It's what you thought you knew but didn't that breaks your heart: GDP revision edition

This week, the Bureau of Economic Analysis will update the national accounts, which include GDP and other key statistics. It could alter the narrative of this recovery and the wisdom of Fed policy.

Source: Getty Images.

Today’s post is a wonky one but important one. A data-driven Fed requires data, and those don’t come down from heaven. I talk here about the data improvements coming this week.

What the Fed does or doesn’t do next hinges on data; in fact, the Fed always calls itself “data-driven.” Fed Chair Powell underscored that multiple times at last week’s press conference after the FOMC meeting:

We will continue to make our decisions meeting by meeting, based on the totality of the incoming data, and their implications for the outlook for economic activity and inflation as well as the balance of risks.

Clearly what we decided to do is maintain the policy rate and await further data. We want to see convincing evidence, really, that we have reached the appropriate level, and we're seeing progress, and we welcome that. But we need to see more progress before we'll be willing to reach that conclusion.

And really what people are saying is let's see how the data come in. You know, we want to see -- you know what we want to see. We want to see that these good inflation readings that we've been seeing for the last three months, we want to see that it's more than just three months, right?

Okay, but how does a data-driven approach work when the data are subject to revision, and especially when the revisions are big?

The data will change soon.

Thursday is the Bureau of Economic Analysis’ comprehensive revision for the national accounts, including GDP, inflation, consumer spending, investment, and much more. The updates will cover the first quarter of 2013 to the first quarter of 2023.¹ We get the third revision to the second quarter of this year, too.

In addition to a massive amount of new source data, the revised series will reflect several methodological changes:

• Improved classification and measures of real estate investment trusts

• Improved measures of regulated investment companies

• New measures of monetary interest paid by Federal Reserve Banks

• Improved measures of the use by industries of financial intermediation services furnished without payment

• Improved measures of housing services [split between rent and included utilities]

• Improved measures of intermediate expenses for meals and entertainment

• Improved measures of National Flood Insurance Program services

• Improved measures of investment in own-account software

• Improved measures of brokers' commissions

• Improved price measures [for cloud computing and solar and wind structures]

Scanning through the list, the connection between the evolving economy and the evolving statistics is clear. Statistical agencies are responsive. Another example: the BEA will start publishing ‘super core’ inflation (core services excluding housing)—a recent focus of the Fed’s—and the addition from BEA is good because they can do the chain aggregation more accurately than anyone else. Quality matters.

Having the best data often takes time and multiple revisions. This week, numbers will change, many of them. The revisions might not make a meaningful difference for headline numbers like GDP, though they often do, especially at comprehensive revisions. The revisions can change the story or at least its emphasis. It did in the Great Recession.

To be clear, revisions are not a sign of problems at the BEA, which runs the comprehensive revision for the national accounts, or other statistical agencies that contribute new source data. And revisions are no reason to whip up conspiracy theories. The U.S. economy is $27 trillion, and it is complex. As one example, GDP estimates are built from the bottom up and cover a wide range of sectors (C+I+G+NX). An incomprehensibly many numbers are subject to revision this week.

It takes time to gather all the data, and some important sources, like the quinquennial Economic Census, which is mandatory for the vast majority of business establishments, are too demanding to collect every year, let alone every quarter. So, while the revisions may upend stories spun off the early data releases, they are signs of the estimation methods constantly improving and more data coming in.

That said, revisions make the job of policymakers and their analysts harder, especially those like the Fed, who are “data-driven” now.

GDP

Given the massive amount of data included in GDP and the frequent methodological improvements, it often revises noticeably. And even the initial releases of GDP often surprise forecasters. Last week, Fed officials doubled their median forecast for real GDP growth this year and then used that additional strength, in part, to justify its higher path for the federal funds rate. We will see how that GDP forecast stands up.

History shows us that revisions can be important. In a prior research note, I showed that during the Great Recession, the Fed staff’s forecast and then-available data for GDP vastly understated the severity of the Great Recession.

Source: Claudia Sahm (2015).

Specifically, in mid-2008, the Fed expected real GDP that year would increase by 1% (first blue bar). Instead, by the end of 2008 (second blue bar), the data showed a slight contraction. With each annual revision (moving further left to right), the estimate for real GDP growth in 2008 became weaker. Currently, the estimate is a decline of 2.5% in 2008. That’s over a 3 percentage point swing from the expectation in mid-2008.

If fiscal and monetary policymakers had known the economy was contracting by 2.5% in 2008—including a drop of 1.6% at an annual rate in the first quarter—they would have likely acted faster and more aggressively. The start of a recession is the best time to soften the free fall, but it was a year and a half later when the revisions showed the scope of the downturn in 2008, and there was no going back.

Inflation

Personal Consumption Expenditure (PCE) prices—the Fed’s preferred measure for its inflation target—will be revised this week, too, and that could affect the trajectory of inflation at a time of heightened interest. Economists Richard Audoly, Martín Almuzara, Richard Crump, Davide Melcangi, and Roshie Xing at the New York Fed show in a research note that revisions to core PCE inflation are frequently notable.

Again, the Great Recession is an example of a large revision. The 12-month core PCE inflation in March 2009 was 1.8 percent in the initial release. But after years of revisions, in the current data, it is 0.8 percent. A one percentage point difference in core inflation is massive, and in this case, putting the economy much closer to the ‘danger zone’ of deflation than was apparent at the time.

Revisions to PCE core inflation are common, even outside of severe recessions. Over two-fifths of monthly inflation—shown in the chart—revised up more than 0.6 or down more than -0.6 percentage point at an annual rate. Revisions of this magnitude might have reshaped policymaking; that’s especially likely in the past three years.

Source: Richard Audoly, Martín Almuzara, Richard Crump, Davide Melcangi, and Roshie Xing (2023.

Monthly revisions can cancel out across twelve months or even three, so they might not show up in our standard inflation metrics. However, the researchers also found that the revisions to core services were larger and more persistent than core goods in the past. And since core services are a much larger share of overall core inflation, it drives the revisions. Plus, given the attention to core services, it is a series to watch in the revisions.²

The Great Recession, which motivated the GDP and the inflation examples above, was a severe downturn, and most revisions are less consequential. However, the pandemic was unprecedented and severe, and its recovery was unusually rapid, so don’t count out big revisions, though it may take several years to see them.

There’s reason to believe GDP growth will revise down.

We won’t know the revisions until Thursday, but there are often regular patterns. And those suggest that GDP could likely revise down. Why is that? Gross Domestic Income (GDI)—conceptually the same as GDP—has been notably weaker than GDP recently.³

Specifically, real GDI declined (-3%) in 2022:Q4 and 2023:Q1 (-2%); it was flat (-0.5%) in 2023:Q2. All this while real GDP increased steadily (2% to 2.5%). That’s a very different picture. And if economic activity were, in fact, closer to GDI, then the puzzle of growth holding up as interest rates shot up and consumer sentiment was in the doldrums would be less puzzling. And most importantly, it would suggest that the Fed has less work to do.

In the past several decades, the initial releases of GDI have done a better job than the initial releases of GDP at predicting output after years of revision. The average of GDI and GDP referred to as Gross Domestic Output (GDO), is a good proxy for actual output.⁴ In an issue brief at the Council of Economic Advisers, I showed GDI estimates (blue bars) in 2007 and 2008 better captured the depths of the Great Recession and the slowdown leading into it (after years of revisions) than GDP (red bars) at the time. If the Fed had relied on GDI more than GDP or an average of the two, they would have had a more accurate picture of economic activity.

The takeaway from this analysis is to use the average of GDP and GDI from the early releases. Recently, that would imply growth of -0.9%, 0.2%, and 1.5% for 2022Q4, 2022Q1, and 2023Q2, respectively, each rate is well below GDP growth (2% to 2.5%) and below the Fed’s projected trend growth (1.9%). According to its forecasts last week, the Fed is using GDP, at least partly, to argue that economic activity is surprisingly strong. And a higher path for the federal funds rate through 2025 will likely be necessary to bring inflation back down to 2%. But with recent GDI growth considerably lower than GDP, that’s a risky bet on how well the economy is doing now.

In closing.

Being data-driven is no easy task.

The Fed is aware that macroeconomic data are subject to revision, and the staff understands the intricacies of the data. But no one has a crystal ball, and policymakers cannot wait for years of revisions. That’s one of many reasons that no one should get worked up about one month of data or even one quarter, and we should always use a wide range of data. The picture of the economy is always incomplete and noisy, especially now with all the disruptions due to the pandemic and the war in Ukraine.

Being data-driven takes more than data; it requires judgment on how to evaluate their quality and how to interpret them. It’s somewhat disconcerting to see the Fed lean hard on recent GDP data, which is one of the most incomplete measures in real-time and is currently out of step with its twin GDI. We will learn more on Thursday, but in all likelihood, it will take years for the dust to settle. We can look back then and see how good the Fed’s policy decisions were.

1. The comprehensive revision occurs every five years and brings in a whole host of new data, especially the quinquennial Economic Census, new Input-Output Tables, several sector-specific annual surveys, and other source data. At the same time, the BEA updates some of its methods to improve the quality.

2. Unlike PCE inflation, the only revision to the CPI is seasonal factors. As a result, the range of the CPI revisions is much narrower than PCE. However, these revisions can still reshape the narrative. New seasonal factors for motor vehicles wiped out what had looked like a clear step down in core CPI inflation late last year—other series, like the unemployment rate, also only revise annually with new seasonals. To use the data policymakers had at the time is why I developed the Sahm rule with real-time data.

3. GDP tracks all expenditures on final goods and services produced in the United States, whereas GDI tracks all income received by those who produced that output. Conceptually, the two should be equal because every dollar spent on a good or service (in GDP) must flow as income to a household, a firm, or the government (and, therefore, must show up in GDI). However, the two rely on different data sources and measurement issues, so discrepancies often arise.

4. Jason Furman was Chair of the Council of Economic Advisers at the time and initiated the issue brief. He is a long-standing proponent of using the average of GDP and GDI. You will see him often tweeting about it on GDP day. And he is in good company—the NBER recession dating committee relies on both GDP and GDI.