Stock market valuation indicators

Five gauges, one table, refreshed on a stated cadence. Each row carries a current reading, the percentile of that reading inside the gauge's own prior decade, and a sentence naming what the reading licenses. The percentile matters more than the level, because it removes the definitional differences that make levels incomparable.

The composite is the mean of the five, 68.0, with a median of 80 and a range from 20 to 90. Which is to say the gauges disagree by 70 percentile points, and the disagreement is the most informative thing in the table.

Resolve it by asking what each gauge is measuring. Four of the five are absolute, comparing a price to an earnings stream or to output with no reference to what else you could buy. The fifth is relative, and compares equities to bonds. That relative gauge reads cheap because the real bond yield is 2.30 percent, which is low against its own history, so equities look tolerable next to a poor alternative. The absolute gauges read dear because the price is high against the earnings, full stop.

What should you do when the gauges disagree?

Both readings are true and they answer different questions. The absolute gauges answer "what real return is this price likely to deliver", and the answer is a poor one. The relative gauge answers "given that I must own something, is equity the better of two options", and the answer is yes. An investor who must be invested acts on the relative gauge. An investor deciding how much to consume, how long to work, or what withdrawal rate to survive on acts on the absolute gauges.

The trailing multiple's 60th percentile against the cyclically adjusted ratio's 90th is the other disagreement in the table, and it has a specific meaning: current earnings sit far above their own ten-year average. Either the recent earnings level is durable, in which case the trailing reading is the honest one and the cyclically adjusted ratio is being dragged down by a stale denominator, or it is a cycle peak, in which case the trailing reading is a trap of exactly the kind the CAPE ratio page describes. That question has an answer you can look for, in margins against their own history and in the profit share of national income.

How should you use the dashboard?

The cadence that works is quarterly for the readings and annual for the histories, with every input carrying the date it was pulled. Anything faster invites you to act on noise; anything slower and the implied premium goes stale as rates move.

Two things the dashboard never licenses. It does not license a date, because none of the gauges has any information about timing and the residual arithmetic on the CAPE ratio page says so explicitly. And it does not license leaving the market, because the floor in your allocation band exists precisely to stop a gauge reading from becoming an exit.

{
  "skill": "Building and refreshing the five-gauge market-temperature dashboard from primary sources, with every input dated and every percentile computed against the gauge's own history",
  "use": "The retrieval here is large and mechanical and the judgement is small, which is the right shape for a tool. Have an assistant write the code that pulls ten years of index earnings, a consumer price index, index-level dividend and buyback yields, a nominal government yield and an inflation expectation into one dated table, then computes the cyclically adjusted ratio, the capitalisation-to-output ratio, the real-yield spread and the two-stage implied premium, and finally the percentile of each reading inside its own prior decade. Require a second, independent implementation of the implied-premium solve (a bisection and a Newton search) so the two can be diffed before either is trusted. Ask it separately for the strongest case that each gauge is currently misleading, because a gauge you have not argued against is a gauge you will over-read.",
  "tools": ["Chat assistants with code execution for the solver and the percentile table", "Python: pandas for the dated panel, scipy.optimize or a hand-rolled bisection for the implied-premium solve", "Document AI for lifting earnings-per-unit series out of index factsheet PDFs into a referenced table"],
  "prompt": "Extraction and computation only; do not offer a market view. From the attached index factsheets and price-index releases, build one table with columns: period, index earnings per unit, earnings definition used, consumer price index, source document and page for each cell. Then compute (a) each year's earnings restated into the latest period's money, (b) the ten-year mean, (c) the cyclically adjusted ratio at the stated index level, (d) the same ratio with the deflator omitted, and (e) the percentile of (c) within a supplied list of ten prior annual readings. Leave any cell you cannot source blank rather than inferring it, and list separately every figure you could not tie to a document.",
  "verify": "Three checks, in order. First, recompute two of the ten deflated earnings figures and the ten-year mean by hand; a single mis-keyed CPI base silently shifts every restated figure and the error is invisible in the output. Second, feed the solved discount rate back through the two-stage model and confirm it reproduces the index level to within 1e-6, and confirm the terminal value's share of the total is reported alongside the premium, because a premium with a 92 percent terminal share is a statement about a convention rather than about a market. Third, check that the earnings definition is identical across all ten years and is named in the output; as-reported and operating earnings differ by enough to move the ratio several points, and a series that switches definitions midway is not a series. Every perishable input carries the date it was pulled, and any figure the model produced without a source document is deleted rather than footnoted.",
  "diy": "The gate is unaided. You restate ten years of earnings into current money with a calculator, average them, divide, and state the reading as a percentile of its own history; you compute the exact real yield from a nominal yield and an inflation expectation and set it beside the earnings yield; you solve a two-stage cash-flow model for the discount rate by trial and interpolation and report the terminal share; and you say in one sentence what each reading does and does not license. A tool that builds the table cannot tell you that the relative gauge and the absolute gauges disagree because the real bond yield is low, which is section 4.8's whole job and yours.",
  "market": "IN"
}