Built 2026-10-10. Sources: CurrentMarketValuation.com model pages (public data to Jun 30, 2026), FRED, Fed Z.1 via FRED, Shiller (Yale), multpl. All charts rebuilt from primary data to Sep-Oct 2026. Gray bands = NBER recessions. This is analysis, not investment advice.
Valuation Fragility Score — 99th percentile since 1970. Higher than the maximum reached in the 12 months before any recession peak in the sample (2000: +1.64 was the previous record). Model-implied next-10-year S&P 500 return: ≈ +0.4%/yr price-only (VFS regression); CAPE regression on real total return: ≈ -0.1%/yr. Wide error bands — these are center-of-distribution estimates, not forecasts.
Cycle Inflection Score — essentially no inflection in progress. Sahm 0.00 (Sep), claims 197k, HY OAS 3.12%, CFNAI -0.04, LEI-R 6m change +2.3% annualized with only 12% of components falling. The one live clock: the curve re-steepened in Dec 2024 — historically recession followed within 1-11 months; we are ~22 months past it, the longest wait in the sample.
Translation: maximum fragility, zero trigger. Every long-horizon model says the next decade pays poorly from here; every short-horizon inflection gauge says the expansion is intact today. Those are not contradictory — they answer different questions, and confusing them is the core flaw in how the CMV models get used.
Your LEI screenshot shows a change pattern: the index rolls over and declines for months before each recession (gray band). The Buffett Indicator shows a level pattern: it grinds to extremes and stays there for years. Levels forecast returns over the next decade (Buffett deviation vs. forward 10y returns: R² = 0.58; real-price deviation: R² = 0.71; CAPE vs. forward 10y real total return: R² = 0.33 — overlapping-sample R², so treat as ranking evidence, not precision). Changes forecast turning points over the next year. Averaging the two families into one score — CMV's members-only aggregate — destroys the timing information in both.
VFS (Valuation Fragility Score) = mean z-score of Buffett trend deviation, log CAPE, real S&P trend deviation, negative Excess CAPE Yield, log price-to-sales. It sets expectations and position-size ceilings. It is not a sell signal: it was above +1 for years in the late 1990s.
CIS (Cycle Inflection Score) = 10 derivative/breadth components (table in §5), each scored 0 / 0.5 / 1, averaged to 0-100. It is deliberately deaf to valuation. Backtest (§4): CIS ≥ 50 fired in 9 episodes since 1970 — 7 matched recession peaks with leads of -6 to +9 months, and the 2 unmatched are the 1970 and 1982 in-recession aftershocks of the double-dip era (same cycles, counted once by the matcher). Crucially, CIS peaked at only 45 during the entire 2022-24 LEI/yield-curve alarm and 35 when Sahm falsely triggered in Aug 2024 — breadth voting refused both false alarms that fooled the famous single indicators.
The regime rule: danger = VFS reached ≥ +1.0 SD at any point in the prior 12 months and CIS ≥ 50 now. (Simultaneous readings almost never coincide — by the time inflection confirms, prices have usually already fallen and VFS has collapsed; in 56 years the simultaneous version fires zero times.) In the sequential regime's 10 months since 1970, the median forward-12-month maximum drawdown was -15.7% vs. -1.7% for all months. Today: VFS condition met, CIS nowhere near — the regime is armed, not triggered.
CIS (red, 0-100, left) with its 50 trigger line; VFS (blue, SD, right); S&P 500 (gray, log, right, toggleable). Watch 2000: VFS peaks first, CIS crosses a year later — the sequence is the signal.
Buffett rebuilt from Fed Z.1 public corporate equities / GDP (level runs hotter than CMV's Wilshire version — 287% vs. 244% — because the numerator is broader; the trend deviation agrees: +81% vs. CMV +77-81%). CAPE: Shiller monthly to Sep 2023, multpl points after; 41.87 on Oct 9, 2026. P/S from 2000 only — its short history is itself a criticism.
LEI-R is my reconstruction of the Conference Board LEI from its 8 FRED-available components (hours, claims, permits, orders, sentiment, S&P, 10Y-FedFunds spread, financial conditions), rebased 2000 = 100. The pattern you spotted is real in the level — and the 6-month annualized change is the tradable version of it. Note 2022-23: LEI-R fell hard (rate shock) while Sahm never moved — dispersion the headline LEI hides.
Baa-Aaa corporate spread (the long-history credit gauge; ICE HY OAS is license-limited to ~3 years on FRED — current reading 3.12%), VIX, Michigan sentiment, policy uncertainty. Sentiment is the current outlier: Michigan at 51.7 (deeply pessimistic) beside a 15.4 VIX and 3.1% HY spreads (calm).
Claims, temp-help employment YoY (the classic first-to-be-cut series), permits YoY, payroll 6m annualized growth. Recessions are labor events; until these roll, LEI/curve alarms stay unconfirmed.
| Signal | Recessions matched | False-positive episodes | Median trigger vs. peak | Episodes |
|---|---|---|---|---|
| Sahm Rule ≥ 0.50 (original) | 8 | 2 | -3 mo | 1970-03→1969-12 (-3m) · 1974-07→1973-11 (-8m) · 1980-02→1980-01 (-1m) · 1981-11→1981-07 (-4m) · 1990-10→1990-07 (-3m) · 2001-06→2001-03 (-3m) · 2003-07→none · 2008-02→2007-12 (-2m) · 2020-04→2020-02 (-2m) · 2024-08→none |
| Sahm ≥ 0.35 (lowered threshold) | 8 | 2 | -2 mo | 1970-02→1969-12 (-2m) · 1974-03→1973-11 (-4m) · 1980-01→1980-01 (+0m) · 1981-11→1981-07 (-4m) · 1990-09→1990-07 (-2m) · 2001-04→2001-03 (-1m) · 2003-06→none · 2007-12→2007-12 (+0m) · 2020-04→2020-02 (-2m) · 2024-04→none |
| Yield curve inversion start (10Y-3M, material ≥ -0.10pp, 1982+) | 4 | 1 | +10 mo | 1989-06→1990-07 (+13m) · 2000-08→2001-03 (+7m) · 2006-08→2007-12 (+16m) · 2019-08→2020-02 (+6m) · 2022-11→none |
| Yield curve re-steepening after inversion | 4 | 1 | +5 mo | 1989-08→1990-07 (+11m) · 2001-02→2001-03 (+1m) · 2007-06→2007-12 (+6m) · 2019-10→2020-02 (+4m) · 2024-12→none |
| Credit widening: Baa-Aaa +50bp from 12m low | 7 | 7 | -3 mo | 1970-08→1969-12 (-8m) · 1974-10→none · 1979-12→1980-01 (+1m) · 1980-12→1981-07 (+7m) · 1982-03→none · 1984-07→none · 1990-12→1990-07 (-5m) · 2001-12→2001-03 (-9m) · 2008-03→2007-12 (-3m) · 2011-10→none · 2015-10→none · 2019-01→2020-02 (+13m) · 2020-04→none · 2022-07→none |
| LEI-R 6m annualized change < -3% (rebuilt LEI) | 6 | 7 | +5 mo | 1973-03→1973-11 (+8m) · 1974-05→none · 1978-11→1980-01 (+14m) · 1980-03→1981-07 (+16m) · 1981-09→none · 1984-08→none · 1988-01→none · 1990-09→1990-07 (-2m) · 2000-12→2001-03 (+3m) · 2002-10→none · 2008-10→none · 2020-03→2020-02 (-1m) · 2022-06→none |
| CIS ≥ 50 (new composite, 10 components) | 7 | 3 | +3 mo | 1970-10→none · 1974-05→1973-11 (-6m) · 1979-10→1980-01 (+3m) · 1980-10→1981-07 (+9m) · 1982-04→none · 1990-08→1990-07 (-1m) · 2000-12→2001-03 (+3m) · 2007-08→2007-12 (+4m) · 2020-03→2020-02 (-1m) · 2025-04→none |
Matching: a trigger matches the nearest NBER peak within -9 to +24 months, each peak once; episodes clustered within 12 months. Positive = triggered before the peak. Samples differ by data availability (curve from 1982, credit proxy full sample). 2020 (COVID) was exogenous — no model led it and none should be graded on it. Sahm's 2003 'false positive' is the jobless-recovery echo of the 2001 cycle. The Dec 2024 re-steepening remains unmatched at ~22 months — either the first clean miss for that signal or a still-open clock; CMV's dashboard handles this by keeping recession risk 'Very High' for a fixed window, which is a rating convention, not evidence.
| Sahm momentum ≥0.20/0.35 | ○ Off (0) |
| Initial claims YoY | ○ Off (0) |
| LEI-R 6m change | ○ Off (0) |
| Curve resteepening window / inverted | ○ Off (0) |
| Baa-Aaa widening, 3m | ● Full (1.0) |
| Temp-help jobs YoY | ◐ Half (0.5) |
| Building permits YoY | ○ Off (0) |
| Chicago Fed CFNAI 3mma | ○ Off (0) |
| Payroll growth, 6m annualized | ○ Off (0) |
| Financial conditions tightening, 3m | ○ Off (0) |
Only temp-help is even half-on (YoY growth under 1.5%). Everything else is off — including the curve component, because the 12-month post-resteepening window from Dec 2024 has expired in this scoring, which is a judgment call worth debating (§7, Q2).
| buffett dev | +2.34 SD |
| log cape | +1.85 SD |
| real dev | +2.38 SD |
| neg ecy | +1.02 SD |
| log ps | +1.69 SD |
Every leg is stretched simultaneously — that unanimity, not any single ratio, is the fragility story. Before the 2001 peak the same table maxed at +1.64 average; before 2007 only +0.30 (that recession was a credit event in a fairly-valued market, and the S&P still halved — VFS measures return risk, not recession risk).
| Model | CMV reading | Criticism (theirs, paraphrased — plus data notes) | Fix / improvement applied here |
|---|---|---|---|
| Buffett Indicator | 244% of GDP, +2.6 SD (CMV, Jun 30) | Ignores rates; US-listed firms earn globally while GDP is domestic, so the ratio drifts up structurally. | Judge only the deviation from its own exponential trend, never the raw level; cross-check with a rates-adjusted variant (ECY). Rebuilt from Fed Z.1 public equities / GDP: 287% level (broader numerator) but +81% above trend, matching CMV's +77 to +81%. |
| CAPE (P/E10) | 39.7 (CMV); 41.87 Oct 9, 2026 (multpl/Shiller) | Ignores rates; 10y averaging lags real profitability shifts; tech-heavy index may sustain higher multiples. | Pair CAPE with Excess CAPE Yield (1/CAPE minus real 10Y yield) so valuation is always rates-relative. ECY now ≈ +0.2pp ex-post (TIPS-based ≈ -0.5pp): thin compensation either way. |
| Price-to-Sales | 3.2, +2.3 SD (CMV) | Ignores profitability entirely; sector mix shift to high-margin tech raises the sustainable ratio; history only starts 2000. | Use it only as a corroborating z-score inside a composite, never standalone; flag the short, single-regime history. |
| Interest Rate Model | +1.93 SD composite (CMV) | No criticisms section on the page; implicitly assumes the 10Y reverts to its 5.81% long-run mean. | Replace mean-reversion-to-history with a neutral-rate (r*) band and show the answer as a range, not a point. |
| S&P 500 Mean Reversion | +88% above trend, +2.4 SD (CMV) | No criticisms section; the page itself concedes deviations can persist for decades. Full-sample trend also peeks at the future. | Keep for context only; my rebuild (Shiller price spliced with FRED, CPI-deflated) gives +93% above trend. Never a timing input. |
| Earnings Yield Gap | -0.27pp, 'Fairly Valued' (CMV) | No criticisms section; crude Fed-model heuristic — equities should normally out-yield bonds for risk, and it uses trailing earnings. | Use forward earnings yield minus real (TIPS) yield, and treat a merely-zero gap at record multiples as fragile, not fair. |
| Q Ratio | ~+2.6 SD / ~81% above its long-run average, Strongly Overvalued (Jun 30, 2026 — from page metadata only) | The CMV page itself is broken server-side (fatal error, empty body — verified via live browser on 2026-10-10), so its criticisms section, if any, is unreachable. On the merits: Tobin's Q shares the Buffett flaws plus replacement-cost measurement error, and intangibles are badly undercounted in modern balance sheets. | Covered conceptually by the Buffett rebuild (same market-value-vs-economy logic; note its +81% trend deviation echoes the Q figure, coincidence of construction, not confirmation); omitted from the composite to avoid double-counting. |
| Yield Curve (10Y-3M) | +0.51pp (CMV dashboard); +1.05pp Oct 2026 monthly avg (FRED) | QE and global safe-haven demand can depress long yields and fake inversions; inversion alone says nothing about severity or timing. | Two fixes, both backtested below: (1) materiality filter (monthly mean below -0.10pp for 2 months — removes the 2025 'touch' inversion at -0.06pp, a T-bill/debt-ceiling artifact); (2) trade the re-steepening, not the inversion — re-steepening led the peak by +1 to +11 months in all 4 sampled cases. |
| Sahm Rule | 0.13pp (CMV, May); 0.00 Sep 2026 (FRED); 0.03 computed | No criticisms section on CMV's page; by construction it confirms recessions ~3 months in rather than forecasting. Its spotless record also quietly ended: it triggered in Aug 2024 (0.53) with no recession — a labor-supply-driven rise (immigration/participation), as Claudia Sahm herself flagged. | Never use it alone or lower the threshold (0.35 triggers at the same time or later relative to the peak in my test — it buys no lead). Require breadth confirmation (claims, payroll momentum, temp help) — exactly what the CIS does. CIS read only 35/100 in Aug 2024 and never confirmed. |
| Leading Economic Indicator | 97.40, momentum +0.75 (CMV, Apr 30); your screenshot: 99.10 Jun 30, 2026 weekly view | Composite hides component dispersion; noisy and revised; the post-2022 slide ran for two years with no recession — a rare prolonged false positive, likely because it over-weights goods/manufacturing and rate-sensitive housing vs. services and labor. | Rebuilt it from 8 available components (LEI-R) and added a diffusion check (share of components falling). LEI-R 6m change did cross -3% in Jun 2022 (false positive, honestly recorded below) — but its diffusion and the labor components disagreed, which is why the composite must vote, not obey one index. |
| State Coincidence | 5 states declining (CMV, May) | No criticisms heading; SD bands on a 0-50 count are illustrative only; identifies recessions but is not tradable. | Use as breadth confirmation inside a composite (diffusion logic), exactly the role diffusion plays in the CIS design. |
| Margin Debt | $1,416B, +1.52 SD YoY change (CMV, May) | No explicit criticisms section found. Raw levels grow with the market; FINRA reporting changes distort long history. | Keep CMV's fix (scale by market value, use 12m change) and treat it as a fragility (VFS-style) input, not a trigger: leverage peaks before tops but with wildly variable lags. |
| Junk Bond Spreads | 2.75% OAS vs 5.19% avg (CMV); ICE HY OAS 3.12% Oct 2026 | No criticisms heading; spreads reflect risk-appetite cycles and structural reach-for-yield demand, so low levels can persist. ICE data on FRED is license-limited to ~3 years, which silently truncates naive backtests. | Use the change, not the level: widening velocity (+50bp from the 12m low on the long Moody's Baa-Aaa proxy) — but know its record: it mostly confirms stress during recessions rather than leading peaks (below). Tight spreads = fragile, not safe. |
| VIX | 16.45 (CMV); 15.4 Oct 8, 2026 (FRED) | No criticisms section; measures expected magnitude of movement, not direction. | Sentiment context only. A low VIX raises fragility (cheap hedges, crowded calm) — it belongs on the fragility side of the ledger. |
| Economic Policy Uncertainty | 221, +1.55 SD (CMV); ~354 early Oct 2026 (daily avg) | No criticisms section; measures policy uncertainty, not economic performance. | Contrarian context only; currently the loudest pessimist in the room while hard data disagrees — a divergence to log, not act on. |
| Consumer Confidence (Michigan) | 49.50, -2.8 SD (CMV, Jun); 51.7 Aug 2026 | No criticisms section; sentiment swings more than fundamentals and has been depressed by price-level (not inflation-rate) frustration since 2022 while spending held up. | Down-weight post-2020 sentiment vs. hard data. It is the single biggest disagreement in the current dashboard: -2.8 SD pessimism alongside 3.1% HY spreads and 197k jobless claims. |
Data files & reproducible scripts: ~/workspace/valuation-model/ (data/ = raw FRED/multpl/Shiller pulls; work/analysis.py, analysis2.py = every number above; work/cmv-catalog.md = the full paraphrased audit notes). Prepared for Michael Davis.
