Which Estimate Counts When a 241 Billion Dollar Statistical Discrepancy Separates GDP From GDI

Two Estimates of One Quarter, and a 241 Billion Dollar Gap

On August 26, 2026, the Bureau of Economic Analysis published its second estimate of second-quarter output. Real gross domestic product increased at an annual rate of 1.5 percent. Real gross domestic income increased 2.2 percent. The average of the two, which BEA also publishes, increased 1.8 percent.

These are not three views of three different things. They are three arithmetic treatments of one quarter. GDP counts what was spent on final output. GDI, in BEA's glossary, is "the costs incurred and the incomes earned in the production of gross domestic product (GDP)"; BEA's primer on the accounts lists its four components as "compensation of employees," "taxes on production and imports less subsidies," "net operating surplus," and "consumption of fixed capital."

In the national accounting framework the two totals are definitionally equal. The difference between them has a name in BEA's own glossary, and the definition is exactly as blunt as it sounds. The statistical discrepancy is "the difference between gross domestic product (GDP) and gross domestic income (GDI)."

In current dollars for the second quarter of 2026, GDP was 32,486.066 billion dollars and GDI was 32,245.142 billion dollars. The discrepancy was 240.9 billion dollars at an annual rate. Dividing that gap by GDP gives 0.74 percent — a figure BEA does not publish and which is derived here from the two published levels.

A gap of that size sounds alarming. It should not. The useful question is not how large the discrepancy is in dollars, but what its change does to the growth rates that markets and policymakers actually read — and whether a one-quarter divergence of 0.7 percentage points carries any signal at all.

Why the Income Side Arrives a Month Behind the Spending Side

The advance estimate of second-quarter GDP, released on July 30, 2026, reported no GDI at all. That is not an oversight. It is the direct consequence of which records exist thirty days after a quarter closes and which do not.

The two sides do not draw on the same records. BEA attributes the resulting gap to "sampling errors, coverage differences, and timing differences," and emphasizes GDP in its releases because of the "more timely source data used to estimate quarterly GDP." Its glossary is blunter still: the two measures differ because their components are estimated using "largely independent and less-than-perfect source data."

Why the gap is recorded on the income side

BEA's primer on the accounts explains where the residual is parked, and the reasoning is a judgment about data quality rather than a neutral convention. "Because the source data used to develop the product-side estimates of the account are based on more comprehensive surveys and censuses, BEA considers them more reliable. Therefore, the statistical discrepancy appears as a component on the income side of the account to equate GDI with GDP."

That placement ranks the source data, not the two published growth rates. The discrepancy is displayed as a separate line on the income side of the account rather than folded into published GDI — in BEA's own summary account it sits between the two totals, as its own numbered line — which is why the two levels still differ by 240.9 billion, and why BEA is able to publish an equal-weighted average of the two at the same time.

The published tolerance, in BEA's own numbers

BEA also quantifies how closely the two track each other once both are available. In its published answer to why the measures differ, the agency reports that "the correlation between the rates of change for the final current quarterly estimates of GDP and GDI is 0.82," and that on annual data "the correlation between GDP and GDI is 0.97." The gap between those two coefficients is the whole story of this article compressed into two decimals. At quarterly frequency the measures disagree meaningfully. At annual frequency they very nearly do not.

A parallel clock in the labor statistics

The same pattern — a fast survey estimate that a slower administrative universe later corrects — governs payroll employment. Each year the Current Employment Statistics survey is benchmarked to the Quarterly Census of Employment and Wages, which BLS describes as "comprehensive counts of employment" that are "derived primarily from state unemployment insurance tax records." The most recent completed cycle is instructive on scale and on lag. Measured against the March 2025 reference month, "total nonfarm employment had a revision of -898,000 or -0.6 percent," and that correction reached the published series with the January 2026 data on February 11, 2026 — roughly eleven months after the month being corrected.

The state-level preliminary estimate for the next cycle was released on August 28, 2026, reporting an average absolute preliminary benchmark revision across all states and the District of Columbia of 0.4 percent for March 2026. Note what that figure is and is not: it is an average of absolute state-level revisions, not a national net revision, and the two can differ substantially when state errors offset.

The Gap Flips Sign More Often Than It Persists

A single quarter of divergence says almost nothing. Six quarters begin to say something. The chart below sets real GDP against real GDI for the six quarters from the first quarter of 2025 through the second quarter of 2026, using the growth rates as currently published.

Three numbers the agencies do not publish

The gap row on the chart is arithmetic applied to BEA's own figures. Three summary statistics follow from it, and none of them appear in any release.

• Mean absolute gap: 1.07 percentage points. Summing the six absolute gaps gives 6.4 points; dividing by six gives 1.0667. On a typical quarter the two measures disagree by roughly one full percentage point of annualized growth.
• Average disagreement: 0.07 percentage points. The simple average of the six quarterly GDP rates is 1.95 percent; the same average for GDI is 2.02 percent. These are averages of annualized quarterly rates, not compounded growth. Averaged across six quarters, the two ledgers describe the same economy to within seven hundredths of a point.
• Sign changes in four of five transitions. The gap runs negative, positive, positive, negative, positive, negative. It reverses direction in four of the five quarter-to-quarter transitions in the window.

The third statistic is the one that matters for interpretation. A gap that persisted in one direction would suggest a systematic problem in one ledger — a coverage gap in the income data, or a chronic mismeasurement on the expenditure side. A gap that flips sign nearly every quarter looks like noise around a shared trend, which is what the annual correlation of 0.97 implies.

Where the second-quarter divergence came from

Because the discrepancy is a level, a quarterly growth divergence is a statement about how that level moved. It stood at 291.5 billion dollars in the first quarter of 2026 and 240.9 billion in the second, a decline of 50.6 billion dollars computed here from the published levels. As a share of GDP the discrepancy fell from 0.91 percent to 0.74 percent.

GDI grew faster than GDP in the second quarter because that level fell; the two statements are the same statement, not a cause and an effect. What moved the level is a separate question, and the profits component — up 400.9 billion dollars in the quarter against 74.4 billion in the first — sits on the income side and arrives late.

What the Discrepancy Is Worth for Dating a Downturn

The most consequential audience for the GDP-versus-GDI question is not a trading desk. It is the committee that dates U.S. business cycles, which treats the two measures as peers rather than as a primary and a footnote.

In its published account of its own procedure, the committee describes the two as "[t]wo measures that are very important in the determination of quarterly peaks and troughs, but that are not available monthly." On weighting it is explicit: "we give equal weight to real GDI." And on when the gap has mattered most, the committee notes that the statistical discrepancy "was particularly important in the recessions of 2001 and 2007-2009."

The committee does not say which measure moved first, and it does not have to. Its stated procedure is narrower: the gap between the two mattered to the dating decision in those two episodes, and the committee "allows sufficient time for standard data revisions in order to assign an accurate peak or trough date." Two episodes is a thin base rate, and it supports only a weak claim — that the gap has at least twice carried information a single-measure reading would have missed. It establishes nothing about which side leads.

For a policy path the implication is narrower than it first appears. A central bank reading a quarter at 1.5 percent versus 2.2 percent is not choosing between expansion and contraction. Both figures sit inside a range that supports the same qualitative description of demand. The divergence changes the confidence interval around the reading, not the sign.

Where This Doesn't Apply

Several conditions void the framework above.

• Level questions, not growth questions. The 0.74 percent ratio says the two ledgers agree closely on the size of the economy, not that either level is correct. Both could share a bias no comparison between them can detect.
• Windows shorter than six quarters. The mean absolute gap of 1.07 points and the sign-flip count are computed over one specific six-quarter window. A different window produces different values, and a six-observation sample supports no confidence statement at all.
• A high correlation is not independent confirmation. GDP and GDI are two estimates of one quantity produced by one agency. The 0.97 figure measures how closely BEA's estimates track each other, not whether either is accurate.
• Averaging is not free. BLS uses GDP, not the average of GDP and GDI, for its official nonfarm business productivity statistics. Its January 2026 review of the three output measures found that the average had lower revisions than either measure alone during the 2000s and 2010s, that this advantage narrowed in the 2020s, and that averaging forecloses the component decomposition that only GDP and GDI separately allow.
• Turning points are the exception the base rate does not cover. The claim that the gap is noise rests on a sample dominated by expansion quarters. In 2001 and 2007-2009 the gap carried information. A framework that treats it as noise will fail precisely when the question is most expensive.

What to Watch Next Week

Three releases land this week, already on the calendar. JOLTS for July 2026 arrives Tuesday, September 1; BLS states that "[t]he JOLTS employment levels are ratio-adjusted to the CES employment levels," so JOLTS inherits whatever the annual CES benchmark later does to those levels rather than checking them independently. Revised Productivity and Costs for the second quarter arrives Thursday, September 3, and will absorb the second-estimate GDP figure of 1.5 percent, because BLS builds its official productivity statistics on GDP rather than on the average — the 2.2 percent income-side reading stays outside that calculation. The August Employment Situation arrives Friday, September 4.

The week of September 7 is where the price side of the same accounts arrives.

• Employer Costs for Employee Compensation, June 2026 — Wednesday, September 9. Compensation of employees is the first of the four components BEA's primer lists on the income side. This release is a separate survey on a different reference period, but it is the reading that lines up most directly with that component between GDI vintages.
• Producer Price Index, August 2026 — Thursday, September 10, and Consumer Price Index, August 2026 — Friday, September 11. Read these for the price side, not for the discrepancy. The statistical discrepancy is defined on current-dollar GDP and GDI, so a change in the deflator moves both real growth rates together and leaves the dollar gap between the two ledgers untouched.
• Beyond the four-week horizon: the third estimate of second-quarter GDP on September 30. If the discrepancy narrows further, published GDI growth falls toward GDP. If it widens, the reverse. Watching the discrepancy level directly is more informative than watching either growth rate alone.

Concrete Framework

A monitoring routine that treats the two ledgers as peers, in order.

1. Record the discrepancy as a level and a ratio each quarter, not as a growth-rate difference. Take current-dollar GDP minus current-dollar GDI, then divide by GDP. For 2026 Q2 that is 32,486.066 minus 32,245.142, equal to 240.9 billion, or 0.74 percent. Use the unrounded levels; the rounded ones do not reproduce the answer.
2. Track the change in that ratio, not its level. Any quarter in which GDI grows faster than GDP is a quarter in which the discrepancy fell, and the size of the fall is the arithmetic equivalent of the growth-rate gap.
3. Keep a running sign column. Log whether the gap is positive or negative each quarter. Persistence in one direction across four or more quarters is the signal that one ledger has developed a systematic problem; alternation is the null case.
4. Never compare a first-vintage GDP quarter with a second-vintage GDI quarter. The advance estimate has no GDI, as the July 30 release for the second quarter shows. Any comparison built on an advance release is comparing one measure with nothing.
5. Apply the annual filter before drawing a conclusion. The quarterly correlation of 0.82 and the annual correlation of 0.97 are BEA's own published figures. A disagreement that survives four quarters of averaging is worth investigating; one that does not, is not.
6. Suspend the noise assumption near a suspected turning point. The committee that dates cycles gives GDI equal weight and cites the discrepancy as particularly important in 2001 and 2007-2009. That is when the gap deserves the most attention and gets the least.
7. Log the correction lag alongside every survey-based series being tracked. For payrolls it is roughly eleven months from reference month to benchmark incorporation. A series whose eventual correction exceeds the monthly change being interpreted cannot support the interpretation.

The honest summary of the second quarter of 2026 is that output grew somewhere between 1.5 and 2.2 percent, that BEA's own midpoint of 1.8 percent is a reasonable single number, and that the width of that range is ordinary rather than alarming. Which estimate counts is the wrong question. The range is the estimate.