Method unchanged since Sep 29, 2026 · what changed then · full methodology below
Conditional market history across the site's instruments at once: which historical months read the same as today on four defined states, and what the S&P 500 did through and after each matching episode. It makes and tests no predictive claim; no validated status exists for it. We present the record; we do not assert predictive skill.
A month matches today only if it reads the same on ALL FOUR states. The four are the State of the Economy regime family, the Forward Stress trajectory, whether the month contains a Capitulation Monitor firing day, and the CAPE valuation tier. The regime family collapses the eleven regime labels to stress / stress-adjacent / benign, and the trajectory reads deteriorating / stable / improving. The valuation tier is expensive p95+ / neither / cheap bottom-quintile, on the point-in-time percentile basis. These four are matched because all four exist for every month since August 1981 — matching on shorter-history series would delete 1981–1998 from the record. Everything else on the page is context, shown but never matched on. The family word describes the regime's distance from contraction, not the market's safety. Today the register is benign-family throughout, and its own May 2007 row was followed by a drawdown of −48.2% at its deepest within the next three years, and stood at −23.8% three years on.
Today's four states are read live — each from its home feature's current surface, classified by the same rules the historical record uses. The analog search side is the settled month record: the monthly rows from August 1981 to the latest inked month, which never move. Now speaks live; history speaks the record.
The analogy: a stock's last-traded price versus the exchange's settled close. Today's states are the last trade — current, and still moving. The month record is the settled close — final, and the only thing history is made of.
Two consequences are accepted and stated rather than hidden. First, the just-closed month joins the searchable record only when its row is inked (the record on this page is settled through October 2026). Second, the in-progress month is never searchable — a month can settle differently from how it traded, and only the settled version enters the record.
Each live reading carries its own as-of stamp, because their sources update on different clocks — monthly composites, daily monitor states, month-end valuation readings.
Contiguous matching months merge into episodes; a single non-matching month splits them, and a one-month match is its own row — no minimum length, no bridging. The matching months up to today, and any other matching month in the six months before today, are today's own episode — shown and marked, excluded from the medians. An episode that ended inside those six months is cut where they begin: its earlier part is a past episode, the rest is today's.
A month counts as a firing month when it contains at least one day the shipped Capitulation Monitor scores at Washout or worse — the same unit its own episode register displays. One word, one meaning across the suite.
Each row is one episode. The columns:
Returns are S&P total return (dividends reinvested monthly, Shiller monthly convention; the basis section below). "During" runs from the episode's first month-end to its last (a one-month episode shows "—"). The forward windows open when the episode's end was known: the next month's label is decided five weeks into the month after it, so every window opens at the second month-end after the episode's last month. "Worst after" is the deepest monthly total-return level within the 36 months after that month-end, from its level. A monthly series has no intra-month lows, so this is a month-end-grain worst. The +1y/+3y/+5y columns are month-end levels 12/36/60 months after that same month-end. Columns are open where the record ends.
Every monthly forward-return table on this site reads S&P total return: the price path with dividends reinvested monthly. That covers this register, the joint cell, the valuation era tables, the ECY lens, the profit-share register, the tested relationships' records, and the State of the Economy context figures. Price-only returns understate the old eras far more than recent ones — dividend yields ran 3–5% in the 1950s–70s against 1.1% now — so cross-era comparisons on a price basis systematically flatter the modern rows. Measured on the record, on one named window: the three years from August 1982 gained +71.6% on price against +96.3% on total return. And across the whole modern record, from January 1947 to October 2026 the S&P's price CAGR is 8.1% but its total-return CAGR is 11.6%.
The series (monthly, 1871→) is derived by the Shiller method from Shiller's own price and dividend-yield data. In that method, TR grows by the price ratio plus one-twelfth of the trailing dividend rate reinvested at each month's price. Its construction is permanently pinned against his published total-return figures (within 0.1pp on every tested window) and against the official S&P 500 TR index over 1988–2023 (within about 0.1pp per year). Two conventions meet on these pages, both stated: the total-return series carries Shiller's monthly price convention, while price charts and the daily-grain tables use last daily closes. In violent months the two can differ by several points on a single window, which washes out in the medians the tables lead with. At the shortest horizons — the three-month columns — that single-window convention shift can exceed the dividend increment itself; read row-level short-horizon figures as monthly-grain history, not tradable precision.
Two table families deliberately stay price-only: the regime page’s period returns and the capitulation monitor’s episode forwards. Both are daily-grain constructions (calendar/recognition day brackets; 63- and 252-trading-day offsets). No honest daily total-return series exists before 1988, and they say so in their own fineprint rather than mixing grains. One disclosure for the deep history: before March 1957 the daily "S&P 500" tape is the S&P 90 composite spliced by the vendor; Shiller's monthly series is the back-cast composite throughout.
Each register row carries the readings at its episode's ENTRY month, as they were known five weeks into that month (the month's decision date):
Context is compared row by row, never averaged.
The summary is the median row plus one dot per closed episode per horizon on a shared scale with a zero line, and a counts caption. The caption reads "X of N ended lower." Zero is the only threshold anywhere; the dots are the rows themselves, and nothing is annotated onto them. Counts, never probabilities.
Not validated against outcomes: this page describes, it does not call. It assembles at load from the other features' maintained records and makes no claim of its own. The Validation page sets that beside every other feature's record.
There are no Confluence-side jobs and no stored Confluence tables: this page assembles at load from the other features' maintained records, so a new month appears here as soon as its inputs land in theirs. The quarterly review sitting resolves the register's open forward columns. There is nothing to grade — the page makes no claims.