of 3 of six baskets pass the a year ahead
How is each number made, and how far can each model be trusted?
In brief
Which models pass the , what the macro model adds, and what the regime tests can measure- The of 3 of the six baskets pass the stability rule a year ahead: tech, semis and growth.→ Exhibit 1
- The gap to the justified level, as the DCF gives it, fails as a one-year target: its error is 1.20 times a random walk's.→ Exhibit 2
- With the macro path known, the model fits the S&P 500's measure with an R² of 0.91.→ Exhibit 3
- Only two regimes held enough quarters since 2017 to measure: 20 for "only the dollar hedges", 15 for "bonds hedge".→ Exhibit 4
- The reconstructed forward P/E sits within 2.1% of FactSet's tech figure in the median month.→ Exhibit 6
| What would change this view | Trigger | Today | Decided by |
|---|---|---|---|
| Tech's normal level stops beating a random walk | tech's error ÷ random walk's at or above 1 at the next re-estimation | 0.88 (13 Nov 2026)not crossed | the next quarterly re-estimation |
| The S&P 500's normal level passes | the S&P 500's error ÷ random walk's below 1 at the next re-estimation | 1.05 (13 Nov 2026)not crossed | the next quarterly re-estimation |
| The reconstruction drifts from FactSet | the median monthly gap to FactSet's tech figure above 3% | 2.1% (114 months)not crossed | every update |
The reasoning behind each point
- RuleThe normal levels of 3 of the six baskets pass the stability rule a year ahead: tech, semis and growth.
Staples' error is 1.73 times a random walk's, the worst of the six.
- JustifiedThe gap to the justified level, as the DCF gives it, fails as a one-year target: its error is 1.20 times a random walk's.
Adjusted for rates, the same gap passes, narrowly, like the gap to each basket's own normal.
- MacroWith the macro path known, the model fits the S&P 500's measure out of sample with an R² of 0.91.
With the macro path forecast, the R² is 0.26, so macro maps scenarios but carries no proven edge as a forecast.
- RegimeOnly two regimes held enough quarters since 2017 to measure: 20 for "only the dollar hedges", 15 for "bonds hedge".
"Nothing hedges" and "bonds and the dollar hedge" held one quarter each, too few to measure.
- HistoryThe reconstructed forward P/E sits within 2.1% of FactSet's tech figure in the median month.
For the S&P 500 the median gap is 2.0%; the history before 13 Nov 2026 is estimated, not consensus.
Next: Sector earnings — the measure the macro model fits
The P/E level and the
Each P/E signal and each basket's against the Fair value board's stability ruleExhibit 2 · The gap to the justified level, as the DCF gives it, fails as a one-year target: its error is 1.20 times a random walk's.
| Signal | Slope (t) | First half | Second half | Error ÷ random walk | Verdict |
|---|---|---|---|---|---|
| Gap to the basket's own normal level | −0.78 (-7.4) | −0.62 (-2.0) | −0.96 (-6.0) | 0.96 | passes |
| Gap to the justified level, rate-adjusted | −0.73 (-9.1) | −0.65 (-1.6) | −0.84 (-6.1) | 0.96 | passes |
| Gap to the justified level, as the DCF gives it | −0.09 (-2.0) | −0.19 (-2.4) | −0.20 (-1.1) | 1.20 | fails |
About this exhibit
Exhibit 1 · The normal levels of 3 of the six baskets pass the stability rule a year ahead: tech, semis and growth.
| Basket | Own normal: error ÷ random walk | Justified, rate-adjusted | Verdict |
|---|---|---|---|
| Tech | 0.88 | 0.81 | passes the rule |
| Semis | 0.91 | 0.95 | passes the rule |
| Growth | 0.88 | 0.91 | passes the rule |
| S&P 500 | 1.05 | 0.99 | fails the rule |
| Value | 1.18 | 1.13 | fails the rule |
| Staples | 1.73 | 1.91 | fails the rule |
About this exhibit
The stretch model against its time-series benchmark
How well the macro model fits the S&P 500's measure , with the macro path known and forecastExhibit 3 · With the macro path known, the model fits the S&P 500's measure out of sample with an R² of 0.91.
| Macro path used | R² | p | Reading |
|---|---|---|---|
| Realised macro | 0.91 | 0.09 | a scenario mapping |
| Forecast macro | 0.26 | 0.13 | no proven edge as a forecast |
About this exhibit
The regime tests
Quarters in each regime since 2017 and the measured effect of a regime on theExhibit 4 · Only two regimes held enough quarters since 2017 to measure: 20 for "only the dollar hedges", 15 for "bonds hedge".
| Regime | Quarters | Effect on the normal multiple |
|---|---|---|
| Only the dollar hedges | 20 | the reference |
| Bonds hedge | 15 | −7% to −10% (t −1.5 to −2.1) |
| Nothing hedges | 1 | not measured: too few quarters |
| Bonds and the dollar hedge | 1 | not measured: too few quarters |
About this exhibit
How forward EPS history is estimated, and what is held until scored
The steps from filings to forward EPS, the check against FactSet, every macro specification tried, and the unscored shares
Exhibit 5 · Steps from filings to each basket's forward EPS, 2017–26
| Step | What is done |
|---|---|
| 1 | A model of next-twelve-month EPS is fitted across the index's members on the consensus of 13 Nov 2026. |
| 2 | It is applied to each past month's filed fundamentals from SEC EDGAR. |
| 3 | Each sector's total is scaled to FactSet's sector forward P/E for that month. |
| 4 | Members are weighted into the six baskets on today's SPY holdings. |
Exhibit 6 · The reconstructed forward P/E sits within 2.1% of FactSet's tech figure in the median month.
About this exhibit
Exhibit 7 · Every macro specification tried for the S&P 500's measure, out of sample
| Specification | R² | p | R² without 2008 | p |
|---|---|---|---|---|
| Direct, seven macro levels | -0.59 | 0.29 | -1.39 | 0.27 |
| Direct, three macro levels | -0.05 | 0.16 | -0.21 | 0.26 |
| Direct, seven macro changes over four quarters | -0.36 | 0.21 | -1.19 | 0.25 |
| Structural, realised macro | 0.88 | 0.08 | 0.64 | 0.04 |
| Structural, realised macro, two lags | -0.41 | 0.24 | -1.75 | 0.52 |
| Structural, forecast macro | 0.38 | 0.13 | -0.29 | 0.21 |
| Structural, forecast macro, two lags | -0.08 | 0.13 | -0.96 | 0.15 |
| Structural, three macro series, forecast | 0.26 | 0.13 | -0.34 | 0.33 |
| Structural, three macro series, realised | 0.91 | 0.09 | 0.79 | 0.06 |
Exhibit 8 · Simulation shares held until scored: the regime's next move and a falling capex year
| What the simulation counts | Share of paths |
|---|---|
| The regime holds for a year | 32% |
| It turns to "nothing hedges" | 27% |
| It turns to "bonds hedge" | 41% |
| Hyperscaler capex falls in 2028 | 9% |
| Hyperscaler capex falls in 2029 | 20% |
| Hyperscaler capex falls in 2030 | 25% |
Appendix
What each test is and what the method is not7 checks, all pass
Every test behind the Sectors & styles pages in one place: the stability rule on each P/E signal and basket, the macro model's out-of-sample record, the regime tests, and how forward earnings before today are estimated.
Exhibit 9 · Definitions
- RuleThe stability rule.
The Fair value board's stability rule: the slope of the next year's change in the multiple on the gap has the sign theory gives (the gap closes) with |t| ≥ 2, both halves of the sample agree in sign, and the target re-estimated at each quarter-end from Q3 2021 to Q4 2025 beats a random walk's error (ratio below 1). A basket passes when its signal passes and its own ratio is below one.
- MacroThe macro model and its two scores.
The macro model sets the S&P 500's trailing measure on industrial production growth, the high-yield spread and the curve, quarterly since 1993. Out of sample it is scored twice: with the macro path as it happened, and with the path a vector autoregression forecast at each origin.
- RegimeThe regime tests.
Each quarter takes the regime that held most of its weeks on the four-regime vote. The effect is the regime's coefficient in the normal-multiple model, in two specifications; a regime with one quarter is not measured.
- HeldWhat is held until scored.
The simulation's shares of paths by what the regime does next, and the capex model's share of paths with a falling capex year, are model output with no record yet: they are printed here, labelled, and nowhere else.
- HistoryThe reconstruction.
Forward EPS before 13 Nov 2026 is a reconstruction (D-2026-11-17-OR6); the page holds it to FactSet's sector forward P/E at every update.
- LimitsWhat the method is not.
No revenue, margin or share-count path; one average return on equity per basket; no charge for capex's cash cost (D-2026-11-17-OR12).
- DatedWhy the records do not move.
The records are the canvas run's of 13 Nov 2026, published as they are under the author's decision of 18 Nov 2026.
How the numbers are made
Yahoo Finance consensus + SEC EDGAR company facts + State Street SPY holdings + FactSet Earnings Insight: forward 12-month P/E by sector (monthly, read from the report's chart) ──▶ the canvas run of 13 Nov 2026: today's consensus by member, the history estimated from filings and scaled to FactSet's sector P/E, the models fitted and tested ──▶ canvas_baskets canvas_baskets ──▶ every file checked against the import's SHA-256 manifest; the forward P/E history held to FactSet's sector figures again ──▶ the checked import the checked import ──▶ each P/E signal and each basket's normal level held to the stability rule: theory's sign with |t| ≥ 2, both halves agreeing, a re-estimated target beating a random walk ──▶ the stability rule the checked import + Federal Reserve Board: industrial production and the high-yield spread ──▶ the macro model's out-of-sample fit for the S&P 500's measure, with the macro path known and with it forecast ──▶ the macro record the checked import ──▶ quarters in each regime since 2017 and the regime's measured effect on the normal multiple; the simulation's unscored shares held apart ──▶ the regime tests the stability rule + the macro record + the regime tests ──▶ tiles, takeaways, falsifier rows and section titles generated from the numbers ──▶ the page
Sources
- licensedYahoo Finance consensus
Source: Yahoo Finance analyst estimates for each S&P 500 member, read on 13 Nov 2026 · https://finance.yahoo.com
- publicSEC EDGAR company facts
Source: SEC EDGAR, the members' filed EPS · https://www.sec.gov/edgar
- licensedState Street SPY holdings
Holdings: State Street Global Advisors (SPDR S&P 500 ETF Trust) · https://www.ssga.com/library-content/products/fund-data/etfs/us/
- licensedFactSet Earnings Insight: forward 12-month P/E by sector (monthly, read from the report's chart)
Source: FactSet Earnings Insight · https://insight.factset.com/topic/earnings
- publicFederal Reserve Board: industrial production and the high-yield spread
Source: Federal Reserve Board · https://www.federalreserve.gov/releases/g17/
Reproduction checks
| Check | Published | Computed | Tolerance | Result |
|---|---|---|---|---|
| Every file of the canvas import of 2026-11-13 has its manifest's SHA-256 (18 files; files that differ) | 0 | 0 | 0 | pass |
| Tech's forward P/E in the canvas against FactSet's Information Technology figure: median absolute gap over 114 months within 3% (worst -15.6% in 2020-02; %) | 0 | 2.0705 | 3 | pass |
| Staples's forward P/E in the canvas against FactSet's Consumer Staples figure: median absolute gap over 114 months within 3% (worst +9.5% in 2026-01; %) | 0 | 1.3631 | 3 | pass |
| S&P 500's forward P/E in the canvas against FactSet's S&P 500 figure: median absolute gap over 115 months within 3% (worst +18.4% in 2020-03; %) | 0 | 1.9663 | 3 | pass |
| The own-normal signal's quarter-ends tested out of sample (basket-quarters) | 108 | 108 | 0 | pass |
| The regime quarters sum to the panel's quarter-ends since 2017 (37) | 37 | 37 | 0 | pass |
| The simulation's regime-path shares sum to one | 1 | 1 | 0 | pass |
- Yahoo Finance consensus — Source: Yahoo Finance analyst estimates for each S&P 500 member, read on 13 Nov 2026
- SEC EDGAR company facts — Source: SEC EDGAR, the members' filed EPS
- State Street SPY holdings — Holdings: State Street Global Advisors (SPDR S&P 500 ETF Trust)
- FactSet Earnings Insight: forward 12-month P/E by sector (monthly, read from the report's chart) — Source: FactSet Earnings Insight
- Federal Reserve Board: industrial production and the high-yield spread — Source: Federal Reserve Board