Corporate Profit Share — Methodology

The construction

Corporate profits after tax, with inventory valuation and capital consumption adjustments (BEA's CPATAX), divided by nominal GDP. Both series are raw inputs computed into the ratio here — this feature consumes no other feature's labels, tiers, or scores. The "with adjustments" concept is deliberate: it measures profits from current production, and its revision behavior differs materially from the unadjusted variant in exactly the stress quarters this page cares about.

The trend lens

Position is measured as distance from the share's own expanding log-linear trend, in units of that fit's residual scatter (sigma). At each quarter an ordinary least-squares line is fit through the natural log of every reading up to and including that quarter, and the reading's distance from the line is divided by the fit's standard error. A quarter is eligible once ten years of prior history exist (40 quarters); the first 24 eligible readings are treated as a stabilization window and excluded from historical summaries. The same mechanics power the CAPE and Buffett sigma-vs-trend indicators; the ±2σ anchor follows Grantham's two-sigma-from-trend definition. The full-history percentile is deliberately demoted to context: the share has sat above its 80th percentile for 88 of the last 91 quarters — an expanding percentile on this series describes a regime, not a signal.

The five-level ladder

Readings map to five levels at ±1σ and ±2σ — the site's standard ladder and colors. Before adoption the ladder passed two pre-registered tests on this series' own history. The first was occupancy: no single level may hold a majority of quarters. The widest, neutral, holds 49.2% of the 254 post-stabilization quarters; pale red 29.5%, deep red 9.4%, pale green 9.1%, deep green 2.8%. The second was outcome differentiation: adjacent levels' forward S&P distributions were compared side by side before the colors were kept. Colors state distance from trend, not a return forecast; what the register says about what followed is in the section above.

What the chart shows — and what an observer would have seen

The page's sigma series runs on today's revised data, labeled as such. That is not what a reader would have seen at the time. Computing the same lens on only the data knowable at each date (true vintages begin December 1991) can tell a different story. The largest divergence in the record is exactly where the mean-reversion claim lived. Through 1998–2000 the reading as perceived then ran from +1.1σ to +2.1σ. Today's revised series puts 9 of those 12 quarters near or below trend — a maximum gap of 1.74σ at 2000Q4 (the median absolute gap across all 140 vintage-true quarters is 0.32σ):

An honest observer of the late 1990s saw margins well above trend; today's data puts every one of them closer to it. Revisions rewrote the story after the fact — which is why this page carries a trust dial and a revision panel, and why the tested claims below were scored on both bases.

What followed — the test, and what it found

The question. Across the whole record, does where the profit share sits on its own ladder say anything about the year that follows — and if some periods ended badly, does anything on this site distinguish them from the rest? The test was run on 56 periods since 1963, of which 55 have a twelve-month outcome (the newest is still open). Forward figures are S&P total return from the month each period's entry quarter was published — when its profits printed, two to four months after the quarter's end. So every window opens when a reader could first have known the level changed.

"Resolved badly" was defined before anything was computed — the next twelve months negative. Five other definitions were counted first and are recorded with the test. They were: below −10% (2 periods), the next three years negative (6 of 55), below the register's own median (27), the worst fifth (11), and the period ending by moving to a redder level (28 of 55). The last is the trap and was rejected: it counts the instrument moving, not the market falling, and those periods' twelve months returned a median of +15%.

What the record says. 6 of 55 periods were followed by a losing year — 11% — against 22% for an arbitrary quarter end over the same years. A level change has been followed by a losing year less often than an ordinary quarter has. The six are five distinct episodes in sixty years (1969, 1983, 1987, 2007–08, 2021; the 2007 and 2008 periods are one housing-led recession), and they are spread across the ladder — one began stretched below trend, three near trend, two stretched above, and none far above.

Almost nothing marks them. 18 conditions were tested against a bar fixed before any of them was computed. The bar required both groups to hold at least five periods, at least three of the bad periods to fall in the flagged group, and the flagged group's rate to be at least twice the unflagged group's. They were: the four-stage regime (three ways), the inflation overlay (two), the stress signal, Forward Stress (two thresholds), Economic Breadth (two) and Housing (two). Also tested were the three valuation lenses and their union, the profit level itself, and the period's own length. Of the eighteen, seventeen fail, and 12 of the 18 point the wrong way — their flagged group did better than the group they left out. Among them are the red levels themselves (9.1% against 12.1%), a strained inflation overlay (8.3% against 13.8%) and an expensive CAPE (6.2% against 12.8%). A split of twelve one way and six the other is what no information looks like.

One condition did stand out. When the housing cycle had already turned past its expansion, 4 of 19 periods ended in a losing year, against 1 of 23 otherwise — across three separate episodes (1987, 2007–08 and 2021). It was one of eighteen conditions tested, so read it as a lead worth watching, not a rule. The profit share's own level still says nothing about what follows. Both bases are recorded. On the revised record, each period read at its entry quarter's end, the same condition flagged 3 of 19 against 2 of 27 (the narrower family, downturn spreading or contraction, 3 of 15 against 2 of 31). That was the bar's minimum, and it failed when any single case was removed. That was measured on September 29, 2026, on the 59 completed periods then in the register; the Housing record it read is no longer kept, so it is not re-measured. On the as-known basis — every period read at the month its profits were published, the housing cycle as it had been published by then — 4 of 19 against 1 of 23, and it holds when any single case is removed. The as-known basis is the site's standard, ruled before this result was seen (PUBLAG-1 ruling 1 and PUBLAG-2, September 28–29, 2026).

The coverage limit, stated. The engines do not reach the whole record. The four-stage regime and Forward Stress speak to 41 of the 55 resolved periods (their records begin 1981), and Housing to 42 (1979). Economic Breadth, CAPE and the allocation lens reach all 55, and the Buffett indicator 26 (its record begins 1999). The register's worst year — 2008 Q2, at −18.8% — is inside every engine's record. Roughly a quarter of the record predates the instruments being asked about it.

Measured October 2026 over 1963–2025, on every period in the register. The other engines were asked only what they read at each date: none of them sets a level, opens or closes a period, or changes what is in the register. This page's arithmetic uses corporate profits, GDP and the S&P, and nothing else.

What followed each level

Colors state distance from trend, not a return forecast. The share has not read below −1σ since 1991 — the green rows are pre-1991 evidence. Forward figures are S&P total return · Shiller monthly convention, from the month each period's entry quarter was published; the medians count only periods whose outcome is complete.

The shape of what followed — today's level

Each dot is one completed period at this level. Red dots ended lower than they began.

The period register

A period is consecutive quarters at one level: it opens the quarter the reading's level differs from the prior quarter's, and re-entry requires leaving first. Forward figures are S&P total return · Shiller monthly convention — dividends reinvested monthly (see "The total-return basis" on the Confluence methodology). Returns run from the month the entry quarter's profits were published — two to four months after the quarter's end, the first month a reader could have known the level changed. So the register reads as it could have been read at the time. It is conditional history, not a tradable signal.

56 periods since 1963, oldest first. "open" marks the period standing now, which has no outcome yet.

Validation

This is the Corporate Profit Share's validation record, kept with its methodology. The Validation page sets it beside every other feature's.

The tested claims

Margin mean reversion — WATCHABLE. The claim is that margins far above their own history come back toward it. Before the test ran, in 1991's terms, we wrote down what would count as coming back and what would count as far above — so the answer could not be chosen after the fact. There have been four episodes since 1991. The three post-2000 episodes never came back at any horizon, on either the data we have today or the data anyone had at the time. The single pre-2000 episode did, at long horizons, on today's data only. Four episodes can neither validate nor kill a claim — it stands WATCHABLE, and the record grows as new episodes complete.

Early recoveries are under-reported — VALIDATED, n=3. All three qualifying recoveries with true-vintage coverage (2001, 2009, 2020) were first reported weaker than they proved — episode revision medians of +1.09, +1.57 and +1.27 points of GDP share against an all-quarter median of +0.85. Three episodes is a small base; the claim is stated only that strongly, wherever it appears.

Revisions, September, and the vintage floor

Corporate profits are absent from BEA's advance GDP estimate: a quarter first prints with the second estimate (roughly two months after quarter end; three for fourth quarters). Each late September the annual update rewrites the level history — the 2008 profit record was still being rewritten in September 2023. This page's response is disclosure, not pretense: the revision panel's left anchors are frozen first prints while its right anchor is living. Each September update is followed by a disclosed re-baseline of the sigma series and its register, documented each year. Everything labeled point-in-time uses only vintages available at that date. True vintages for the ratio begin December 1991 — the earlier history is shown on today's basis and never enters a point-in-time computation.

Why this number — where the share comes from, and how its level is named

The share of GDP on the current-read card is the Kalecki–Levy identity's eight signed legs, summed — CPATAX = A557 + A889 + DIVIDEND − PSAVE − FGDEF − SLDEF + NETFI − (GDP − GDI), each leg as a share of GDP — and the identity closes at $0.0B in every one of the 318 quarters since 1947. So the card's expander prints columns that are exact by construction. The share of GDP column (each leg's level) sums to the share, and the this quarter column (each leg's level now minus last quarter's) sums to the share's change. The four quarters column does likewise, because the page already speaks in that horizon. There is no residual. The statistical discrepancy is one of the eight legs — a real, small term BEA publishes — shown as a named row, never a plug. The only difference from a printed figure is rounding: the card prints the share at 0.01, and BEA publishes each leg at $0.1B. The "why" line under the share is the this-quarter column in words: the largest movers in the change's direction, the largest against it, the rest as one net. It is an ordering of the arithmetic, not a ranking of what matters.

The read is rule-based, and the page prints it as one. The margins-vs-trend word is the five-level house ladder over σ — the margin's distance from its expanding log-linear trend in the fit's residual sigmas. The direction word compares σ now with σ four quarters ago (rising, falling or flat). The table below prints that rule. The page used to carry this rule live, under the three cards, with the distance from each line that would change the reading. It does not any more: the test above found that the level a period starts at makes no difference to the year that follows, so a line explaining how the level was named was explaining a classification with no consequence.

+2.0 exactly is deep red, +1.0 exactly pale red, −1.0 exactly pale green, −2.0 exactly deep green (the house ladder's boundary semantics). The numbers on the why line and in the table wear their sign on the site's cell tokens — light below the bar, deep at or beyond it. The bar is 0.4 pp for a quarter's move and 0.9 pp for four quarters (the 75th percentile of the absolute moves of the eight legs over every quarter since 1947), and one point of GDP for a level.