CAPE alone answers one question: how expensive are stocks against their own earnings power? It cannot answer the question investors actually face: expensive compared to what? The alternative to owning stocks is owning bonds. The Excess CAPE Yield asks the two-sided question: take the CAPE earnings yield (1/CAPE — at a CAPE of 40.90 stocks "yield" 2.44%), then subtract what ten-year Treasuries yield after inflation. What is left is the equity premium on Shiller’s convention — the margin stocks offer over the alternative.
A worked example, on the latest month the measure has: in August 2026 the ten-year Treasury paid 4.68% while the CAPE yield was 2.44% — nominally, bonds appear to out-yield stocks. But earnings are claims on real activity; they inflate with prices, while a bond coupon is fixed. Subtracting trailing ten-year inflation (3.35%) leaves a real bond yield of 1.33% — and the comparison flips: stocks out-yield real bonds by +1.12 points. Using the nominal rate would misread every high-inflation decade in the record: the 1970s look bond-favorable nominally while bonds were losing purchasing power every year.
On the other three lenses, high readings mean expensive. This lens inverts: ECY is the margin stocks offer over bonds, so a THIN margin is the stretched state. Low ECY = expensive. The tier colors follow the meaning, not the direction of the number: the bottom of the distribution (≤p10) is deep red — "thin vs bonds" — and the top (>p90) is deep green — the wide-margin states history produced after crises, 1974–82 and late 2008. The percentile pill on the lens page carries the same inverted tint: today’s p11.9 renders warm where a CAPE reading that high would render cold.
The chain, August 2026: CAPE 40.90 (multpl, end-of-month) → CAPE yield 2.44%. GS10 4.68% minus trailing ten-year CPI-NSA inflation 3.35% → the convention real rate 1.33%. ECY = 2.44 − 1.33 = +1.12 points, the 11.9th percentile of the measure’s own 73-year history — the pale red tier, in a spell running since May 2025, the longest pale-red spell in the register. Stocks still out-yield real bonds — the zero line is +1.12 points below — but the margin is thinner than in roughly 88% of the record.
Zero is not a percentile convention — it is arithmetic: below zero, ten-year Treasuries out-yield stocks in real terms. In 73 computed years the state has occurred once: May 1999 → Nov 2000, 19 months, spanning the dot-com top. The S&P was +7.8% one year after the crossing and −15.8% three years after. One episode is a fact, not a base rate — we list it and stop.
Five tiers, fixed percentile boundaries — ≤p10, p10–25, p25–75, p75–90, >p90 — plus the zero overlay. Every percentile is computed on an expanding window: each month judged only by history knowable then — 1975 is ranked against 1953–1975, never against 2026. The boundaries were frozen before any outcome data was examined, and no boundary other than these percentiles and zero exists. The register collapses tier eras by the sitewide convention (at least two consecutive months); the raw monthly census behind it — 83 boundary events, 21 lasting a single month — is stated here because monthly grain chatters at percentile boundaries, and hiding that would overstate the register’s crispness.
Shiller’s convention derives the real rate from realized trailing inflation; the market’s TIPS yield states one directly (the site’s series since 2003). They are different instruments and they disagree in a regime-dependent way: after inflation spikes the trailing window drags the convention below the market — today 1.33% vs 2.40%, a gap of −1.07 points — wide, though short of the paired record’s extreme of −2.27 in December 2008. Both render on the lens page with the gap stated, not reconciled: on the market’s rate, today’s ECY would be thinner still.
From the registers: the widest margins cluster in 1974–1982 — ECY above +8 for long stretches while inflation crushed real bond yields — and briefly in December 2008. The thinnest cluster at 1965–69, 1996–2001 (the only below-zero spell), and the current era. The tier medians are published as computed, not as a story: read from deep red to deep green they rise with the inversion at one year (+11.3% to +23.6%), and do not rise at three (+27.9% to +51.1%). The extreme tiers hold 9 eras and 7 eras respectively; we publish the counts and decline the inference.
Latest-basis, labeled: the CAPE leg revises (earnings backfill; source-basis differences in the newest months), so readings shown are today’s revised history — a point-in-time ECY is not constructible and we say so rather than simulate one. October 2025 has no reading: the BLS published no CPI for that month, and this site publishes only observations that exist — Shiller’s file bridges that month with an estimated CPI and publishes an ECY on it; we decline that conduct. His pre-1953 history rides his own rate splice and CPI cells — it appears only in the expanded chart, dashed and attributed, context never series. Forward returns are S&P total return · Shiller monthly convention (see "The total-return basis" on the Confluence methodology).
The measure: Shiller, Black & Jivraj, "Making Sense of Sky-High Stock Prices," Project Syndicate, November 30, 2020; carried since as a column of Shiller’s living dataset. The construction here: his file’s own arithmetic — GS10 minus the 120-month ratio-annualized CPI-NSA change — reproduced against his published column to ±0.0003 at the verification probes. Legs: FRED GS10 (1953-04→) and CPIAUCNS (1913→); CAPE via multpl; TIPS DFII10 (2003→). Descriptive-only — no validated status exists for this lens.