Securities in Bank Credit Grew 9.7 Percent in July. The Next Three Wednesdays Fell.

The Federal Reserve's H.8 release dated September 4, 2026 shows securities in bank credit at commercial banks growing at a 9.7 percent seasonally adjusted annual rate in July. That is the highest of the four monthly readings the table prints for 2026 and more than three times its second-quarter pace of 3.0 percent. Two columns to the right, the same series is lower on each of the last three Wednesdays the release publishes.

Both numbers are correct. They are not measurements of the same thing, and the release says so in its own column headings.

What the September 4 release actually prints

Table 1 is labeled "Percent change at break adjusted, seasonally adjusted, annual rate." Table 2 is labeled "Seasonally adjusted, billions of dollars." Their columns do not line up: Table 1 ends with a month, July 2026, Table 2 with four Wednesdays, August 5, 12, 19 and 26. Every figure below comes from the Federal Reserve Board's H.8 release of September 4, 2026.

For securities in bank credit, Table 2 prints monthly levels of 5,682.6 billion dollars in January, 5,721.2 in February, 5,743.1 in March, 5,733.7 in April, 5,756.3 in May, 5,786.0 in June and 5,832.7 in July, then Wednesdays: 5,830.0, 5,813.2, 5,811.7 and 5,796.3 on August 5, 12, 19 and 26. July is the highest figure in the row and every August observation sits below it. The fall from August 5 to August 26 is 33.7 billion dollars, or 36.4 billion from the July monthly level.

Table 1 prints -2.0 percent for April, 4.7 for May, 6.2 for June and 9.7 for July. Read down the rate column and the series accelerates; read across the level row and it turned over roughly six weeks ago. Neither row has been revised; both were published on the same page on the same day.

A percent change and a level are not the same measurement

The Federal Reserve Board's H.8 About page states: "Weekly levels represent estimated aggregates for selected bank asset and liability items as of close-of-business Wednesday. Monthly levels are pro rata averages of weekly data."

A monthly level is not an end-of-month reading but an average of the Wednesdays inside the month, so a month-over-month change is an average against an average, referenced near mid-month rather than at the end. July's 9.7 percent compares a July average centered near mid-July with a June average centered near mid-June. On the day it prints it describes a midpoint seven weeks back, and the newest observation in the release is nine days old.

Securities rose steeply into July, lifting the July average above June's. The level then turned. That turn appears at once in the weekly column and cannot reach the rate column until the August monthly rate is computed.

Most of that August rate is already determined by numbers the release has printed. The simple average of the four published August Wednesdays is 5,812.8 billion dollars, 19.9 billion below the July monthly level of 5,832.7, or about -4.0 percent annualized. The Board's figure will differ, because monthly levels are pro rata rather than simple averages and weekly data are revised, but a series printing +9.7 percent today is tracking toward roughly -4 percent for the month after.

The same table runs the reversal the other way

Commercial and industrial loans do the mirror image, which rules out a story about one row behaving oddly. Table 1 prints 15.7 percent for April, 12.0 for May, 6.9 for June and -2.4 for July, the only negative monthly reading in the row.

Table 2 prints the C&I level at 2,905.1 billion dollars in June and 2,899.3 in July, and that 5.8 billion decline produces the -2.4 percent. The Wednesdays after read 2,925.2, 2,933.9, 2,945.7 and 2,956.9: up in every published August week, 31.7 billion across three weeks and 57.6 billion above the July monthly level. Their average, 2,940.4 billion, annualizes to roughly 18 percent against July.

So both components of bank credit reversed, in opposite directions, in the four weeks after the month whose rates are on the page. Bank credit, their sum, moves far less: 19,735.6 billion in July against an August Wednesday average of 19,809.2, about 4.6 percent annualized versus the printed 5.5. The aggregate conceals both reversals, which is why composition rows matter more than the headline here.

Why the printed rate and the printed levels do not reconcile exactly

A second gap survives even after the timing problem is handled. Annualize the published June and July levels for securities, 5,786.0 to 5,832.7, and the answer is 10.1 percent. Table 1 prints 9.7. Doing that for every row, not the row you came for, is what makes it legible.

Table 1 is break adjusted; Table 2 is not. The About page sets out what that removes and, in the next sentence, how: "All percent changes have been adjusted to remove the effects of nonbank structure activity of $5 billion or more, as well as the effects of accounting rule changes, such as the initial consolidation of off-balance-sheet vehicles (FAS 166/167) and the adoption of the Current Expected Credit Loss (CECL) standard (ASU 2016-13). These adjustments use a ratio procedure like that used to account for panel shifts, designed to make past levels comparable with current levels."

The second sentence sets the order of operations, and it reverses the intuitive one. The adjustment is not applied to a growth rate after the rate is computed. It is applied to past levels, scaling them to be comparable with current ones, and the rate is taken from the adjusted series. Hence percent change at break adjusted rather than break adjusted percent change: the adjustment sits upstream of the arithmetic, on a level path Table 2 does not print.

Recomputed that way, the July 2026 gaps run as follows, in points of annualized growth.

• Treasury and agency securities - levels imply 15.12 percent, the table prints 14.2. Gap 0.92.
• All other loans and leases - levels imply 11.03, the table prints 10.5. Gap 0.53.
• Other securities - levels imply -10.59, the table prints -11.1. Gap 0.51.
• Securities in bank credit - levels imply 10.13, the table prints 9.7. Gap 0.43.
• Consumer loans - levels imply 7.17, the table prints 6.9. Gap 0.27.
• Total assets 0.21, bank credit 0.17, deposits 0.06, loans and leases 0.05, C&I loans 0.03, real estate loans -0.01.

The gaps are spread across the release and the largest is not in securities. What matters more than the ranking is the relation between a component and the aggregate above it. Loans and leases agrees with its levels to within 0.05 of a point, which would ordinarily read as no adjustment there. Yet two of its four components carry gaps that dwarf it: all other loans and leases at 0.53, about eleven times the aggregate, and consumer loans at 0.27, about five times, while C&I and real estate carry none to speak of. Securities does it in reverse: its 0.43 is smaller than either component beneath it, 0.92 and 0.51.

This is the aggregation effect from earlier, arriving a second time. An aggregate matching its own levels is not evidence that no adjustment touched it, only that whatever touched its components did not survive addition. The release does not publish which adjustment produced which gap: the About page names the categories removed and the procedure used, and attaches neither to a row.

Adjustments of this kind are disclosed on the Federal Reserve Board's H.8 notes on the data, entry by dated entry: definitional changes and structure events that move balances between rows and put a step in the level series without any lending having been made or unmade. An entry dated to one month says nothing about another month's gap.

Applying that check to July returns nothing. The three most recent entries on the notes page are dated April 1, 2026, January 7, 2026 and December 31, 2025; none falls in June, July, August or September 2026. The July gaps are not accounted for by anything the page discloses, and this article does not close them.

What is being measured, and by whom

The About page describes the sample: "The H.8 estimates are primarily based on the FR 2644 reporting form-reported weekly by a voluntary authorized panel of 850 domestically chartered banks and foreign-related institutions-and quarterly Consolidated Reports of Condition and Income (Call Reports) for those commercial banks that are not included in the FR 2644 panel." The Federal Reserve Board's FR 2644 reporting form page adds the qualification that matters: participation is voluntary, the frequency is weekly as of the close of business each Wednesday, and "Smaller banks (those with assets less than $5 billion) have an option to report monthly."

Weekly movement is observed only for the part of the panel reporting weekly, and estimated both for smaller banks taking the monthly option and for the universe outside the panel, so a weekly change is a partial observation scaled up rather than a census. And the scaling is itself revised: the Federal Reserve Board's H.8 technical Q&As describe benchmarking, in which the weights applied to weekly releases since the previous Call Report are updated once new Call Report data become available.

Where This Doesn't Apply

This reads one release. It does not forecast bank behaviour.

• Three weekly observations are a small sample. The August reversal rests on three consecutive declines in a series that routinely moves several billion dollars a week. One large institution restructuring a portfolio can produce that pattern without saying anything about the system.
• The simple average used here is not the Board's method. Monthly levels are pro rata averages, so weeks straddling a boundary are split. The August estimates of about -4 percent for securities and 18 percent for C&I loans will shift.
• The break-adjustment finding is a July finding. It shows several rows in which the printed rate and the printed levels differ, by 0.92 of a point at the widest. It establishes no standing bias, and does not identify what was adjusted.
• Weekly data are revised. Benchmarking can move the August Wednesdays before the August monthly rate is ever printed, changing the arithmetic without anyone having read it wrong.
• None of this touches the twelve-month picture. Securities in bank credit are 4.4 percent above their July 2025 level and C&I loans 8.7 percent above theirs.

What to Watch Next Week

The H.8 is a Friday release carrying data through the Wednesday nine days earlier, so the next edition on that cadence covers the week ending September 2.

• Whether securities hold below 5,800 billion dollars. August 26 is 5,796.3; a September 2 print back above 5,830 would make the August decline a level step rather than a trend.
• Whether C&I loans hold above 2,950 billion. August 26 is 2,956.9, after three consecutive weekly increases.
• The August monthly rates when they appear. Securities near -4 percent and C&I near 18 percent are what the published Wednesdays imply. A materially different print means the weeklies were revised or a break adjustment was applied.
• A new entry on the notes page dated August or September. Thrift conversions and reclassifications are disclosed there.

Three branches are worth separating in advance. In the first, the September Wednesdays continue in the August direction, making the July rates stale and the shift from securities into loans real. In the second, levels revert toward their July values, making August a one-off portfolio adjustment and leaving July the better signal. In the third, the August weeklies are themselves revised at benchmarking, in which case neither reading measured what it appeared to, and the branch is resolved only when the revised series is published.

Concrete Framework

1. Read the units line above each table before the numbers. On the H.8 the two lines differ in one phrase, break adjusted.
2. Establish what the most recent column is. A month and a Wednesday are not comparable, and a table's newest column is not necessarily the release's newest information.
3. Recompute the printed rate from the printed levels for every row, not the row you came for. Annualize the month-over-month change and compare. One row tells you nothing about the others.
4. Compare each component with the aggregate above it. An aggregate matching its levels can sit on components that do not. Agreement at the top is not evidence of agreement underneath.
5. Extend the level path past the last monthly column. Average the published weeks that follow and annualize. That approximates next month's rate, weeks before it prints.
6. Check the notes page for the months in question before treating a level step as economic activity; record it when the page says nothing.
7. Mark the reading provisional until benchmarking. Current-quarter weights update when the next Call Report arrives.