Macro ReviewHow to read
Cross Asset · Fundamentals · Long-run growth

The valuation assumes 4.0% ; estimated from data since 1990, 5.0%

What terminal growth does the valuation assume, and is it consistent with the data?

BEA · 4 Nov 2026 · data to Q2 2026SEP 16 Sept 2026 · Valuation inputs 16 Nov 2026Company filings to Q2 2026Updated 17 Nov 2026
Terminal growth the valuation assumes4.0%long-run real growth 2.0 + inflation 2.0 · , 16 Sep 2026
Nominal GDP, growth over four quarters+6.3%terminal growth 4.0% · Q2 2026
Long-run growth state, nominal GDP5.6%2.9%95% band, with parameter uncertainty8.4%the model's trend · Q2 2026 · parameters taken as known 3.0–8.3assumption inside the band
Profits' growth over GDP, a year+2.2pp1990–2019 average · standard error 1.5not distinguishable from zero
Index EPS against profits, a year−0.9pp1990–2019 average · standard error 7.1not distinguishable from zero
Long-run rate estimated from the data, nominal GDP5.0%2.5%95% likelihood interval8.8%since 1990 · φ free 4.8 · assumed 4.0% · LR p 0.63assumption not rejected

In brief

What the numbers say, and what would change the reading
  • The model puts nominal GDP's trend growth at 5.6% in Q2 2026, and the 4.0% assumption sits inside the band around it.→ Exhibit 1
  • Retention and return on equity have sustained 1.7% growth a year since Q2 2015, below the 4.0% assumed.→ Exhibit 3
  • Nominal GDP grew 6.3% in the four quarters to Q2 2026, against the 4.0% the valuation assumes.→ Exhibit 2
  • Estimated rather than assumed, nominal GDP's long-run rate since 1990 is 5.0% with φ held, 4.8% with φ free.→ Exhibit 4
  • Per-share earnings outside finance outgrew nominal GDP by 7.7 points a year from Q2 2015.→ Exhibit 9
What would change this view
What would change this viewTriggerTodayDecided by
Nominal GDP's trend returns to the assumed ratethe model's trend (the filtered state) ≤ 4.0% a year5.6 (Q2 2026)not crossedthe next BEA GDP estimate for Q3 2026 (not on the release calendars)
Profits stop outgrowing the economyprofits less nominal GDP, four-quarter log growth, below 0+10.6 (Q2 2026)not crossedBEA corporate profits, Q3 2026, 25 Nov 2026
Index EPS falls behind profits by more than usualEPS less profits, four-quarter log growth, below −0.9 (the 1990–2019 average)−0.1 (Q2 2026)not crossedBEA corporate profits, Q3 2026, 25 Nov 2026; Shiller's earnings for Q3 2026
The Committee moves the assumed ratelonger-run real GDP + PCE medians ≠ 4.04.0 (16 Sep 2026)not crossedthe next projections, 9 Dec 2026
The reasoning behind each point
  • Trend
    The model puts nominal GDP's trend growth at 5.6% in Q2 2026, and the 4.0% assumption sits inside the band around it.

    The filtered long-run growth state (continuously compounded), 95% band 3.0 to 8.3. On nominal GDP: estimated freely, φ goes to its upper bound (0.99) while σ_η stays positive (0.090) — the state moves, and fixing φ at 0.95 imposes faster reversion to μ̄; it is not a state the data cannot find. It has been above the assumed rate since Q4 2020. The same model on real GDP and on prices puts real growth's trend at 2.0 (band 1.3–2.8) and the deflator's at 5.2 (band 4.4–6.1), 7.3 together — so most of the gap to 4.0 is prices. The LR test of the bound rule's φ on nominal GDP is 6.69 (p 0.010; free φ 0.990, at its bound): the data reject the fixed 0.95.

  • Three legs
    Retention and return on equity have sustained 1.7% growth a year since Q2 2015, below the 4.0% assumed.

    The fundamental leg is retention × return on equity, averaged quarter by quarter (error 0.7); counting buybacks as payout leaves little retained, and per-share growth adds a share-count effect of 2.2 a year. The statistical leg is Long-run growth's estimate from nominal GDP, 5.0% (error 1.9); the assumed rate is not rejected there (p 0.63).

  • Growth
    Nominal GDP grew 6.3% in the four quarters to Q2 2026, against the 4.0% terminal growth the valuation assumes.

    In log points, the scale the model and the gaps use: nominal GDP +6.1, national-accounts profits after tax +16.7 and the index's trailing reported earnings per share +16.6 over the four quarters to Q2 2026. Since Q1 1960 the three have averaged 6.2, 7.2 and 6.5 log points a year.

  • Estimated rate
    Estimated rather than assumed, nominal GDP's long-run rate since 1990 is 5.0% with φ held, 4.8% with φ free.

    The same model with μ̄ freed: the profile likelihood over μ̄ on 146 quarters, φ at 0.95 as in the page's fit, the same jump dummies; standard error 1.89. The likelihood-ratio test of the assumed 4.0% gives 0.23 (p 0.63): the data do not reject it at 5%. The 95% likelihood interval is 2.50–8.75. With φ estimated too, on this sample it goes to its lower bound (0.80), the fastest reversion the model allows, and μ̂ to 4.8% (p 0.24 against 4.0). The 4.0% itself is the Committee's longer-run real GDP 2.0 + PCE inflation 2.0 (16 Sep 2026), the rate every other fit here reverts to.

  • Wedge
    Per-share earnings outside finance outgrew nominal GDP by 7.7 points a year from Q2 2015.

    Revenue grew +1.2 points a year against GDP, the margin added +4.4 and the share count +2.2. The index's own EPS ran +4.0 points a year ahead of GDP over the same years. The filings cover today's members only: a company that grew into the index is counted from the start, which raises measured growth.

  • Profits and EPS
    For profits and EPS the data cannot separate trend from cycle, so the model holds their trend at the assumed 4.0%.

    On profits after tax: 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 index EPS: 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 cycle absorbs their growth above the assumed rate: it averages +0.8 and +2.0 log points a year over the sample and stands at +5.4 and +28.8 in Q2 2026. That is a statement about the model on these series, not evidence that profits trend at 4.0: the gaps below measure what they have done.

  • Gaps
    Profits outgrew nominal GDP by 2.2 points a year in 1990–2019, but with a standard error of 1.5 the gap is not distinguishable from zero.

    Averages of four-quarter log growth differences over 120 quarters with Newey–West standard errors: profits less GDP +2.20 (1.54), index EPS less profits −0.92 (7.06, not distinguishable from zero either), EPS less GDP +1.28 (7.63) — the Earnings page's gaps, reproduced. A gap held for ever would put profits on a path to exceed the economy; over 1960–1989 profits less GDP averaged −0.6. In the year to Q2 2026: profits +10.6 over GDP, EPS −0.1 over profits.

Next: Valuation — the model that fades to this rate

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

The model's trend for nominal GDP is 5.6%; the 4.0% assumption sits inside its band

The model's long-run growth state for nominal GDP with its band, against the assumed rate, and four-quarter growth of nominal GDP, profits and index earnings since 1960

Exhibit 1 · Growth and the model's trend with its band against the assumed rate

% a year, quarterly since Q1 1960: growth (thin), the model's trend and band (bold), the cycle, the rate the data choose; dashed, the assumed 4.0%
About
About this exhibit
Switchable chart (GDP, profits, EPS), quarterly: quarterly growth annualised, the model's trend with its 95% band, the cycle, and, for GDP, the long-run rate the data choose, against the assumed rate.
Nominal GDP: the model's trend against the assumed rate
% a year (continuously compounded), quarterly since Q1 1960: growth, the model's trend with its 95% band, and the cycle.
The thin line is the quarterly growth the model is fitted to (annualised). The bold line is the trend the model read from it at each quarter, shaded to its 95% band, and the cycle is the rest. The trend reverts to the assumed rate (4.0) with persistence φ 0.95. In Q2 2026 it stands at 5.6 (3.0 to 8.3). It has been above the assumed rate since Q4 2020.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.longrun-3

Exhibit 2 · Four-quarter growth of nominal GDP, profits and index EPS since 1960

Four-quarter growth in log points (%), quarterly, Q1 1960 to Q2 2026; dashed, the assumed terminal growth.
About
About this exhibit
Three lines, quarterly over the whole record: four-quarter growth of nominal GDP, national-accounts profits after tax and S&P 500 trailing EPS, in log points; dashed, the assumed terminal growth.
Growth of the economy, profits and index earnings since 1960
Growth over four quarters in log points, quarterly, Q1 1960 to Q2 2026; dashed, the assumed terminal growth.
100 × the change in the log over four quarters — the scale the model and the gaps use, where a fall of a half and a doubling are the same distance ().
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller · data to Q2 2026cross.longrun-1

Is the assumed rate consistent with the fundamentals?

Growth reinvestment can sustain (retention × return on equity), the estimated from the economy, and the rate assumed, each with its error

Exhibit 3 · Long-run growth read by reinvestment, the GDP estimate and the assumption

Growth reinvestment sustains, the rate estimated from nominal GDP, and the rate the valuation assumes, % a year, each with its error, to Q2 2026.
LegQ2 2026Mean or estimateStandard error95% intervalWhat it is
Fundamental: retention × return on equityQ2 2026; the mean since Q2 2015 with its Newey–West error8.21.70.720.3 to 3.1estimation error
Statistical: the long-run rate estimated from nominal GDPprofile likelihood, Q1 1990–Q2 2026; the trend today 5.6% (3.0 to 8.3)–5.01.892.5 to 8.8estimation error
Assumed: the Committee's longer-run medians addedreal GDP 2.0 + PCE inflation 2.0, 16 Sep 2026–4.0––assumed values
About this exhibit
Table: three estimates of long-run growth, % a year, each with its error: reinvestment (retention × return on equity), the rate estimated from nominal GDP, and the rate the valuation assumes.
The long-run rate, three ways
Growth reinvestment sustains, the rate estimated from nominal GDP, and the rate the valuation assumes, % a year, each with its error.
The fundamental leg measures the growth of total earnings; per-share growth adds the share-count effect, 2.2 a year on average since Q2 2015. Buybacks count as payout here, which leaves little retained. The statistical leg is the Long-run growth page's estimate; the assumed rate is not rejected against it (p 0.63). The fundamental mean less the assumed rate: −2.3 points.
Source: U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; the Long-run growth page; Macro Review calculations · data to Q2 2026cross.payout-11

Assumed 4.0% a year; estimated from the data since 1990, 5.0%

The Committee's longer-run real growth and inflation added, and every page that uses the same number; the same model with the rate estimated, the bound rule tested, and the trend split into real growth and prices

Exhibit 4 · Nominal GDP's long-run rate assumed and estimated, and its trend's parts

Nominal GDP, % a year, Q1 1990–Q2 2026: assumed 4.0% against the rate the data choose, with errors; below, its trend's real and price parts.
About
Long-run rateEstimateStandard error95% likelihood intervalLR test of 4.0%p
Assumed — every other fit on the page4.0––––
Estimated, φ fixed at 0.95 (the page's model)5.001.892.50 to 8.750.230.629
Estimated, φ estimated (0.800)4.750.632.75 to 7.251.380.239
Trend in Q2 2026The model's trend95% bandIts assumed rate
Nominal GDP5.63.0 to 8.34.0
Real GDP2.01.3 to 2.82.0
Prices (GDP deflator)5.24.4 to 6.12.0
Real + prices, fitted apart7.3–4.0

The bound rule's φ, tested: with φ free the log-likelihood is -303.68 against -307.02 at 0.95 — LR 6.69, p 0.010; free φ 0.990, at its bound. The deflator is anchored at the Committee's longer-run PCE inflation; the GDP deflator is not PCE.

About this exhibit
Two tables: nominal GDP's long-run rate assumed and estimated two ways, with errors and the test of the assumed rate; then the model's latest trend for nominal GDP, real GDP, prices and the two summed, each against its assumed rate.
The long-run rate estimated, not assumed
Nominal GDP, Q1 1990 to Q2 2026 (146 quarters): the rate that maximises the model's likelihood, against the assumed 4.0%; then the trend split into real growth and prices. % a year, log points.
The same model and jump dummies, with μ̄ freed: its profile likelihood, the standard error from the profile's curvature, the 95% likelihood interval, and the likelihood-ratio test of the assumed rate (χ² with one degree of freedom). A persistent trend (φ near 1) pins the long-run rate loosely, so its interval is wide. estimation error
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.longrun-13

Exhibit 5 · The assumed terminal growth, its parts, and where it is used

The Committee's longer-run medians, 16 Sept 2026, added: an assumption, not an estimate; below the line, the pages using it.
About
ItemValueWhat it is
Longer-run real GDP growth2.0the Committee's median, 16 Sept 2026
Longer-run PCE inflation2.0the same table
: long-run nominal growth4.0%the two added — 1.00 log points a quarter in the model
Earnings: its long-run rate4.0the same number
Valuation: the two-stage identity's g_L4.0the same number
Valuation: g_L in its fair-value inputs4.0the same number
What is priced: its SEP medians added4.0the same number
Earnings: the profits fan's rate, less its gap6.20 − 2.20a gap held for the fan's eight quarters only
Earnings: the trailing-EPS fan's rate, less its gap5.28 − 1.28the robustness case
Source: Federal Reserve (Summary of Economic Projections) · data to 16 Sept 2026cross.longrun-2

Profits outgrew GDP by 2.2 points a year in 1990–2019, within the noise

Profits growth less nominal GDP growth, and index EPS growth less profits growth, with their averages over stated windows and standard errors

Exhibit 6 · Growth gaps: profits less nominal GDP, and index EPS less profits

Gaps in four-quarter growth, log points (pp), quarterly since 1961: profits less nominal GDP, and index EPS less profits; dashed, the first's 1990–2019 average.
About
About this exhibit
Two lines, quarterly since 1961: four-quarter growth of profits less nominal GDP, and of index EPS less profits, log points; dashed, the 1990–2019 average of the first.
How far profits and earnings run ahead of the economy
Gaps in four-quarter growth, log points, quarterly since 1961: profits less nominal GDP, and index EPS less profits.
The second gap is the bridge the Earnings page uses from its profits fan to EPS. In the year to Q2 2026: +10.6 and −0.1.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller · data to Q2 2026cross.longrun-5

Exhibit 7 · Average growth gaps by window, with standard errors

Average gaps in four-quarter growth, pp a year, by window since 1960, with standard errors; below, trends at Q2 2026 against the assumed rate.
About
WindowQuartersProfits less GDPEPS less profitsEPS less GDP
1990–2019Q1 1990 – Q4 2019120+2.20SE 1.54−0.92SE 7.06+1.28SE 7.63
1960–1989Q1 1960 – Q4 1989120−0.61SE 1.85−0.72SE 2.26−1.33SE 1.76
2020 to the latest quarterQ1 2020 – Q2 202626+1.83SE 1.82+1.22SE 6.11+3.05SE 6.71
1960 to the latest quarterQ1 1960 – Q2 2026266+0.90SE 1.12−0.62SE 3.40+0.28SE 3.60
The year to Q2 20261+10.57−0.13+10.43
The model's trend, Q2 2026Trend95% bandLess the assumed rateMean growth since Q1 1960Cycle, sample mean
Nominal GDPthe trend moves5.653.0 to 8.3+1.656.20+0.13
NIPA profits after taxtrend and cycle not separable: held at the assumed rate4.004.0 to 4.0±0.007.16+0.83
S&P 500 trailing EPStrend and cycle not separable: held at the assumed rate4.004.0 to 4.0±0.006.54+1.98
About this exhibit
Two tables: average growth gaps (profits less GDP, EPS less profits, EPS less GDP) by window with standard errors; then each series' model trend in the latest quarter against the assumed rate.
The gaps on average, with standard errors, and each series' trend
Average gaps in four-quarter growth, log points a year, by window, with Newey–West standard errors; a gap within two errors of zero is noise.
Newey–West standard errors over 4 lags (the four-quarter overlap). 1990–2019 is the Earnings page's sample; its averages are that page's gaps.
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Macro Review calculations · data to Q2 2026cross.longrun-6

Which sectors carry the index's growth, and where the valuation takes it

Q3 2026 earnings growth for the seven sectors FactSet's report quotes, and the hand-off to the Valuation page, where enters the fair value

Exhibit 8 · Which sectors carry the index's growth, as FactSet quotes them

Q3 2026 earnings growth, year over year, for the 7 sectors FactSet's report quotes, 11 Sept 2026.
SectorQ3 2026 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%
About this exhibit
Growth only: sector weights and earnings levels are in none of the sources, so the sectors cannot be summed to the index or given a process of their own.
Source: FactSet Earnings Insight · data to 11 Sept 2026cross.longrun-7

Why do per-share earnings outgrow the economy?

Per-share earnings growth since 2017 split into revenue against GDP, the margin and the share count, beside the index's own earnings per share

Exhibit 9 · Per-share earnings growth against GDP split into its parts

Average growth, log points (%) a year, Q2 2015 to Q2 2026 (eleven years), companies outside finance; the index's own EPS below.
PartA yearWhat it is
Revenue6.3sales growth
Margin+4.4the change in net income ÷ revenue
Share count+2.2the falling share count
Per-share earnings12.9the three added
Nominal GDP5.2the economy, the same years
Revenue less nominal GDP+1.2sales ahead of the economy
The wedge: per-share earnings less nominal GDP+7.7revenue's gap + margin + share count
The index's EPS less nominal GDP, Q2 2015 to Q2 2026+4.0S&P 500 trailing EPS 9.3 against GDP 5.2, the Long-run growth page's series
About this exhibit
Table, outside finance: average yearly growth of revenue, margin and share count adding to per-share earnings, against nominal GDP, and the index's own EPS against GDP.
Per-share earnings against nominal GDP, in parts
Average growth a year, log points, Q2 2015 to Q2 2026 (11 years), companies outside finance; the index's own EPS below.
Per-share growth = revenue + the change in the margin + the falling share count, each a year's average of the quarterly links; less nominal GDP's growth over the same years, revenue's gap to GDP + the margin + the share count = the wedge. Today's members only: a company that grew into the index is counted from the start, which raises measured growth.
Source: U.S. Securities and Exchange Commission (EDGAR); State Street Global Advisors; U.S. Bureau of Economic Analysis; Robert J. Shiller; Macro Review calculations · data to Q2 2026cross.profitability-16

Exhibit 10 · Growth since Q2 2015: per-share earnings, revenue and the economy

Cumulative growth from Q2 2015, log points, quarterly to Q2 2026.
About this exhibit
Each line adds up its quarterly growth from zero, so the distance between two lines is how much one outgrew the other, in , which add across quarters where percentages do not.
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-17

Next: terminal growth enters the fair value on Valuation — each half-point on it moves the gap to fair value by about 4.1 points, the required return held.

Robustness, and the Earnings model reproduced

The long-run state when one thing changes; the Earnings page's model re-run against its published parameters

Exhibit 11 · Robustness: the model's trend when one thing changes

The trend at the last quarter of each sample with its 95% band, then φ, σ_η and the log-likelihood of each fit.
FitGDP: stateProfits: stateEPS: stateGDP: the fitProfits: the fitEPS: the fit
The page's fitsφ by the rule, Q1 1960 – Q2 2026, μ̄ 1.0005.653.0 to 8.34.004.0 to 4.04.004.0 to 4.0φ 0.95 · σ_η 0.193log-likelihood -307.0φ 0.95 · σ_η 0.000log-likelihood -788.3φ 0.95 · σ_η 0.000log-likelihood -893.1
Persistence estimated freelythe same jump dummies, the default start5.823.7 to 7.94.004.0 to 4.04.004.0 to 4.0φ 0.99 · σ_η 0.089log-likelihood -303.7φ 0.80 · σ_η 0.000log-likelihood -788.1φ 0.80 · σ_η 0.000log-likelihood -892.6
1960–1989steps one to three from the default start5.942.9 to 9.04.024.0 to 4.04.014.0 to 4.0φ 0.95 · σ_η 0.233log-likelihood -156.4φ 0.95 · σ_η 0.000log-likelihood -355.2φ 0.95 · σ_η 0.000log-likelihood -283.9
1990 to the latest quartersteps one to three from the default start4.903.2 to 6.64.014.0 to 4.04.004.0 to 4.0φ 0.95 · σ_η 0.081log-likelihood -133.6φ 0.95 · σ_η 0.000log-likelihood -435.5φ 0.95 · σ_η 0.000log-likelihood -533.0
The anchor converted strictlyμ̄ 0.981 a quarter (100 × ln 1.04 ÷ 4)5.653.0 to 8.33.923.9 to 3.93.923.9 to 3.9φ 0.95 · σ_η 0.196log-likelihood -307.2φ 0.95 · σ_η 0.000log-likelihood -788.4φ 0.95 · σ_η 0.000log-likelihood -893.1
About this exhibit
Log points a year. For the strict conversion the trend is read against 3.92. other models the 95% bands are the trend's own estimation error estimation error
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.longrun-10

Exhibit 12 · The Earnings page's model, run again on this page

The Earnings page's two fits on the same data: every coefficient, standard error and log-likelihood against that page's.
Parameterthe default fan's fit, NIPA profitsthe robustness case, trailing EPS
PublishedRe-run herePublishedRe-run here
φ — persistence of the long-run state0.950fixed0.950fixed0.950fixed0.950fixed
σ_η — long-run state shock0.000at bound0.000at bound0.000at bound0.000at bound
ρ1 — cycle, first lag1.303SE 0.2281.303SE 0.2280.854SE 0.0690.854SE 0.069
ρ2 — cycle, second lag-0.608SE 0.177-0.608SE 0.177-0.383SE 0.066-0.383SE 0.066
σ_ν — cycle shock0.997SE 0.3900.997SE 0.3906.919SE 0.3006.919SE 0.300
σ_ε — observation shock4.205SE 0.2414.205SE 0.2410.000at bound0.000at bound
Log-likelihood-786.599-786.599-892.906-892.906
μ̄, log points a year6.201 = anchor + wedge +2.2015.281 = anchor + wedge +1.281
Largest difference0.0000.000
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.longrun-11

Appendix

The model, its parameters, the assumed rate and its copies, the gaps, sources and checks153 checks, all pass
About this page

The terminal growth the valuation assumes — nominal growth of 4.0% a year, the Committee's longer-run medians added, an assumption rather than an estimate — tested against the data: the same trend-and-cycle model on nominal GDP, national-accounts profits and the index's earnings per share, and how far profits and earnings have run ahead of the economy. Beside it, the growth the companies' retained earnings can sustain at their return on equity, from their own filings: the same rate read three ways. And why per-share earnings have outgrown the economy, in parts: revenue, the margin and the falling share count. The rate then enters the fair value on the Valuation page.

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
SymbolMeaningFromAssumed or estimated
yty_tnominal GDP, national-accounts profits after tax and the index's trailing EPS, each in turn, 100 × the quarterly change in logsBEA NIPA Tables 1.1.5 and 1.12; Robert J. Shiller–
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 = 4.0 a year, 1.00 a quarter, the same on all three seriesFOMC 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

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 on each series. Identification — On nominal GDP: estimated freely, φ goes to its upper bound (0.99) while σ_η stays positive (0.090) — the state moves, and fixing φ at 0.95 imposes faster reversion to μ̄; it is not a state the data cannot find. On profits after tax: 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 index EPS: 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 parameter table follows (cross.longrun-4).

The long-run rate estimated, the bound rule tested, the trend split

ℓp(μˉ)=max⁡θ ℓ(θ; μˉ),μˉ^=arg⁡max⁡μˉℓp(μˉ),se⁡(μˉ^)=[−ℓp′′(μˉ^)]−1/2\ell_p(\bar\mu) = \max_{\theta}\ \ell(\theta;\ \bar\mu), \qquad \hat{\bar\mu} = \arg\max_{\bar\mu} \ell_p(\bar\mu), \qquad \operatorname{se}(\hat{\bar\mu}) = \big[-\ell_p''(\hat{\bar\mu})\big]^{-1/2}
LRμˉ=2 [ℓp(μˉ^)−ℓp(4.0)]∼χ12,LRϕ=2 [ℓ(ϕ^)−ℓ(ϕ=0.95)]∼χ12LR_{\bar\mu} = 2\,\big[\ell_p(\hat{\bar\mu}) - \ell_p(4.0)\big] \sim \chi^2_1, \qquad LR_{\phi} = 2\,\big[\ell(\hat\phi) - \ell(\phi = 0.95)\big] \sim \chi^2_1
Δln⁡GDPtnom=Δln⁡GDPtreal+Δln⁡Pt,Pt=GDPtnom/GDPtreal×100\Delta\ln \mathrm{GDP}^{nom}_t = \Delta\ln \mathrm{GDP}^{real}_t + \Delta\ln P_t, \qquad P_t = \mathrm{GDP}^{nom}_t / \mathrm{GDP}^{real}_t \times 100
SymbolMeaningFromAssumed or estimated
ℓ(θ;μˉ)\ell(\theta;\bar\mu)the model's Kalman-filter log-likelihood at a given long-run rate, θ the other parametersmaximum likelihood–
μˉ^\hat{\bar\mu}the long-run rate the data choose, nominal GDP Q1 1990 – Q2 2026, φ at the page's 0.95profile likelihood, a grid then a bounded refinementestimated
ϕ^\hat\phiφ estimated with the page's jump dummies, on the full samplemaximum likelihood (the better of two starts)estimated
GDPtreal\mathrm{GDP}^{real}_t, PtP_treal GDP and the implicit price deflatorBEA NIPA Tables 1.1.6 and 1.1.5 (checked against Table 1.1.9)–

The profile's curvature is taken at ± 0.25; the 95% likelihood interval is the set of grid points within χ²₁(0.95) ÷ 2 of the maximum. φ's free estimate sits at its upper bound, so the test is against the bounded alternative. The trend split fits the same model separately to real growth (anchored at the SEP's longer-run real growth, 2.0) and to the deflator (anchored at its longer-run PCE inflation, 2.0 — the GDP deflator is not PCE); separate fits need not add to the nominal fit's trend.

The gaps between profits, EPS and the economy

wˉX,Y=1n∑t(Δ4ln⁡Xt−Δ4ln⁡Yt)×100,se⁡(wˉ) Newey–West, 4 lags\bar w_{X,Y} = \frac{1}{n}\sum_{t} \big(\Delta_4 \ln X_t - \Delta_4 \ln Y_t\big)\times 100, \qquad \operatorname{se}(\bar w)\ \text{Newey–West, 4 lags}
wˉEPS,GDP=wˉΠ,GDP+wˉEPS,Π\bar w_{\mathrm{EPS},\mathrm{GDP}} = \bar w_{\Pi,\mathrm{GDP}} + \bar w_{\mathrm{EPS},\Pi}
SymbolMeaningFromAssumed or estimated
Πt\Pi_tprofits after tax with inventory and capital-consumption adjustmentsBEA NIPA Table 1.12 line 15–
GDPt\mathrm{GDP}_tnominal GDPBEA NIPA Table 1.1.5–
EPSt\mathrm{EPS}_tthe index's trailing 12-month reported earnings per shareRobert J. Shiller–
wˉ\bar wthe average gap in four-quarter log growth over a window; 1990–2019 is the Earnings page's sampleestimated hereestimated

A regression of each difference on a constant, Newey–West errors over 4 lags (the four-quarter overlap), no small-sample correction. A gap whose mean is within 1.96 standard errors of zero is described as not distinguishable from zero.

Exhibit 13 · The trend-and-cycle model on three series: parameters and standard errors

Maximum likelihood, Q1 1960 – Q2 2026, 266 quarters each; Hessian standard errors, none for a parameter at a bound.
About
ParameterNominal GDPNIPA profits after taxS&P 500 trailing EPS
φ — persistence of the long-run state0.950fixed0.950fixed0.950fixed
σ_η — long-run state shock0.193SE 0.0400.000at bound0.000at bound
ρ1 — cycle, first lag1.302SE 0.1891.275SE 0.2420.854SE 0.065
ρ2 — cycle, second lag-0.666SE 0.200-0.566SE 0.192-0.382SE 0.063
σ_ν — cycle shock0.174SE 0.0751.076SE 0.4236.923SE 0.301
σ_ε — observation shock0.602SE 0.0384.199SE 0.2440.000at bound
φ estimated freely0.990σ_η 0.0900.800σ_η 0.0000.800σ_η 0.000
Cycle: one turn every9.7 quarters11.2 quarters7.8 quarters
Jump quarters (a dummy each)Q2 1978, Q2 2020, Q3 2020Q4 2008, Q1 2009, Q3 2020Q4 2008, Q2 2009, Q4 2009, Q1 2010
Log-likelihood-307.02-788.27-893.07
Starting points (the best kept)292929

What the data can say. On nominal GDP: estimated freely, φ goes to its upper bound (0.99) while σ_η stays positive (0.090) — the state moves, and fixing φ at 0.95 imposes faster reversion to μ̄; it is not a state the data cannot find. On profits after tax: 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 index EPS: 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 long-run state printed is the filtered one (what the model knew at each quarter), with a 95% band from its filtered variance; where σ_η is at zero the state sits at the assumed rate by construction and the band is only the filter's start-up funnel, so none is drawn. The gaps section measures what profits and EPS have actually done.

About this exhibit
μ̄ = 1.00 log points a quarter (the assumed rate). φ is fixed when, estimated freely, it reaches a bound or the trend's shock collapses — the Earnings page's rule. estimation error
Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); Macro Review calculations · data to Q2 2026cross.longrun-4

Exhibit 14 · Definitions

  • Model-conditional
    What the last open section is, and is not.

    Model-conditional, not a forecast: the last open section applies the Valuation page's two-stage identity, at that page's expected inputs copied and checked, with the long-run rate moved by ±0.5 and ±1.0pp, and grows trailing earnings along the identity's fade. It says what the identity implies under stated inputs; it does not say where earnings or the index will go.

  • Assumed rate
    One long-run rate, stated, and who owns it.

    One nominal long-run growth rate for the whole site: the September SEP's longer-run median for real GDP growth (2.0) plus its longer-run median for PCE inflation (2.0), 4.0% a year, projection table of 16 Sep 2026. It is an assumption — the Committee's view of where nominal growth settles — and the page states it rather than estimating it. The same number is the Earnings page's long-run rate, the base its two growth fans add their wedges to (6.20 and 5.28 less 2.20 and 1.28), the g_L of the Valuation page's two-stage identity and What is priced's two medians added; a check compares each copy with the projection table read here on every run. Who owns it: the Valuation page's g_L is the rate the fair value uses and is read by this page, so this page is made after Valuation and takes it from there — Valuation reading this page's rate would make each wait for the other.

  • Units
    Log points, and how the anchor enters the model.

    The model is estimated on quarterly log growth, 100 × Δ log. The anchor enters it as 4.0 log points a year, 1.00 a quarter — the conversion the Earnings page applies to the same number, so the two pages' models share their anchor exactly. Read strictly as 4.0% simple growth it is 3.92 log points a year (0.981 a quarter); the robustness table refits at that value. Charts and wedges on the page are in log points, where a fall of a half and a doubling are the same distance; the tile for nominal GDP's four-quarter growth is the percentage change a reader would find in BEA's release.

  • The model
    One state space, one procedure, three series.

    One model, three series. y_t = μ_t + c_t + ε_t (+ intervention dummies), with μ_t = μ̄ + φ(μ_(t−1) − μ̄) + η_t the long-run growth state reverting to the anchor (φ in [0.80, 0.99]), c_t = ρ1 c_(t−1) + ρ2 c_(t−2) + ν_t a stationary cycle and ε_t the observation shock — the Earnings page's state space, priors and bounds, and its procedure: estimate freely; if φ reaches a bound or σ_η collapses, fix φ at 0.95; take the quarters whose standardised one-step error exceeds 3 as jumps and re-estimate with a dummy for each; do that from the default start and a fixed grid of 28 others and keep the highest likelihood. Settings are read from the Earnings page's specification. Sample Q1 1960 – Q2 2026, 266 quarters, for all three. The long-run state printed is the filtered one (what the model knew at each quarter), with a 95% band from its filtered variance; an independent Kalman filter reproduces the likelihood, the states and the variance.

  • Identification
    What each series can and cannot tell the model.

    What the data can say. On nominal GDP: estimated freely, φ goes to its upper bound (0.99) while σ_η stays positive (0.090) — the state moves, and fixing φ at 0.95 imposes faster reversion to μ̄; it is not a state the data cannot find. On profits after tax: 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 index EPS: 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 long-run state printed is the filtered one (what the model knew at each quarter), with a 95% band from its filtered variance; where σ_η is at zero the state sits at the assumed rate by construction and the band is only the filter's start-up funnel, so none is drawn. The gaps section measures what profits and EPS have actually done.

  • Reproduction
    The Earnings page's model, re-run.

    The Earnings page's own specification, re-run here with the same functions: its default fan's fit (NIPA profits, μ̄ = the anchor + the profits wedge 2.201, with the start grid) and its robustness case (trailing EPS, μ̄ = the anchor + the EPS wedge 1.281, the default start) return every published coefficient, standard error and log-likelihood to the three decimals published (largest differences 0.000 and 0.000), with the same jump quarters. The EPS line on this page is that model with the wedge taken out of μ̄.

  • Gaps
    Growth differences and their errors.

    The wedges are differences of four-quarter log growth (100 × Δ4 log): profits after tax less nominal GDP, index EPS less profits (the bridge the Earnings page uses to carry its profits fan to EPS) and index EPS less nominal GDP, which is the sum of the other two. Each window's mean is a regression on a constant with Newey–West errors over 4 lags — the four-quarter overlap — and no small-sample correction, checked against statsmodels. 1990–2019 is the Earnings page's wedge sample and reproduces its published wedges.

  • Hand-off
    Terminal growth on the Valuation page.

    Each half-point on terminal growth moves the Valuation page's gap to fair value by about 4.1 points, the required return held; the sensitivity itself lives on the Valuation page.

  • Path
    The earnings path over the fade.

    The earnings path: growth falls linearly from g_S to g_L over the 6-year fade, year t's rate is the average of that line over the year, g_S + (g_L − g_S)(t − ½) ÷ 6, and the year after grows at g_L. The growth above g_L summed over the fade is H(g_S − g_L), the identity's own term — a check. It starts from Shiller's trailing reported EPS ($295.52, Q3 2026); the Valuation identity prices trailing earnings, so the path is reported earnings, not FactSet's operating estimates.

  • Sectors
    Growth only, as quoted.

    Q3 2026 earnings growth for the 7 sectors in the author's capture of FactSet's report (11 Sep 2026), copied from the Earnings page and checked. The capture carries growth only: sector weights and earnings levels are in none of the sources, so the sectors cannot be summed to the index or modelled here.

  • Robustness
    One departure at a time.

    Three departures, one at a time. Persistence estimated freely, with the same jump dummies. The sample split at 1990 Q1 (the start of the Earnings wedge sample), each half estimated with steps one to three from the model's default start — the procedure the Earnings page runs on its trailing series; on a half-sample the start grid finds degenerate maxima on the cycle's stationarity boundary with every shock variance at zero, a spike in the likelihood rather than an estimate — and the long-run state at each half's last quarter: 1960–1989: GDP 5.9, Profits 4.0, EPS 4.0; 1990 to the latest quarter: GDP 4.9, Profits 4.0, EPS 4.0. And the anchor converted strictly. None of them is the page's estimate.

  • Not modelled
    What this page does not estimate.

    Not modelled: sector earnings processes (no weights or levels are in the sources); the share count and buybacks (no source the site holds carries them). The long-run rate is assumed in every fit but one: the estimated-rate test frees it on nominal GDP since 1990.

  • Dates
    Data dates.

    Data dates: the NIPA flat file released 4 Nov 2026 (data to Q3 2026; the file of 2026-11-17); Shiller's earnings to Q3 2026 (the workbook of 2026-11-17); SEP projection table 16 Sep 2026; the Valuation page's inputs 16 Nov 2026; FactSet quoted 11 Sep 2026; data read on 17 Nov 2026.

  • Licence
    Whose numbers these are.

    NIPA and the FOMC's projections are published by federal agencies for reuse. Shiller's earnings series is shown with its citation. FactSet's sector growth rates are the author's dated quotes. The model, the wedges and the identity's sensitivity are computed here.

Source: U.S. Bureau of Economic Analysis; Robert J. Shiller; Federal Reserve (Summary of Economic Projections); FactSet Earnings Insight; Macro Review calculations · data to Q2 2026cross.longrun-12

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.