perfect) approximation of the Wage Growth Tracker series (something we could have used if we only had rounded earnings data in the past).
When the January 2022 CPS PUF data become available in February, we will produce the trimmed-mean version of the overall Wage Growth Tracker and add it to the Atlanta Fed's Wage Growth Tracker data set. We are currently exploring if a similar approach will produce useful alternatives to the Wage Growth Tracker for other ways to view the data, such as those for job switchers versus job stayers, or by average wage level. Watch this space for updates.
January 12, 2022
Hybrid Working Arrangements: Who Decides?
Even after—or should we say "if"?—working from home eventually becomes less of a necessity, it's likely to stick around in a hybrid form, with some working days performed at home and some in the office. (This recent study , coauthored by three of this post's authors, also makes this case.) Still, much remains undetermined about how that hybrid arrangement will work and who at the firm decides how many and which days employers will require workers to be onsite.
To shed some light on how hybrid working arrangements are working, we posed a few special questions to executives in our Survey of Business Uncertainty (SBU) last July and again last month (December 2021). Specifically, we asked, "Does your firm currently have employees who work remotely?" If they said yes, we followed that up with the question, "Who decides which days and how many days employees work remotely?" Respondents selected options ranging from fully decentralized to company-determined schedules. (The results between the July and December surveys were nearly identical, so we've combined them here to simplify this discussion.) Among firms in our panel, 53 percent have employees who work remotely, and their survey responses are interesting (see chart 1).
As you can see, respondent firms are roughly split, with about 30 percent leaving the decisions up to their employees, 30 percent giving teams (or team leads) decision rights, and nearly 40 percent indicating the decision on how many and which days employees will be remote resides at the company (management) level.To dig into these results a bit further, we looked at who makes these decisions over working arrangements by industry and firm size. Given the differences across the industrial sector's ability to work from home (see research by Jonathan Dingel and Brent Neiman), we find it somewhat surprising that little difference exists across industries about whether the decision to work remotely is fully decentralized, made at the team level, or determined by the company (see chart 2).
However, we see a stark difference when comparing these decision rights by firm size. More than half of the smallest firms in our panel (those with fewer than 25 employees) allow the employees to decide how and when to come into the office, compared to just 10 percent of larger firms (with 250 employees or more). Instead, these larger firms have left decisions about remote work with the team. Although that's certainly far from a rigid, top-down approach, it can suggest a need for coordination among teams, and this variation highlights remote work's big trade-off: balancing employee choice with the coordination that work life sometimes requires.
Allowing employees to choose their teleworking days has the benefit of flexibility, letting them to plan their work schedules around some nonwork commitments. But it has the cost of limiting face-to-face meetings, as on any given day of the week larger teams will likely find one or more members working remotely, which forces meetings partly or completely online. In our discussions with larger firms, they highlight the importance of face-to-face interactions and so have been promoting team- or company-level coordination. Interestingly, smaller firms appear to be walking another path: providing greater individual choice. Which one of these approaches becomes prevalent should become clear by the summer, when employees can (hopefully) return to the office. Though the future of office work appears to be a hybrid one, the form of decision making that will dominate that future has yet to be determined.
- Business Cycles
- Business Inflation Expectations
- Capital and Investment
- Capital Markets
- Data Releases
- Economic conditions
- Economic Growth and Development
- Exchange Rates and the Dollar
- Fed Funds Futures
- Federal Debt and Deficits
- Federal Reserve and Monetary Policy
- Financial System
- Fiscal Policy
- Health Care
- Inflation Expectations
- Interest Rates
- Labor Markets
- Latin AmericaSouth America
- Monetary Policy
- Money Markets
- Real Estate
- Saving Capital and Investment
- Small Business
- Social Security
- This That and the Other
- Trade Deficit
- Wage Growth