Macro ReviewHow to read
Cross Asset · Fundamentals · Earnings

Index EPS grew +18.0% in the year to Q2 2026, behind national-accounts profits

What does the index earn, and what supports it?

BEA · 4 Nov 2026 · data to Q2 2026Shiller earnings to Q3 2026 · FactSet quoted 11 Sept 2026Company filings to Q2 2026Updated 17 Nov 2026
Trailing EPS, reported$295.52+25.6% on a year earlier · Q3 2026
Consensus EPS growth, 2027+15.1%Calendar 2026 +31.6% · FactSet, 11 Sep 2026
After-tax profit share of GDP11.9% 9.8% · Q2 2026 · highest since 1990
EPS growth expected for 2027+4.7%+1.7%the site's models+5.8%default model's median · its record: no better than a fixed mean · Q2 2026no edge in its record

In brief

What the numbers say, and what would change the reading
  • Trailing reported EPS is $295.52 for the year to Q3 2026, +25.6% on a year earlier.→ Exhibit 1
  • Index EPS grew +18.0% in the year to Q2 2026, against +18.2% for national-accounts profits after tax.→ Exhibit 6
  • After-tax profits are 11.9% of GDP, the highest since 1990, against a of 9.8%.→ Exhibit 7
  • EPS trailed profits by 0.1 points in the year to Q2 2026; the 1990–2019 average was −0.9.→ Exhibit 8
  • FactSet's consensus has calendar-2026 EPS up +31.6% on revenue up +12.1%, so the margin term carries most of the growth.→ Exhibit 3
What would change this view
What would change this viewTriggerTodayDecided by
Profits stop taking a record share of the economyafter-tax profit share ≤ 9.8% of GDP (its 2015–19 average, the Profits page's test) for two quarters11.91% (Q2 2026)not crossedBEA GDP and profits release, 25 Nov 2026
Index EPS stops trailing national-accounts profitsEPS growth less after-tax profits growth > −0.1pp (half today's gap)−0.2pp (Q2 2026)not crossedBEA GDP and profits release, 25 Nov 2026; Shiller's quarterly earnings
Calendar-2027 EPS growth lands outside the fanreported EPS growth for calendar 2027 outside −9% to +22% (the fan's 10–90 band)not yet reported (earnings to Q3 2026)pendingShiller's earnings for Q4 2027
Consensus moves halfway to the modelFactSet calendar-2027 EPS growth < 9.9% (halfway from +15.1% to the model's +4.7%)+15.1% (11 Sep 2026)not crossedthe next FactSet Earnings Insight quote
The reasoning behind each point
  • Where EPS stands
    Trailing reported EPS is $295.52 for the year to Q3 2026, +25.6% on a year earlier.

    Shiller's series (reported, GAAP earnings at quarter-end months): $235.25 a year earlier. FactSet's bottom-up (operating) estimates, as quoted on 11 Sep 2026: calendar 2026 $361.38 (31 Aug 2026; $340.49 earlier), Q3 2026 $89.69, revised +1.7% since 30 June; growth +28.7% in Q3, +26.3% in Q4, +31.6% for 2026 and +15.1% for 2027. The two measures differ by construction: reported earnings carry write-downs and one-offs that operating estimates leave out. The season's beat rate is not in the FactSet report this page quotes. Shiller's workbook dated 16 Nov 2026 restated Q4 2025 to Q2 2026: Q2 2026 is $262.59, where this page had printed $295.39.

  • National accounts
    Index EPS grew +18.0% in the year to Q2 2026, against +18.2% for national-accounts profits after tax.

    Pre-tax profits with inventory and capital-consumption adjustments +20.8%; domestic nonfinancial +21.1% and domestic financial +23.1%. Index EPS is running 0.2pp below the economy-wide measure: the index is not the economy — a few sectors' margins, buybacks and the reported-versus-operating gap all sit in that difference. Shiller's earnings run a quarter ahead, to Q3 2026 (+25.6% on a year earlier). BEA publishes profits for that quarter on 25 Nov 2026, so the comparison here is made in Q2 2026.

  • Margins
    After-tax profits are 11.9% of GDP, the highest since 1990, against a 2015–19 average of 9.8%.

    The margin gap (the share less its average over the last 20 quarters, 10.8%) is +1.11pp. Domestic nonfinancial profits are 9.8% of GDP against 6.7% in 2015–19; the pre-tax share is 14.5%. A share this far above its average is the earnings level the index is priced on; the question is whether it persists, not whether it grows. The Corporate profits and margins page reads the same 11.91% for Q2 2026, and its test — the share back at 9.8% for two quarters — is the one threshold this page uses.

  • Bridge
    EPS trailed profits by 0.1 points in the year to Q2 2026; the 1990–2019 average was −0.9.

    The fan is carried from profits to EPS by that average gap in four-quarter log growth, −0.92 points a year over 120 quarters. Its standard deviation is 47.9: the bridge moves the fan, it does not widen it. Last year's gap of −0.1 is 0.0 standard deviations above that average: the gap swings widely from year to year, and the fan carries its average, not its swings.

  • What drives it
    FactSet's consensus has calendar-2026 EPS up +31.6% on revenue up +12.1%, so the margin term carries most of the growth.

    Read as (1 + EPS growth) ÷ (1 + revenue growth) − 1, the margin term is +17.4% in 2026 and +5.5% in 2027 (revenue +9.1%, EPS +15.1%); it is arithmetic on the quotes, not a finding. The quoted net margin is 14.9% for Q3 2026 against 13.0% a year earlier and a five-year average of 12.4%. By sector, Q3 earnings growth runs from +2.5% to +105.7% across the 7 sectors quoted; 42 companies guided Q3 EPS down and 72 up. The share count is in none of the sources this page reads, so buybacks sit inside the margin term.

  • Terminal growth
    Beyond the fan's eight quarters the valuation assumes terminal growth of 4.0%, the Committee's long-run nominal GDP.

    The Committee's longer-run medians added: real GDP 2.0 + PCE inflation 2.0 (projection table of 16 Sep 2026). Over its 8 quarters the fan reverts to 6.20, which adds the 1990–2019 average gap between profits and nominal GDP (2.20); on the Valuation page that gap fades to zero over 6 years, because a gap over GDP held for ever implies a profit share rising without limit.

  • Income mix
    Per-share earnings outside finance grew 31.1 log points in the year to Q2 2026, revenue adding 11.2.

    In simple terms that is +36.5%. The margin added +19.3 log points and the falling share count +0.7; the parts add to the whole by construction. Across every company, per-share earnings grew 30.0 log points (+35.0%). These are the companies' own filings, summed; the index's EPS is S&P's and differs by float and membership.

  • Per share
    Outside finance, the share count added 0.7 log points to per-share earnings in the year to Q2 2026.

    Measured as a holder who keeps a fixed fraction of every company: the change in that holder's earnings less the change in the companies' total earnings. Net buybacks raise it; issuance for acquisitions and employee pay lowers it. Across every company, the share-count effect is the one Payout prints.

  • Taxes
    In the filings taxes take 19.2% of pretax income; the national accounts put it at 17.7%.

    The two differ in coverage (every corporation in the national accounts; the index's companies in the filings) and in what counts as profit (the national accounts adjust for inventory valuation and depreciation).

Next: Payout & reinvestment — where the earnings go

Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve Board; EIA; FactSet Earnings Insight; U.S. Securities and Exchange Commission (EDGAR); Macro Review calculations · data to Q2 2026cross.earnings-read

What does the index earn, reported and as quoted?

Trailing per share in dollars since 1980, and the operating earnings and growth the consensus quotes, read beside each other and never spliced

Exhibit 1 · S&P 500 trailing reported earnings per share, since 1980

S&P 500 trailing twelve-month reported EPS, dollars, quarterly since 1980; data to Q3 2026.
About
About this exhibit
Line of S&P 500 trailing twelve-month reported EPS in dollars, quarterly since 1980.
Earnings per share, reported: the level in dollars
S&P 500 trailing 12-month reported EPS, dollars, quarterly since 1980; data to Q3 2026.
Q3 2026: $295.52, +25.6% on Q3 2025 ($235.25). earnings carry write-downs and one-offs that the operating (bottom-up) estimates FactSet quotes leave out; earnings levels are index data of S&P Dow Jones Indices, shown with attribution.
Source: Robert J. Shiller · data to Q3 2026cross.earnings-21

Exhibit 2 · What the index is expected to earn, as FactSet quotes it

FactSet Earnings Insight, 11 Sept 2026: operating (bottom-up) earnings and growth, read beside Shiller's reported level, never spliced to it.
Calendar-2026 EPS, bottom-up31 Aug 2026 · earlier estimate $340.49$361.38
Q3 2026 EPS, bottom-up31 Aug 2026 · estimate revised +1.7% since 30 Jun$89.69
EPS growth, Q3 2026year over year, blended+28.7%
EPS growth, Q4 2026year over year, estimated+26.3%
EPS growth, calendar 2026bottom-up+31.6%
EPS growth, calendar 2027bottom-up — the near-term rate Valuation prices+15.1%
Forward 12-month P/Etrailing 25.9× · Valuation carries the multiple with SPY's price19.1×

Not quoted: beat rate and aggregate surprise — the capture does not carry the season's beat rate; the author's weekly transcription would.

Source: FactSet Earnings Insight · data to 11 Sept 2026cross.earnings-22

What is driving EPS?

Sales, margins and sectors as FactSet quotes them; the share count is in none of the sources, so buybacks sit inside the margin term

Exhibit 3 · Consensus growth by period and sector in revenue, EPS and margin

Revenue growth beside EPS growth, %, the margin's contribution, net margins and sector growth for the quarter being reported; FactSet Earnings Insight, 11 Sep 2026.
PeriodRevenue growthEPS growthMargin's contribution
Q3 2026+11.9%+28.7%+15.0%
Calendar 2026+12.1%+31.6%+17.4%
Calendar 2027+9.1%+15.1%+5.5%
Net marginQ3 2026Q2 2026Q3 20255-year average
S&P 500, as quoted14.9%17.0%13.0%12.4%

Guidance for Q3 2026: 42 companies guided EPS down, 72 up. Not quoted: share count and buybacks — none of the sources this page reads carries the index share count; the margin's contribution is read as EPS growth less revenue growth.

Sector, Q3 2026 earnings growth (year over year)Growth
Energy+105.7%
Information Technology+63.0%
Communication Services+50.8%
Materials+30.9%
Health Care+6.0%
Financials+3.5%
Consumer Staples+2.5%

The 7 sectors the report quotes; the others are not in the capture. Descriptions of what the consensus expects, not the site's view.

About this exhibit
Tables of FactSet consensus revenue growth, EPS growth and the margin's contribution by quarter and calendar year, quoted net margins, and earnings growth by sector for the quarter being reported, with guidance counts.
Sales, margins and sectors, as FactSet quotes them
Revenue growth beside EPS growth, the margin term, net margins and Q3 growth by sector; FactSet Earnings Insight, 11 Sept 2026.
The margin term is (1 + EPS growth) ÷ (1 + revenue growth) − 1: what is left of EPS growth once sales are taken out, arithmetic on the quotes. The share count is in none of the sources this page reads, so buybacks sit inside that term.
Source: FactSet Earnings Insight · data to 11 Sept 2026cross.earnings-23

What are the earnings made of?

Revenue and the margin on it from the companies' own filings, and per-share earnings growth split into revenue, margin and the share count

Exhibit 4 · Per-share earnings growth and the share count's part, since 2016

Four-quarter growth, log points (pp), quarterly since 2016: per-share earnings for all companies and outside finance, and the falling share count's part outside finance
About this exhibit
Lines of four-quarter per-share earnings growth for all companies and outside finance, and the falling share count's part outside finance, log points, quarterly from Q3 2016.
Per-share earnings growth from the filings, and the share count's part
Growth over four quarters, log points, quarterly from Q1 2016: a holder of a fixed fraction of every company.
Per-share earnings growth is total earnings growth plus the share-count effect: a holder who owns a fixed fraction of every company gains when share counts fall. Per-share earnings growth is a holder's who owns a fixed fraction of every company: total earnings growth plus the share-count effect, where the share-count effect is the change in that holder's earnings less the change in the companies' total earnings. Outside finance it splits further into revenue growth and the change in the margin. Each quarter is a link on one sample of companies; the four-quarter figure adds four links, so the parts add to the whole. A company whose share count moves by more than a split's factor in a quarter (below 0.55 or above 1.8 times) is left out of that link. These are the companies' own filings, summed; the index's EPS is S&P's and differs by float and membership.
Source: U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; Macro Review calculations · data to Q2 2026cross.profitability-12

Exhibit 5 · Per-share earnings growth outside finance split into revenue, margin and share count

Four-quarter growth, log points (pp), quarterly to Q2 2026: revenue, net margin change and falling share count add to per-share growth (the line).
About
About this exhibit
Stacked bars of four-quarter per-share earnings growth outside finance split into revenue, margin change and share count, with total per-share growth as a line, quarterly (the latest 20 quarters by default).
Per-share earnings growth in three parts, outside finance
Growth over four quarters, log points: revenue, the change in the net margin and the falling share count add to per-share growth (the line).
Each quarter's link is taken on one sample of companies and four links are added, so the parts add to the whole exactly. The margin part swings most: in a recession it is the margin, not revenue, that falls. 500 companies held by State Street's S&P 500 fund on 13 Nov 2026; today's members only before that date.
Source: U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; Macro Review calculations · data to Q2 2026cross.profitability-13

Do the national accounts support it?

Profits against index earnings, the profit share against its , and the gap between EPS growth and profits growth against its long-run average

Exhibit 6 · S&P 500 EPS growth against national-accounts profit growth, since 1980

Four-quarter growth, %, quarterly since 1980; index EPS reported (Shiller), profits from the national accounts; data to Q2 2026.
About
About this exhibit
Lines of four-quarter growth of S&P 500 reported EPS against national-accounts profits after tax, pre-tax profits and domestic nonfinancial profits, %, quarterly since 1980 (ten years by default).
Index EPS against national-accounts profits
Growth over four quarters, %, quarterly since 1980; index EPS reported (Shiller), profits from the national accounts. Data to Q2 2026.
The index is not the economy: a few sectors' margins, buybacks and the reported-versus-operating gap sit in the difference. Over the full history reported EPS growth reached +793% in 2010 off the collapsed 2009 base; the axis leaves that quarter out and says so.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller · data to Q2 2026cross.earnings-12

Exhibit 7 · After-tax profits as a share of GDP, against the 2015–19 average

After-tax profits as % of nominal GDP, quarterly since 1980, against the 2015–19 average; data to Q2 2026.
About
About this exhibit
After-tax profits as % of GDP, quarterly since 1980, with the 20-quarter average and domestic nonfinancial and financial shares, against the 2015–19 average.
The profit share of GDP
After-tax profits as % of nominal GDP, quarterly since 1980, against the 2015–19 average; data to Q2 2026.
Profits after tax with inventory and capital-consumption adjustments. Margin gap (the share less its average over the last 20 quarters) +1.11pp. The Corporate profits and margins page reads the same 11.91% for Q2 2026, and its test — the share back at 9.8% for two quarters — is the one threshold this page uses.
Source: U.S. Bureau of Economic Analysis · data to Q2 2026cross.earnings-13

Exhibit 8 · Profit growth to EPS growth: the calendar-year projection against last year

Four-quarter growth, log points (%): profits plus the EPS-over-profits gap gives EPS; calendar 2027: model median, 1990–2019 average gap; beside, the year to Q2 2026.
Calendar 2027, the fan's medianThe year to Q2 2026, as reported
Profits after tax, growth (log points)+5.5+16.7
+ gap of EPS over profits (log points)−0.92−0.13
= EPS growth (log points)+4.6+16.6
EPS growth, %+4.7%+18.0%
About this exhibit
Table bridging profit growth to EPS growth in log points: for the calendar year the model's median profits, the average gap and EPS, beside the latest reported year.
The bridge from profits to EPS, and last year's gap
Four-quarter log growth: the fan adds the 1990–2019 average gap; last year's actual gap beside it.
The average gap is −0.92 log points a year over 120 quarters, with a standard deviation of 47.9: the fan is shifted by the average and not widened by the spread. Last year's gap is 0.0 standard deviations above the average.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Macro Review calculations · data to Q2 2026cross.earnings-26

How much do taxes take, and how clean are the earnings?

The tax share of profits in the filings against the national accounts, and the gap between operating and as-

Exhibit 9 · Tax share of profits: company filings against the national accounts

Income taxes ÷ pretax income, trailing year, %, quarterly since Q2 2015; the national accounts' taxes ÷ profits before tax.
About this exhibit
Lines of income taxes ÷ pretax income for the index's companies (filings) and taxes ÷ profits before tax for every corporation (national accounts), trailing year, quarterly since Q4 2015.
Taxes as a share of profits: the filings against the national accounts
Income taxes ÷ pretax income, trailing year, %, quarterly since Q2 2015; the national accounts' taxes ÷ profits before tax.
The two differ in coverage — every US corporation in the national accounts, the index's companies in the filings, abroad included — and in what counts as profit: the national accounts adjust for inventory valuation and depreciation. The national-accounts line is the Corporate profits and margins page's.
Source: U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; U.S. Bureau of Economic Analysis; Macro Review calculations · data to Q2 2026cross.profitability-14

Exhibit 10 · S&P 500 operating and as-reported earnings against company filings, by quarter

S&P Dow Jones Indices' trailing-year earnings, $bn, for the index, Q2 2023 to Q3 2025, beside the filings' net income summed over the current members.
Year toOperating, S&PAs reported, S&PGap, % of operatingNet income, the filingsFilings ÷ as reported
Q3 20252,1651,9957.8%2,0991.05
Q2 20252,0471,8838.0%2,0021.06
Q1 20251,9961,78210.7%1,9561.10
Q4 20241,9671,72612.2%1,9071.11
Q3 20241,9011,63813.8%1,8151.11
Q2 20241,8391,59813.1%1,7851.12
Q1 20241,8071,60611.1%1,7341.08
Q4 20231,7871,6109.9%1,7291.07
Q3 20231,7561,54112.3%1,6551.07
Q2 20231,7421,51513.0%1,5911.05
About this exhibit
Table of S&P's trailing-year operating and as-reported earnings for the index, $bn, the gap as % of operating, the filings' net income and its ratio to as-reported, by quarter.
Operating against as-reported earnings, and the filings against both
S&P Dow Jones Indices' trailing-year earnings for the index, $bn, beside the filings' net income summed over today's members.
Operating earnings leave out what the company calls one-off — write-downs, restructuring, gains on sales — so the gap between the two measures how much is set aside. The filings' net income is company totals as filed (no float adjustment) for today's members; it ran between 0.90 and 1.25 times S&P's as-reported earnings in 10 of 10 quarters. The S&P figures are the Valuation page's.
Source: S&P Dow Jones Indices; U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; Macro Review calculations · data to Q2 2026cross.profitability-15

What does the valuation assume?

The Committee's longer-run real growth and inflation added — the rate the valuation fades to after the near term, and the gap over it the model's paths carry on the Outlook pages

Exhibit 11 · How the assumed long-run growth rate adds up, in its parts

The Fed's longer-run real GDP growth plus PCE inflation, % a year (projections of 16 Sep 2026), then the profit gap added for eight projected quarters
Longer-run real GDP growththe Committee's median, 16 Sept 2026+2.0%
Longer-run PCE inflationthe same table+2.0%
Terminal growth: long-run nominal GDPthe two above, added — the rate the Valuation page's two-stage model fades to+4.0%
+ the fan's gap, profits − nominal GDPaverage over 120 quarters, Q1 1990 – Q4 2019, in logs (median +1.89); held for the fan's 8 quarters only+2.20
= the rate the fan reverts to, profitslog points a year; carried to EPS by −0.92, 5.286.20
The same on trailing EPS (robustness)4.0 + the EPS − nominal GDP gap +1.28 (median +6.47)5.28
The one-stage variant's sensitivities3.00: the September report's g, the growth a 0.85 payout sustains — a sensitivity since the one-rate decision; 5.28: the robustness-case fan's assumed long-run rate held for ever — trailing EPS, 4.0 + wedge 1.283.00 · 5.28
About this exhibit
Table building the assumed growth rate: the Fed's longer-run real GDP growth plus PCE inflation; plus the model's profit gap for the projected quarters; the trailing-EPS robustness and the one-stage sensitivities.
Terminal growth handed to Valuation, in its parts
The Committee's long-run real growth and inflation added: 4.0% a year, an assumption; the fan's gap over it lasts 8 quarters.
After the near-term years the two-stage model fades to this , over 6 years (a convention of this site, not estimated — printed as such). The gap is what profits (or index earnings per share) have grown faster than the economy: buybacks, the index's own composition and the difference between national-accounts profits and reported EPS all sit in it. Held for ever, a gap over GDP would imply a profit share rising without limit, so it fades to zero. The one-stage variant on the Valuation page runs at the same rate and prints its result at the sensitivities in the last row as well.
Source: Federal Reserve (Summary of Economic Projections); U.S. Bureau of Economic Analysis; Robert J. Shiller; Macro Review calculations · data to 16 Sept 2026cross.earnings-20

What does the index itself earn, and what has the quarter being reported shown?

Earnings per index unit read through the fund that holds the index — quarter by quarter on its holdings since 2019, checked against S&P's own figures — and the companies that have filed the next quarter so far

Exhibit 12 · Earnings per index unit from filings against S&P's as-reported EPS

Trailing four-quarter earnings, dollars per index unit, quarterly from Q2 2020; dashed, S&P's as-reported EPS at each quarter-end.
About
About this exhibit
Lines of trailing four-quarter earnings per index unit summed from the filings through SPY's holdings against S&P 500 as-reported EPS from Shiller's file (dashed), dollars, quarterly.
Earnings per index unit, from the filings, against S&P's own
Trailing four quarters, dollars per index unit, quarterly from Q2 2020; dashed, S&P's as-reported EPS at each quarter-end.
Each member's quarter over its diluted share count, times the shares SPY held in it at that quarter's end (its Form N-PORT), over the value of those holdings, times the S&P 500's close — the fund's share of each company and the index's divisor cancel. Point in time: the members of each quarter, those since removed included, each figure as first reported. Against S&P's figure it sat within 0.97–1.03 in 10 of the 10 quarters S&P's buyback release covers, and within 0.97–1.05 in 21 of 25 quarters overall (outside: Q1 2021 0.941, Q4 2022 1.051, Q1 2026 1.080, Q2 2026 1.114). On today's holdings (13 Nov 2026) the year to Q2 2026 is $292.19, a trailing multiple of 25.9.
Source: U.S. Securities and Exchange Commission (EDGAR, Form N-PORT); State Street Global Advisors; Robert J. Shiller; Yahoo Finance; Macro Review calculations · data to Q2 2026cross.profitability-19

Exhibit 13 · Earnings per index unit against S&P's as-reported EPS, by quarter

Trailing four quarters, dollars per index unit, the 10 quarters S&P's buyback release covers; the band written before the series was computed.
Year toIndex basisS&P as reportedRatioInside 0.97–1.03
Q3 2025233.33235.250.992yes
Q2 2025222.23222.530.999yes
Q1 2025214.01216.690.988yes
Q4 2024206.89210.170.984yes
Q3 2024195.10200.270.974yes
Q2 2024197.03196.761.001yes
Q1 2024193.03191.391.009yes
Q4 2023191.51192.430.995yes
Q3 2023185.96184.251.009yes
Q2 2023179.26181.010.990yes
About this exhibit
S&P's figure is its four-quarter as-reported total as Shiller's file carries it at the quarter-end month. What separates the two: S&P counts each company's EPS as reported (net income to common shareholders), the filings net income over the diluted count; the fund's small cash and timing differences; members the reading cannot place (0.24% of the fund at most).
Source: U.S. Securities and Exchange Commission (EDGAR, Form N-PORT); Robert J. Shiller; Macro Review calculations · data to Q2 2026cross.profitability-20

Exhibit 14 · Q3 2026 net income growth of companies that have filed so far

444 of 500 companies, 83.9% of the index's weight, have filed Q3 2026; the reading starts at 60%.
Q3 2026 against Q3 2025GrowthOver
Net income, the companies that have filed+27.7%444 companies, 84% of the index's weight
Reweighted to the index's sectors+32.8%sectors carrying 100% of the index's net income a year earlier
Revenue outside finance+14.9%the companies outside finance that have filed
SectorThe index's weightThe filers' weightFiled, % of the sectorNet income growthShare of the index's net income, Q3 2025
Information Technology35.7%28.0%66%+86.4%24.4%
Financials12.7%15.0%100%−7.9%24.7%
Health Care10.5%11.8%95%+25.3%7.8%
Consumer Discretionary9.5%9.8%87%+9.6%9.6%
Communication Services8.9%10.3%96%+11.8%9.1%
Industrials8.2%9.4%96%+7.3%8.0%
Consumer Staples5.2%4.7%76%+14.1%5.7%
Energy3.6%4.3%100%+156.8%4.3%
Utilities2.1%2.5%98%+32.6%3.5%
Real Estate1.9%2.2%100%−3.5%1.2%
Materials1.7%1.9%94%+32.0%1.7%
About this exhibit
A status card until most of the index has filed: the companies that have filed the quarter being reported and their share of the index's weight; net income growth prints once filers carry the threshold.
Q3 2026 so far, with 84% of the index's weight reported
Net income of the 444 companies that have filed Q3 2026, against their own Q3 2025; their sectors beside the index's.
The quarter after Q2 2026 is shown only on its own card, only once the companies that have filed it carry at least 60% of the index's weight, and only as the growth of their net income over the same quarter a year earlier (the companies that have filed both): no figure is estimated for a company yet to file. Large companies file first, so the sample leans to them; the card sets the sample's sector weights beside the index's and reweights the growth to the index's sectors — each sector's growth over its own filers, weighted by the sector's share of the whole index's net income a year earlier. It never enters a tile, a takeaway or the history.
Source: U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; Macro Review calculations · data to Q2 2026cross.profitability-18

Carried forward, the statements put trailing earnings on this path

: the base's trailing earnings per index unit for eight quarters with its band, and the year-by-year path to the terminal state

Exhibit 15 · Trailing earnings per index unit, with the model's quarterly projection

Dollars per index unit, trailing four quarters; history to Q2 2026, then the model's base-case median with 25th–75th and 10th–90th bands of its draws; model-conditional
Range: simulated shocks
About
About this exhibit
Line of trailing earnings per index unit (the index basis from the fund's holdings, Shiller's series beside it) to the latest filed quarter, then the base case's median projection for the next eight quarters with its bands
Trailing earnings per index unit, and the base's next eight quarters
Dollars per index unit, four quarters to each quarter's end: SPY's holdings back to 2020, Shiller's series before and beside it; then the base from Q2 2026, with the 25th–75th and 10th–90th of its draws.
The index's income statement, payout and balance sheet carried forward by their own equations: revenue from nominal GDP, the dollar and oil; a pretax margin that reverts toward its 2017–26 mean; dividends by Lintner's rule; buybacks, issuance and dilution to the share count. Model-conditional: what the statements imply if the equations hold, not a forecast of what will be filed.
Source: U.S. Securities and Exchange Commission (EDGAR: the members' filings; Form N-PORT); State Street Global Advisors; Yahoo Finance; BEA; Robert J. Shiller; Federal Reserve Board; EIA; Macro Review calculationscross.earnings_outlook-15

The regression's tables and the quoted multiples

The top-down regression's coefficients and fitted values, its quarters, the jumps in the trailing-EPS fit, the forward multiples as quoted, and every quarterly series

Exhibit 16 · EPS growth regression on macro inputs and margins in each sample

Errors in pp: four-quarter EPS growth on nominal GDP, the dollar, oil and margins, fitted Q1 1990 to Q4 2019; headline ex-2008–10, alternative 2008–10 included
About
RegressorCoefficientHAC SEtCoefficient, 2008–10 includedHAC SEt
Constant−31.78811.328−2.81+29.95448.924+0.61
Nominal GDP, 4-quarter growth (%)+8.5132.513+3.39+1.6667.074+0.24
Broad dollar, 4-quarter change (%)−1.2030.677−1.78−4.7592.729−1.74
WTI, 4-quarter change (%)−0.0400.133−0.30+0.4260.562+0.76
Margin gap, lagged four quarters (pp)+3.9742.919+1.36−16.77816.396−1.02
R² · quarters0.397 · n = 1120.119 · n = 120
Out of sampleQuartersHeadline RMSE2008–10 included RMSESame as last yearZero growthMean error (headline)
Q1 2022 – Q2 2026 (2020–21 excluded)1821.8pp27.9pp41.1pp17.5pp−14.8pp
Alternative sample: 2020–21 included, Q1 2020 – Q2 20262626.6pp37.8pp53.4pp34.1pp−8.3pp

Headline sample: Q1 1990 – Q4 2019, Q4 2008 – Q3 2010 excluded (112 quarters; R² 0.40). 8 base-effect quarters (Q4 2008 – Q3 2010: reported EPS growth −89% to +793% off the collapsed 2008–09 base) dominate the as-written fit (R² 0.12, nominal-GDP t 0.24), and its paths (+51–55% four-quarter growth) are not credible as an expectation. Excluding them is a sample choice, not a change of specification; the author can revert it with the page's single headline-sample switch. The specification as written (Q1 1990 – Q4 2019, 120 quarters; R² 0.12, constant 30.0) is printed as the sensitivity in the model table, on the fitted-values chart and as a second path, with both out-of-sample scores side by side.

About this exhibit
Table of the regression's coefficients, standard errors and t for four-quarter EPS growth on nominal GDP growth, the dollar's change, WTI's change and the lagged margin gap, on two samples, plus out-of-sample errors against naive benchmarks.
The regression's coefficients, two samples
Four-quarter EPS growth on nominal GDP, the dollar, oil and margins; headline ex-2008–10, alternative 2008–10 included; Newey–West errors.
Dependent: S&P 500 trailing 12-month reported EPS (Shiller), 4-quarter growth, %. OLS; Newey–West standard errors, 4 lags. Headline sample Q1 1990 – Q4 2019, Q4 2008 – Q3 2010 excluded; alternative sample Q1 1990 – Q4 2019.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve Board (H.10); EIA; Macro Review calculations · data to Q2 2026cross.earnings-1

Exhibit 17 · EPS growth regression: actual against fitted, with predictions after 2019

Four-quarter EPS growth in log points (%), actual and fitted; fitted to Q4 2019, predicted from realised inputs after.
About this exhibit
Lines of actual four-quarter reported EPS growth, the headline regression's fitted values and predictions after its sample, and the alternative sample's, in log points, with the 2020–21 base effects and the out-of-sample period shaded.
The regression's fit and its predictions after 2019
Four-quarter EPS growth in log points, actual and fitted; fitted to Q4 2019, predicted from realised inputs after.
Log points (100 × log of one plus the percentage growth) — the scale the model is read on, where a fall of a half and a doubling are the same distance. Fitted over Q1 1990 – Q4 2019 and predicted from realised inputs afterwards (shaded from Q1 2020). The 8 base-effect quarters Q4 2008 – Q3 2010 reached −89% to +793% off the collapsed 2009 base, which is 219 log points here; the 2020–21 quarters are the pandemic's own collapse and rebound. Both are shaded, and both are on the axis.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve Board (H.10); EIA; Macro Review calculations · data to Q2 2026cross.earnings-3

Exhibit 18 · EPS growth regression against actual growth and naive benchmarks, by quarter

Actual four-quarter EPS growth, %, against the regression's prediction and naive benchmarks; rows before Q1 2022 are the 2020–21 sample.
QuarterActualHeadlineErrorLast year2008–10 included
Q2 2026+18.0%+23.1%−5.1+13.1%+72.5%
Q1 2026+11.6%+26.3%−14.7+13.2%+67.4%
Q4 2025+12.2%+18.1%−5.9+9.2%+47.3%
Q3 2025+17.5%+16.5%+0.9+8.7%+46.7%
Q2 2025+13.1%+10.0%+3.1+8.7%+29.5%
Q1 2025+13.2%+2.1%+11.1+9.3%+8.1%
Q4 2024+9.2%+9.3%±0.0+11.4%+12.8%
Q3 2024+8.7%+12.3%−3.7−1.6%+20.5%
Q2 2024+8.7%+17.7%−9.0−5.9%+29.7%
Q1 2024+9.3%+17.4%−8.1−11.5%+34.0%
Q4 2023+11.4%+25.8%−14.4−12.7%+49.1%
Q3 2023−1.6%+30.8%−32.4+6.7%+38.9%
Q2 2023−5.9%+25.8%−31.7+21.1%+16.9%
Q1 2023−11.5%+31.2%−42.7+54.4%+12.6%
Q4 2022−12.7%+27.1%−39.8+110.2%−6.2%
Q3 2022+6.7%+43.2%−36.5+78.6%−1.4%
Q2 2022+21.1%+51.2%−30.1+60.0%+20.1%
Q1 2022+54.4%+61.9%−7.5+10.2%+46.3%
Q4 2021+110.2%+69.7%+40.5−32.5%+79.4%
Q3 2021+78.6%+64.7%+13.9−26.1%+72.6%
Q2 2021+60.0%+119.1%−59.2−26.6%+168.3%
Q1 2021+10.2%+7.4%+2.8−13.4%+81.1%
Q4 2020−32.5%−21.4%−11.1+5.3%+26.9%
Q3 2020−26.1%−30.8%+4.7+1.9%+9.4%
Q2 2020−26.6%−92.7%+66.0+10.4%−34.6%
Q1 2020−13.4%−7.7%−5.8+16.4%+11.0%
About this exhibit
Table of actual four-quarter EPS growth against the regression's prediction, its error and two naive benchmarks (last year, the 2008–10-included fit), quarter by quarter since 2020.
The regression out of sample, quarter by quarter
Actual four-quarter EPS growth against the regression's prediction and two naive benchmarks; rows before Q1 2022 are the 2020–21 sample.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve Board (H.10); EIA; Macro Review calculations · data to Q2 2026cross.earnings-2

Exhibit 19 · Quarters with outsized EPS shocks, and the high-volatility state's probability

Quarters whose standardised one-step error exceeds 3, p = 1.50% a quarter, to Q2 2026; sizes resampled with their signs.
About
QuarterStandardised errorSize (log points)Dummy coefficient
Q4 2008−9.80−111.6−63.9
Q2 2009+3.23+36.8+25.6
Q4 2009+8.18+93.2+89.9
Q1 2010−5.12−58.3−8.2

Two-state Markov-switching variance on the same series: standard deviations 3.9 and 38.1 log points, expected durations 61 and 8 quarters; smoothed probability of the high-variance state at Q2 2026 0.7%. Highest recent readings: Q4 2008 100%, Q4 2009 100%, Q1 2009 100%, Q3 2009 100%.

About this exhibit
Table of quarters whose standardised one-step error exceeds the jump cut, with size and dummy coefficient, plus a two-state switching-variance result (state volatilities, durations, current probability).
Jumps in the trailing-EPS fit, and a switching-variance check
Quarters whose standardised one-step error exceeds 3, p = 1.50% a quarter; sizes resampled with their signs.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.earnings-6

Exhibit 20 · The forward multiples and growth quotes, as FactSet printed them

FactSet's forward P/E at quarter ends and its calendar-year EPS growth, as quoted; refreshed only with a new quote.
QuoteValueDateWhere it sits
Calendar 2026 EPS growth+31.6%11 Sept 2026FactSet Earnings Insight, 11 Sep 2026, quoted in the September 2026 issue §2.1
Calendar 2027 EPS growth+15.1%11 Sept 2026FactSet Earnings Insight, 11 Sep 2026, quoted in the September 2026 issue §2.1
Forward 12-month P/E22.0×31 Dec 2025the Valuation page's quarter-end quotes (the premium implied that day on a payout basis, +2.68pp — not today's implied premium, which the Valuation page prints)
Forward 12-month P/E19.7×31 Mar 2026the Valuation page's quarter-end quotes (the premium implied that day on a payout basis, +3.02pp — not today's implied premium, which the Valuation page prints)
Forward 12-month P/E20.4×30 Jun 2026the Valuation page's quarter-end quotes (the premium implied that day on a payout basis, +2.73pp — not today's implied premium, which the Valuation page prints)
Forward 12-month P/E19.7×1 Sept 2026the Valuation page's quarter-end quotes (the premium implied that day on a payout basis, +2.53pp — not today's implied premium, which the Valuation page prints)
Forward 12-month P/E19.1×11 Sept 2026the Valuation page's quarter-end quotes (the premium implied that day on a payout basis, +2.50pp — not today's implied premium, which the Valuation page prints)
Forward P/E, latest18.66×16 Nov 2026the Valuation page's multiple, moved with SPY's price
Source: FactSet Earnings Insight · data to 16 Nov 2026cross.earnings-16

Exhibit 21 · Every quarterly series on the page

Latest value, changes over a quarter and a year shaded against 2015–19, percentile since 1990; oil as its change only.
SeriesLevel1q1yPct since 1990Level, recentAs of
Index earnings
S&P 500 trailing 12-month reported EPSRobert J. Shiller's series, reported (GAAP) earnings at quarter-end months$295.52+12.5%+25.6%100Q3 2026
National accounts: growth
Corporate profits with IVA and CCAdj, 4-quarter growthNIPA Table 1.1220.8%+9.51pp+16.31pp90Q2 2026
Profits after tax with IVA and CCAdj, 4-quarter growthNIPA Table 1.1218.2%+8.37pp+11.22pp86Q2 2026
S&P 500 trailing 12-month reported EPS, 4-quarter growthShiller, reported earnings25.6%+7.62pp+8.15pp83Q3 2026
Domestic nonfinancial profits, 4-quarter growthNIPA Table 6.16D, from 200221.1%+10.91pp+21.77pp82Q2 2026
Domestic financial profits, 4-quarter growthNIPA Table 6.16D, from 200223.1%+4.86pp−0.17pp83Q2 2026
National accounts: margins
After-tax profits, % of GDP11.91%+0.62pp+1.19pp100Q2 2026
Pre-tax profits with IVA and CCAdj, % of GDP14.46%+0.91pp+1.73pp100Q2 2026
After-tax share, trailing 20-quarter average10.80%+0.02pp+0.11pp100Q2 2026
Margin gap: share − its trailing 20-quarter average1.11 pp+0.60pp+1.09pp78Q2 2026
Domestic nonfinancial profits, % of GDPfrom 20019.76%+0.71pp+1.19pp100Q2 2026
Domestic financial profits, % of GDPfrom 20013.03%+0.15pp+0.41pp93Q2 2026
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve Board (H.10); EIA; Macro Review calculations · data to Q2 2026cross.earnings-14

Appendix

The model, its parameters, each step of the calculation and every source172 checks, all pass
About this page

Step one of the valuation chain: where index earnings stand — reported earnings per share in dollars since 1980, and the operating earnings and growth the consensus quotes beside them — what drives them as FactSet quotes them, and what they are made of in the companies' own filings: revenue, the margin and the falling share count, the taxes paid and how much is set aside as one-off, whether the national accounts support them, and the terminal growth the valuation assumes after the near term. What the model expects earnings to do next — its fan, with its inputs and its record — is read on the Scenarios page under Outlook.

The growth process: an unobserved-components model

yt=μt+ct+∑jβjdjt+εt,εt∼N(0,σε2)y_t = \mu_t + c_t + \textstyle\sum_j \beta_j d_{jt} + \varepsilon_t, \qquad \varepsilon_t \sim N(0, \sigma_\varepsilon^2)
μt=μˉ+ϕ (μt−1−μˉ)+ηt,ηt∼N(0,ση2),ϕ∈[0.80,0.99]\mu_t = \bar\mu + \phi\,(\mu_{t-1} - \bar\mu) + \eta_t, \qquad \eta_t \sim N(0, \sigma_\eta^2), \qquad \phi \in [0.80, 0.99]
ct=ρ1ct−1+ρ2ct−2+νt,νt∼N(0,σν2)c_t = \rho_1 c_{t-1} + \rho_2 c_{t-2} + \nu_t, \qquad \nu_t \sim N(0, \sigma_\nu^2)
μ0∼N(μˉ, 100),(c0,c−1)′∼N(0,Σc),  vec⁡Σc=(I−Tc⊗Tc)−1vec⁡Qc\mu_0 \sim N(\bar\mu,\ 100), \qquad (c_0, c_{-1})' \sim N(0, \Sigma_c),\ \ \operatorname{vec}\Sigma_c = (I - T_c \otimes T_c)^{-1}\operatorname{vec} Q_c
μˉ  restricted, not estimated;if ϕ^ reaches a bound or σ^η→0: ϕ:=0.95\bar\mu \;\text{restricted, not estimated};\qquad \text{if } \hat\phi \text{ reaches a bound or } \hat\sigma_\eta \to 0:\ \phi := 0.95
simulation: εT+h∗=σεz+J 1{u<p},J resampled from the jump quarters’ errors, p=#jumps/n\text{simulation: } \varepsilon^{*}_{T+h} = \sigma_\varepsilon z + J\,\mathbf 1\{u < p\},\quad J \text{ resampled from the jump quarters' errors},\ p = \#\text{jumps}/n
SymbolMeaningFromAssumed or estimated
yty_tNIPA corporate profits after tax with inventory and capital-consumption adjustments, 100 × the quarterly change in logsBEA NIPA Table 1.12 line 15–
μt\mu_tthe long-run growth state (the trend), reverting to μ̄estimated by the Kalman filterestimated
μˉ\bar\muthe assumed long-run rate: the Committee's longer-run real GDP growth 2.0 + PCE inflation 2.0 + the 1990–2019 average gap of profits over nominal GDP (+2.20) = 6.20 a year, 1.550 a quarterFOMC Summary of Economic Projections; BEAassumed
ϕ\phihow slowly the state returns to μ̄ each quarterestimated, or fixed by the bound ruleassumed
ctc_tthe cycle, a stationary second-order autoregressionestimated by the Kalman filterestimated
ρ1\rho_1, ρ2\rho_2the cycle's persistence and turnmaximum likelihoodestimated
djtd_{jt}data: 1 in jump quarter j, 0 otherwise — a quarter whose one-step error in the Gaussian fit is beyond 3 standard deviationsthe first-pass fit–
βj\beta_jthe size of jump quarter j's impulsemaximum likelihoodestimated
ση\sigma_\eta, σν\sigma_\nu, σε\sigma_\varepsilonthe shocks' standard deviations: trend, cycle, observationmaximum likelihoodestimated
μ0\mu_0, Σc\Sigma_cthe start: the trend known at μ̄ with variance 100 (diffuse in practice), the cycle at its stationary covariance given ρ and σ_νset by the model, not estimatedassumed
JJ, ppthe jump draw of the simulation: one of the identified jump quarters' prediction errors, with probability p a quarterthe jump list; not in the estimated likelihoodassumed

Maximum likelihood through the Kalman filter, standard errors from the Hessian over the parameters not at a bound; Q1 1960 – Q2 2026, quarterly, 266 observations; the best of 29 starting points. Identification — On NIPA profits after tax (the fan): estimated freely, σ_η goes to zero (0.000; φ 0.80) — the data cannot separate a slow trend from the cycle, so the trend is held at the assumed rate μ̄ and the spread of the paths is the cycle's and the jumps'; φ is then fixed at 0.95 by the rule. On trailing reported EPS (the robustness case): estimated freely, σ_η goes to zero (0.000; φ 0.80) — the data cannot separate a slow trend from the cycle, so the trend is held at the assumed rate μ̄ and the spread of the paths is the cycle's and the jumps'; φ is then fixed at 0.95 by the rule. The fan draws its paths from the smoothed state and its covariance at the last quarter; each simulated quarter's observation shock is Gaussian plus, with probability p = 1.1% (the share of jump quarters), one of the 3 identified jumps resampled with its sign — a mixture the estimated likelihood does not contain, so the fan's tails are the simulation's assumption, not an estimate. The parameter table follows (cross.earnings-5).

From the profits fan to index EPS, in dollars and at two horizons

EPST+h=EPST exp⁡ ⁣[(∑s=1hyT+s+h4 b)/100]\mathrm{EPS}_{T+h} = \mathrm{EPS}_T \, \exp\!\Big[\big(\textstyle\sum_{s=1}^{h} y_{T+s} + \tfrac{h}{4}\, b\big) / 100\Big]
b=mean⁡1990–2019(Δ4ln⁡EPSt−Δ4ln⁡Πt)b = \operatorname{mean}_{1990–2019}\big(\Delta_4 \ln \mathrm{EPS}_t - \Delta_4 \ln \Pi_t\big)
g2027=EPSQ4 2027 / EPSQ4 2026−1,gS=(EPST+8 / EPST)1/2−1g_{2027} = \mathrm{EPS}_{\mathrm{Q4}\,2027} \,/\, \mathrm{EPS}_{\mathrm{Q4}\,2026} - 1, \qquad g_S = \big(\mathrm{EPS}_{T+8} \,/\, \mathrm{EPS}_T\big)^{1/2} - 1
SymbolMeaningFromAssumed or estimated
EPST\mathrm{EPS}_Ttrailing 12-month reported EPS at the origin, $262.59 (Q2 2026)Robert J. Shiller–
yT+sy_{T+s}quarterly profits growth on each of 20,000 simulated pathsthe model above–
Πt\Pi_tprofits after tax with inventory and capital-consumption adjustmentsBEA NIPA Table 1.12 line 15–
bbthe average gap between EPS and profits growth, −0.92 log points a year over 120 quarters (standard deviation 47.9), held fixed over the horizonBEA; Robert J. Shillerestimated
g2027g_{2027}calendar-2027 growth, median +4.7% — compared with consensus on this pagethis page–
gSg_Sthe eight-quarter rate, median 5.77% a year — the near-term growth of the two-stage modelthe Valuation page–

One set of paths read three ways: in dollars, as growth over the calendar year and as a rate over eight quarters. The bridge is a level shift in logs, so every percentile moves by it; the spread of the gap itself is not carried into the fan. Consensus is operating (bottom-up) earnings and Shiller's series is reported (GAAP): the two are drawn side by side, never spliced.

Exhibit 22 · The trend-and-cycle model's parameters, with standard errors

Maximum likelihood on profits after tax (the fan) and on trailing EPS (robustness); Hessian errors, none at a bound.
About
ParameterNIPA profits, with jump dummies (default)Trailing: free (Gaussian)Trailing: Gaussian, φ pinnedTrailing: with jump dummies
φ — persistence of the growth state0.950fixed0.800at bound0.950fixed0.950fixed
σ_η — growth-state shock0.000at bound0.000at bound0.001at bound0.000at bound
ρ1 — cycle, first lag1.303SE 0.2281.026SE 0.0841.025SE 0.0850.854SE 0.069
ρ2 — cycle, second lag-0.608SE 0.177-0.569SE 0.072-0.567SE 0.075-0.383SE 0.066
σ_ν — cycle shock0.997SE 0.3907.536SE 0.7527.553SE 0.7576.919SE 0.300
σ_ε — observation shock4.205SE 0.2416.646SE 0.6076.640SE 0.6040.000at bound
Log-likelihood-786.60-1,024.65-1,025.12-892.91

On NIPA profits after tax (the fan): estimated freely, σ_η goes to zero (0.000; φ 0.80) — the data cannot separate a slow trend from the cycle, so the trend is held at the assumed rate μ̄ and the spread of the paths is the cycle's and the jumps'; φ is then fixed at 0.95 by the rule. On trailing reported EPS (the robustness case): estimated freely, σ_η goes to zero (0.000; φ 0.80) — the data cannot separate a slow trend from the cycle, so the trend is held at the assumed rate μ̄ and the spread of the paths is the cycle's and the jumps'; φ is then fixed at 0.95 by the rule. Parameters at a variance bound have no Hessian standard error and are printed as 'at bound', never as a number; the other errors invert the Hessian over the identified parameters. An independent Kalman filter reproduces the likelihood at the estimates (−786.5988 on profits, −892.9059 on trailing EPS).

Assumed long-run rate over the fan's 8 quartersNIPA profits (default)Trailing EPS (robustness)Source
SEP longer-run real GDP growth2.002.00FOMC projection table, 16 Sept 2026
SEP longer-run PCE inflation2.002.00FOMC projection table, 16 Sept 2026
Wedge over nominal GDP+2.20+1.28profits (left) or EPS (right) less nominal GDP, Q1 1990 – Q4 2019 mean, 120 quarters, in logs; medians +1.89 and +6.47
μ̄6.205.281.550 and 1.320 a quarter · bridged to EPS the left one is 5.28
Beyond the horizon: the one long-run rate4.0nominal GDP growth, no wedge: the rate Valuation's two-stage identity fades to (the card cross.earnings-20)
About this exhibit
The fan's fit is on national-accounts profits after tax, Q1 1960 – Q2 2026 (266 quarters; the best of 29 starting points); the robustness case on S&P 500 trailing 12-month reported EPS (Shiller), quarterly log growth × 100, Q1 1960 – Q2 2026 (266 quarters). Maximum likelihood; errors from the Hessian over the identified parameters. Cycle: one turn in 10.8 quarters on the NIPA fit, 7.8 on the trailing series. What is identified on the trailing series is an AR(2) around a fixed assumed rate, with jumps.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.earnings-5

Exhibit 23 · Definitions

How the numbers are made, step by step

  1. The national accounts. Source: BEA NIPA Tables 1.12, 6.16D and 1.1.5 (the quarterly flat file). Transform: profits, taxes and GDP by line; growth and the after-tax profit share. Check: the tables' own identities to the million; BEA's published changes reproduced; the Profits page's copies agree (17 checks, all passing).
  2. Index earnings. Source: Robert J. Shiller's workbook: trailing 12-month reported earnings. Transform: quarter-end earnings per share in dollars and its four-quarter growth. Check: the workbook's real columns reproduced; the series complete and current (11 checks, all passing).
  3. The top-down regression. Source: steps 1–2, the Board's H.10 dollar and EIA's WTI. Transform: OLS with Newey–West errors, 1990–2019; eight-quarter paths at 4, 5 and 6% nominal GDP; out of sample from 2022. Check: coefficients and errors equal statsmodels; fitted + residual = actual; the path band is the out-of-sample error (10 checks, all passing).
  4. The long-run rate. Source: the FOMC's September projection table. Transform: longer-run real GDP + PCE inflation medians; the wedge of EPS over nominal GDP, 1990–2019. Check: the rate is the SEP's medians added; one rate on Earnings and Valuation (8 checks, all passing).
  5. The growth process. Source: steps 1 and 4. Transform: core.ucm: an unobserved-components model of profits growth, jump dummies, a grid of starting values. Check: the likelihood reproduced by an independent Kalman filter; the published estimate the best of every start (11 checks, all passing).
  6. The fan, bridged to EPS. Source: step 5 and the EPS–profits bridge. Transform: 20,000 simulated paths; every percentile shifted by the 1990–2019 mean EPS-minus-profits wedge. Check: quantiles ordered; the grid's median the fan's; the bridge a level shift of every percentile; the long-run state converges to the rate (32 checks, all passing).
  7. The fan's record. Source: step 5, refitted at every year-end from 2005. Transform: the four-quarter fan against what happened. Check: coverage and RMSE recomputed from the stored errors (3 checks, all passing).
  8. What the index is quoted to earn. Source: FactSet Earnings Insight (the author's 11 Sep capture); the Valuation page's forward multiples. Transform: read, never recomputed; the margin's contribution as an identity. Check: every printed figure equals the capture; the copies equal the Valuation page's (7 checks, all passing).
  • Quoted
    Where EPS stands and what drives it, as FactSet quotes it.

    Where EPS stands, as quoted: the FactSet Earnings Insight figures are the author's capture of the report dated 11 Sep 2026, read and never recomputed (a check compares every printed figure with the file): bottom-up (operating) EPS for calendar 2026 and Q3 2026 in dollars, the Q3 estimate's revision since 30 June, earnings growth by quarter and calendar year, revenue growth, net margins, guidance counts and Q3 earnings growth by sector. The margin's contribution is read from the quotes as (1 + EPS growth) ÷ (1 + revenue growth) − 1, an identity, not an estimate. Not quoted: the season's beat rate and aggregate surprise (not in the capture) and the share count (none of the sources this page reads carries it). Shiller's series is reported (GAAP) earnings and FactSet's is operating: the two are printed side by side and never spliced.

  • One threshold
    The Profits page owns the profit share and its test.

    One owner, one threshold: the after-tax profit share and its test are the Profits page's, the threshold its computed 2015–19 average. This page copies its reading and its test — the share back at its 2015–19 average for two consecutive quarters — and checks the copy against the source page and against its own arithmetic on every run (11.91% for Q2 2026; threshold 9.8%). The halfway threshold this page printed in earlier reports (the share halfway back to its average) is retired: two pages testing one number at two thresholds is the disagreement the register listed.

  • The fan here
    Why a model-conditional object leads this page.

    The fan is read on the Scenarios page under Outlook, where every forward-looking object of the site lives, beside the regression's paths under stated nominal growth; this page keeps what is observed and estimated. The fan is labelled model-conditional wherever it appears: it is not a forecast, and its inputs are printed with it — the fitted process on national-accounts profits after tax, the assumed long-run rate of 6.20 over 8 quarters, the bridge to EPS and 20,000 simulated paths from Q2 2026.

  • Paths
    Model-conditional, not forecasts.

    The eight-quarter paths are model-conditional projections, not forecasts: each is the regression's arithmetic on a stated nominal GDP growth rate (4, 5, 6%), the dollar and WTI held at their latest monthly prints (Oct 2026 and Oct 2026, so their four-quarter changes run off the actual history and are zero from Q4 2027) and the margin gap closing in equal steps from +1.11pp to zero over eight quarters, with the regressor lagged four quarters as in the estimation. The band is ± the headline model's out-of-sample root-mean-square error since Q1 2022.

  • Spec
    The regression.

    Quarterly OLS of the four-quarter growth of the S&P 500's trailing 12-month reported earnings per share (Shiller's series at quarter-end months) on the four-quarter growth of nominal GDP (Table 1.1.5), the four-quarter percentage change of the Board's broad dollar index, the four-quarter percentage change of the WTI spot price and the margin gap (after-tax profit share of GDP less its trailing 20-quarter average) lagged four quarters; estimation window Q1 1990 – Q4 2019. Standard errors are Newey–West with Bartlett weights over 4 lags and no small-sample correction; checks reproduce coefficients, errors and R² of both estimations with statsmodels.

  • Sample
    The headline sample and the switch.

    Headline sample: Q1 1990 – Q4 2019, Q4 2008 – Q3 2010 excluded (112 quarters; R² 0.40). 8 base-effect quarters (Q4 2008 – Q3 2010: reported EPS growth −89% to +793% off the collapsed 2008–09 base) dominate the as-written fit (R² 0.12, nominal-GDP t 0.24), and its paths (+51–55% four-quarter growth) are not credible as an expectation. Excluding them is a sample choice, not a change of specification; the author can revert it with the page's single headline-sample switch. The specification as written (Q1 1990 – Q4 2019, 120 quarters; R² 0.12, constant 30.0) is printed as the sensitivity in the model table, on the fitted-values chart and as a second path, with both out-of-sample scores side by side.

  • Out of sample
    Two naive benchmarks; 2020–21 as a sensitivity.

    Out of sample: 18 quarters Q1 2022 – Q2 2026, predicted with the in-sample coefficients and the realised inputs; headline RMSE 21.8pp against 41.1pp for 'same as last year' (growth four quarters earlier) and 17.5pp for zero growth; the 2008–10 included coefficients 27.9pp. 2020–21 are excluded from the headline score and shown as a sensitivity (from Q1 2020: 26.6, 53.4 and 34.1pp over 26 quarters; 2008–10 included 37.8pp).

  • Top-down
    What the Valuation page reads.

    The topdown block is the expected earnings of the Valuation page's one-stage comparison (the one-stage column and the macro view of cross.valuation-8); that page's headline two-stage identity takes its near-term growth from the growth fan instead. The block is the headline model's four-quarter EPS growth at Q4 2027 under nominal GDP growth of 4, 5 and 6% as a low–mid–high band (+5.6% / +14.1% / +22.6%), the out-of-sample RMSE (21.8pp) and the quoted calendar-2027 consensus (+15.1%, gap −1.0pp). A check ties band.mid to paths.mid at that quarter.

  • Process
    Model-conditional, not a forecast.

    The process fan is a model-conditional distribution, not a forecast: it is what the fitted state-space model — an AR(2) around a fixed assumed rate, with jumps — implies for the next 8 quarters, simulated from its smoothed state on Q2 2026 with the estimated shock variances and the identified jump distribution. It says nothing about news after that quarter.

  • Process spec
    The state-space model.

    Unobserved components on y_t = 100 × Δ log of Shiller's trailing 12-month reported earnings per share at quarter-end months, Q1 1960 – Q2 2026 (266 quarters): y_t = μ_t + c_t + ε_t, with μ_t = μ̄ + φ(μ_{t−1} − μ̄) + η_t the long-run growth state (φ restricted to [0.80, 0.99]), c_t = ρ1 c_{t−1} + ρ2 c_{t−2} + ν_t a stationary AR(2) cycle and ε_t the observation shock. State [μ_t, c_t, c_{t−1}], transition [[φ,0,0],[0,ρ1,ρ2],[0,1,0]], intercept [(1−φ)μ̄,0,0], design [1,1,0]; φ through a logistic map, the AR(2) pair through statsmodels' stationary transform, the standard deviations through exp; μ initialised at the assumed long-run rate with variance 100 and the cycle at its stationary distribution. Estimated by maximum likelihood (statsmodels MLEModel); standard errors from the Hessian. Everything is in quarterly log points; the page annualises by multiplying by four, and cumulative growth is exp(Σ y ÷ 100) − 1.

  • Long-run rate
    One rate across Earnings and Valuation, and the fan's wedge over it.

    One long-run rate across Earnings and Valuation (the author's decision of 23 September 2026): long-run nominal GDP growth, 4.0% a year = the September SEP's longer-run median for real GDP growth (2.0) + its longer-run median for PCE inflation (2.0), projection table of 16 Sep 2026. It is an assumption, not an estimate. The growth fan's assumed long-run rate (the anchor in earlier reports) adds the 1990–2019 mean wedge of its own series over nominal GDP — 2.20 for NIPA profits after tax (the default fan, assumed rate 6.20) and 1.28 for trailing index EPS (the robustness case, assumed rate 5.28) — and holds it over the model's 8-quarter horizon only. Valuation's two-stage identity takes near-term growth from the fan (nipa-bridged fit, 2 years) and fades it linearly to the long-run rate over 6 years (a convention of this site, not estimated — printed as such), so the wedge fades to zero: held for ever, a wedge over GDP would imply a profit share rising without limit. Valuation reads its g_L from the same SEP table and checks on every run that it equals the rate published here (the Valuation page is made after this one). Its one-stage identity runs at the same rate and prints its gap to fair value at 3.00 (the September report's g, the growth a 0.85 payout sustains — a sensitivity since the one-rate decision) and 5.28 (the robustness-case fan's assumed long-run rate held for ever — trailing EPS, 4.0 + wedge 1.28) as sensitivities, not as the site's rate.

  • Assumed long-run rate
    Where the robustness case's μ̄ comes from.

    Robustness case, the trailing series: its assumed long-run rate μ̄ is 5.28 log points a year = SEP longer-run real GDP growth 2.0 + SEP longer-run PCE inflation 2.0 (September 2026 medians, projection table of 16 Sep 2026) + the EPS-minus-nominal-GDP wedge 1.28, the Q1 1990 – Q4 2019 mean of (four-quarter log growth of trailing EPS − four-quarter log growth of nominal GDP) over 120 quarters. The wedge is computed in logs because the model is: on simple percentage changes the same mean is 19.6pp, a number the 2008–10 base effects (−89% to +793%) own outright and which would put the anchor near 24% a year. Sensitivity ±0.5pp on μ̄ is in the robustness table. Like the default fan's (6.20, with the profits wedge 2.20), this assumed rate holds over the model's horizon only; the site's one long-run rate is 4.0, nominal GDP growth with no wedge.

  • Identification
    What could and could not be estimated.

    On NIPA profits after tax (the fan): estimated freely, σ_η goes to zero (0.000; φ 0.80) — the data cannot separate a slow trend from the cycle, so the trend is held at the assumed rate μ̄ and the spread of the paths is the cycle's and the jumps'; φ is then fixed at 0.95 by the rule. On trailing reported EPS (the robustness case): estimated freely, σ_η goes to zero (0.000; φ 0.80) — the data cannot separate a slow trend from the cycle, so the trend is held at the assumed rate μ̄ and the spread of the paths is the cycle's and the jumps'; φ is then fixed at 0.95 by the rule. Parameters at a variance bound have no Hessian standard error and are printed as 'at bound', never as a number; the other errors invert the Hessian over the identified parameters. An independent Kalman filter reproduces the likelihood at the estimates (−786.5988 on profits, −892.9059 on trailing EPS).

  • Jumps
    Interventions and the switching-variance sensitivity.

    Jumps: quarters whose standardised one-step prediction error in the Gaussian fit exceeds 3 in absolute value — Q4 2008, Q2 2009, Q4 2009, Q1 2010 (4 of 266, p = 1.50% a quarter). The model is re-estimated with those quarters as impulse dummies in the observation equation, and the simulation draws a jump with probability p, resampling its size from the identified prediction errors with their signs (−112, +37, +93, −58 log points); a symmetric resampling is in the sensitivity table. A second pass on the re-estimated model would flag Q4 2008, Q1 2009, Q2 2009, Q3 2009, Q1 2020, Q1 2021: the procedure is one round by specification, and those quarters stay in the sample. As a further sensitivity a two-state Markov-switching variance model (statsmodels MarkovRegression, switching variance) is fitted to y_t directly: standard deviations 3.9 and 38.1 log points, expected durations 61 and 8 quarters, smoothed probability of the high-variance state at Q2 2026 0.7%. That model is fitted to y_t itself, so its variances contain the cycle: the row is a variance sensitivity, not a decomposition.

  • Non-overlapping
    The same model on NIPA profits, and the bridge to EPS.

    The default fan is the non-overlapping fit (the author's decision of 23 September 2026; the trailing-series fan is the robustness case): the same model, the same rules and the same assumed-rate construction on the quarterly log change of NIPA corporate profits after tax with inventory valuation and capital consumption adjustments (Table 1.12 line 15, quarterly SAAR), Q1 1960 – Q2 2026 (266 quarters). Trailing earnings per share are a four-quarter moving sum, so the quarterly change of that sum shares three of its four quarters with the previous one and carries a moving-average structure by construction; profits after tax are the quarter's own flow. Anchor 6.20 log points a year = SEP longer-run real GDP 2.0 + SEP longer-run PCE 2.0 + the 1990–2019 profits-minus-nominal-GDP log wedge 2.20 (120 quarters; median 1.89). 3 quarters carry an intervention dummy (Q4 2008, Q1 2009, Q3 2020); cycle ρ1 1.30, ρ2 −0.61, period 10.8 quarters. The bridge to EPS is additive in logs: EPS four-quarter log growth = profits four-quarter log growth + −0.92 log points, the 1990–2019 mean of the difference over 120 quarters (standard deviation 47.9 log points — the bridge is a level shift of the fan, not a narrowing of it), so every percentile of the calendar-2027 fan moves by it (a check). Bridged calendar-2027 EPS growth: median +4.7% (10–90 −9.3% to +21.8%), 80% of paths below the quoted consensus, against +2.3% and 66% in the robustness case on the trailing series. The two fits share the SEP medians and the wedge construction, so the gap between their medians is the dependent variable, not the assumed rate. On this series the likelihood with the jump dummies has more than one local maximum, so the estimation is run from 29 starting values (the model's default and a fixed grid declared in the page's specification) and the highest likelihood is kept: −786.60, against −786.60 from the default start alone. The starts reached 6 maxima; the bridged calendar-2027 median at each is +4.7% (log-likelihood −786.60, 7 starts, used); +5.8% (log-likelihood −787.52, 3 starts); +5.7% (log-likelihood −787.60, 1 start); +5.8% (log-likelihood −788.84, 3 starts); +4.8% (log-likelihood −789.75, 2 starts); +5.5% (log-likelihood −791.37, 13 starts). That spread is estimation uncertainty the fan's own bands do not carry.

  • Cycle period
    What the AR(2) implies for the length of a turn.

    The cycle's period: the companion roots of c_t = ρ1 c_(t−1) + ρ2 c_(t−2) are complex when ρ1² + 4ρ2 < 0, and the pair r e^(±iθ) turns once every 2π ÷ θ quarters with cos θ = ρ1 ÷ 2√(−ρ2). On the trailing series ρ1 0.854, ρ2 −0.383: complex roots, period 7.8 quarters. On the non-overlapping series 10.8 quarters. A check recomputes each period from the argument of the root.

  • Switching variance
    The fan's second preset.

    The second fan preset simulates the fitted two-state Markov-switching model itself — y_t = c[s_t] + σ[s_t] ε_t with the estimated transition matrix (P(stay in the low-variance state) 0.984, P(stay in the high-variance state) 0.877), the regime at the origin drawn from the smoothed probabilities of Q2 2026 (high-variance state 0.7%) — rather than the unconditional two-variance mixture used in the sensitivity table. Constants 1.91 and −0.34 log points a quarter, standard deviations 3.9 and 38.1. It carries no cycle and no jump list: it is the variance-regime reading of the same series, and its calendar-2027 median is +7.9% against +2.3% for the headline fan. Check: started from its ergodic distribution (88.3% / 11.7%), the mean regime constant over 20,000 paths × 8 quarters is 1.649 against the ergodic mean 1.646; the Monte Carlo standard error of that mean is 0.0043 log points (the paths are independent, so it is the standard deviation of the 20,000 path means ÷ √n) and the tolerance is three of them rounded up to the next hundredth, 0.02.

  • Process backtest
    Recursive out of sample and the benchmarks.

    The default fan's record: at each year-end from Q4 2005 the fan's whole construction is repeated on the data to that date only — the profits-minus-GDP wedge and the EPS-minus-profits bridge averaged from 1990 to the origin (to 2019 at the latest), the anchor from them, the bound rule, the jump dummies and the fixed start grid (609 fits converged, 0 failed) — and four quarters of profits growth simulated (20,000 paths) and carried to EPS by that origin's bridge. It is scored on the same origins and target as the trailing model below: RMSE 49.8 log points against 49.4 for a fixed mean, mean error +1.6, 80% band 12 of 20 (Kupiec p 0.04). Its band carries the simulation's spread only — not the bridge's own spread or the estimation's — and the record shows what that costs. Recursive out of sample: the whole two-step estimation is repeated on the data to each year-end from 2005 to 2025 (21 refits) and the next four quarters' cumulative growth simulated (20,000 paths). 20 of those have all four quarters observed and are scored; 1 is still pending. Benchmarks on the same origins: a random walk with drift on y (ŷ_{T+h} = y_T + hδ̂, δ̂ the mean change of y in the training sample, predictive variance 30 δ̂-variances for the four-quarter sum) and a fixed mean (the training sample's mean four-quarter sum with its own sample variance). Point forecasts are each method's median, so the percentage column is the log column transformed; RMSE is reported in log points (the headline) and on simple four-quarter percentage growth for comparability with the macro regression above, and the log score is the Gaussian predictive density at the realisation on the log scale (higher is better). The random walk's four-quarter log forecast reaches 568 log points at one origin (it carries a single quarter of 2009 forward four times), so its percentage-scale error is arithmetic, not a comparable number: read that benchmark in log points. The macro regression's out-of-sample RMSE (21.8pp, Q1 2022 – Q2 2026) is a different window and a different object — quarterly four-quarter growth from realised macro inputs — so the table also prints both models on the 6 year-end quarters they share. The assumed long-run rate is today's SEP median at every origin, not the SEP of the day: the backtest tests the process, not real-time anchor selection.

  • National accounts
    NIPA tables and the profit share.

    Anchor exhibits: NIPA corporate profits with inventory valuation and capital consumption adjustments (Table 1.12 lines 13–15) and by industry (Table 6.16D: domestic financial, domestic nonfinancial, rest of the world; the NAICS-basis table starts in 2001 Q1, so its growth rates start in 2002 Q1), against Shiller's 12-month trailing reported EPS at quarter-end months. The profit share is after-tax profits ÷ nominal GDP; its norm is the 2015 Q1 – 2019 Q4 average (9.81%, σ 0.30); the margin gap is the share less its trailing 20-quarter average. Percentiles run from 1990 Q1; display from Q1 1980.

  • Dollar
    H.10 broad index, spliced.

    The Board's nominal broad dollar index (goods and services, Jan 2006 = 100) starts in January 2006; the discontinued broad index (V0. series in the H.10 package, 1973–2019) is spliced on by the ratio of the two in Jan 2006 (0.9075), so four-quarter changes before 2007 are the old index's. Quarterly values are averages of the Board's monthly averages.

  • Oil
    WTI, change only.

    WTI is EIA's monthly Cushing, Oklahoma spot price, averaged to quarters. EIA attributes the spot series to Refinitiv/LSEG and its display rights are unresolved, so the level is not printed anywhere on the site; only the four-quarter percentage change enters the regression. Latest month Oct 2026 (10/2026).

  • Shiller
    Index earnings, quoted.

    Robert J. Shiller's data (shillerdata.com, workbook dated 16 Nov 2026; the Yale copy at econ.yale.edu stopped updating in September 2023) — monthly 12-month trailing reported earnings interpolated by Shiller from S&P's quarterly figures, so the quarter-end months carry the reported values; latest earnings month Sep 2026. This workbook restates 3 quarters the page had printed from the one before it: Q2 2026 is $262.59, was $295.39; Q1 2026 is $241.89, was $261.72; Q4 2025 is $235.81, was $240.63. Academic data shown with its citation: the trailing reported earnings per share is printed in dollars (the author's decision of 21 September 2026 allows the level with attribution; earlier issues printed growth rates only), and the index's price and dividend levels are not used on this page. A check reproduces the workbook's real-earnings column from its nominal and CPI columns.

  • Consensus
    Quoted, not downloaded.

    Consensus is quoted, not downloaded: the forward P/E quotes are the Valuation page's (FactSet Earnings Insight as quoted in the September 2026 issue; 11 Sep 2026: 19.1×, carried to 16 Nov 2026 at 18.66× with SPY), and the calendar-year EPS growth points are the September report's dated quotes (Calendar 2026 EPS growth +31.6%, Calendar 2027 EPS growth +15.1%, FactSet 11 Sep 2026), checked against the author's 11 Sep capture file. They are bottom-up operating earnings, drawn at the fourth quarter of each year, where the four-quarter growth of trailing earnings is the calendar-year growth; the model's dependent variable is reported (GAAP) earnings.

  • Dates
    Data dates.

    Data dates: the NIPA flat file released 4 Nov 2026 (data to Q2 2026); Shiller's earnings to Sep 2026; H.10 daily to 13 Nov 2026, monthly to Oct 2026; SEP projection table 16 Sep 2026; FactSet figures quoted 11 Sep 2026; data read on 17 Nov 2026.

  • Licence
    Whose numbers these are.

    NIPA, H.10 and the FOMC projection tables are published by federal agencies for reuse. Shiller's earnings series is shown in dollars with its citation; EIA's WTI spot enters as a change only. FactSet's figures appear as the author's dated quotes, which is the only form the site holds them in. The regression, the growth process, the fan and the paths are computed here.

Source: U.S. Bureau of Economic Analysis; Federal Reserve Board; EIA; Robert J. Shiller; FactSet Earnings Insight; Macro Review calculations · data to Q2 2026cross.earnings-17

Exhibit 24 · Where a series stops, and the rule applied

Each gap below is a period the source itself does not cover on the basis the page uses.

BEA Table 6.16D, Q1 1980 – Q4 2000: the NAICS-basis industry table starts in 2001 Q1 (Tables 6.16B/C carry the SIC-basis history); shown as missing, not spliced · Shiller earnings, Oct 2026 – Nov 2026: earnings not yet reported for the latest months; growth ends at the last quarter-end month with a value · BEA national accounts, Q3 2026: profits for the quarter are not yet published (due 25 Nov 2026); the national-accounts series end at Q2 2026 and index earnings are set against profits there; the page's models use earnings to that quarter

Source: U.S. Bureau of Economic Analysis; Robert J. Shiller · data to Q2 2026cross.earnings-18

Before you read

Macro Review is a personal research site, still being built.

Nothing on it is investment advice, a recommendation or a solicitation.

Check every figure for yourself. Each names its publisher — confirm it there before you rely on it.

Figures can be revised, delayed or wrong.

Cones, fans, scenarios, fair values and “what followed” tables are conditional on the model each page names: they describe, they do not predict.

No warranty is given; use is at your own risk.