Methodology
How HornScout reads the market
How we turn prices, filings and market data into readings you can check: what we measure, against which history, how to read each scale and where the limits are. We don't predict prices or give buy or sell advice.
One reading, start to finish
US stress on Sep 15, 2026, figure by figure.
An "elevated" state and percentile 21 in the same layer look contradictory. They aren't, and following the reading step by step shows why.
Observed figure
Every day, for the US companies in the universe, we compute three measures: how much their prices move over 20 sessions (realized volatility, annualized), how far they have fallen from their high and what share sit more than 15% below it. On Sep 15, 2026, median volatility was 8.9%.
Against its history
Each measure is compared with its own daily history: volatility, over 5,439 days. Its position in that history is its percentile.
State rule
Each of the three measures is classified with fixed bands on its own percentile: elevated from the 75th percentile, stressed from the 90th and shock from the 97th. The layer's state is that of the most strained measure.
Result
Volatility came in at percentile 21: lower than on most days in that history. Even so, the state is "elevated", because one of the drawdown measures pushes it there.
How to read it
Prices are moving little by their usual standard, but many companies remain far below their highs. Both are true at once, which is why they are shown together.
Limit
It describes the US market that day; it does not anticipate the next. The oldest history carries survivorship bias, and the reading changes every morning.
Four steps, in order
How a company is read here.
The same four steps for every company, from what is on record to what changed since you started following it.
The company
What it does, what its record holds and how its price got here: the public profile of every company.
See the profiles →Its sector and market
Seven layers of its region's market — liquidity, credit, regime and the rest — each against its history.
Explore the context →The precedents
What followed similar moments in the past: the distribution of outcomes and how many observations support it; never a forecast.
See the Lab →Since you started following it
Add it to the Watchlist and review what changed since you added it: drift, notices and agenda. Written theses are coming soon.
See the watchlist →
What Context is (and what it's for)
Context is the reading of each region's market, layer by layer: regime, breadth, sector cycles, stress, credit, rates and liquidity. It places a company before you look at it closely. It is not a ranking or a buy signal.
Where the data comes from
We use closing prices and volume for companies in the US and Europe, together with the SEC filings we cover. European registers are not yet included. We store and process them ourselves: what you see are readings computed on stored data, not a live query to the market.
The layers and what they measure
For each layer we tell you what it measures, with which data, against which history and how to read its scale. We don't publish every formula; the example at the top walks through one complete reading, figure by figure.
- Regime and breadth — whether the market as a whole is rising, falling or moving sideways, and what share of companies trade above their 200-session average.
- Sector cycles — which phase each sector is in (leading, extending, losing leadership, lagging, deteriorating, improving) and how many days it has been there.
- Stress — three measures taken on the companies themselves: realized volatility, drawdown from highs and the share of companies more than 15% below; the state takes the most strained, and the percentile shown is the volatility one. Precedent matching uses a composite series built from the same inputs.
- Credit — whether borrowing costs more or less than in its history, measured by the spread over government debt (each region's exact series sits next to the reading).
- Rates and liquidity — rates: the 10-year yield level and its recent move (US only); liquidity: only the direction of flow against a year ago, not how much there is.
Data health: freshness and coverage
Each reading carries the health of the data behind it: current, partial, stale or insufficient. It changes how far you can rely on the reading, not its direction; if data is out of date, we say so instead of implying a cleaner reading. It is different from the Research Lab's evidence grade and the Watchlist's A–D health, which answer other questions.
Percentiles and history
A percentile places today's value against its own history: p80 means 80% of the days in that history had a lower value. It describes how high or low it is, not what will happen, and its meaning depends on the layer: in stress or credit, rising means strain; in breadth, rising means support.
The universe we look at
We work over a universe of stocks curated by us, not over an index like the S&P 500 or the IBEX. That means what you see reflects our selection and cleaning, not 'the whole market'.
Honest limits
- Survivorship bias — history tends to over-represent what still trades.
- Limited coverage — if a layer or a stock has little data, we flag it.
- Zero flags is not zero risk — no alerts means our data found nothing, not that there is no risk.
- History being extended — we are adding companies that stopped trading back into the history. Meanwhile, Research Lab cohorts and precedents with long histories over-represent the survivors and their figures may move; daily prices and each morning's Context are not affected by this.
How the Lab builds its reads
In the Research Lab you define a condition, gather the past moments when it held and compare what followed against two controls. Twelve things worth knowing before you read a number:
The twelve rules for reading a Research Lab number
- Population versus episode — a distribution of company-day observations (thousands, overlapping) and one of independent windows (a few) describe different samples; read their spread in terms of that unit.
- Regime cutoffs are calibrated on the full sample, including a 2005-2007 warm-up window, before they're used to label any single day.
- "Lift" names four different comparisons; always read it next to its baseline. Against the market: the cohort's hit rate divided by the market's under the same conditions. Against the cohort: a band's or slice's rate divided by the whole cohort's. In signals: the median of a variable among winners divided by the universe's, which is not a hit rate. By entry month: that month's rate divided by its control, which the view itself names.
- Preview and Cohort look at different universes: Preview requires today's data to be fresh and excludes extreme drawdowns; the historical Cohort doesn't.
- Returns are price only — dividends aren't included.
- Eligibility is frozen at the last weekly build: a symbol blocked today drops out of its ENTIRE history, not just today's read.
- Deep history is survivorship-biased by construction — it's built from companies still trading today. Delisted companies are being restored to the record; that work isn't complete yet.
- Earnings anchors: US coverage goes back to 2002 and keeps growing; Europe starts in 2023.
- The cohort clock shows WHEN the windows happened — month by month, over the market regime of the moment — because the same number of windows can be decades of history or a handful of months from a single crisis. Its last months are hatched: their forward window is still open, so they have no label yet.
- Two controls and a stability rule — every cohort is compared with the market under the same conditions and with its own sector without the regime filter. A slice by era or regime counts if it holds at least 200 observations; if the sign of lift flips in any of them, the reading is flagged unstable and its evidence grade drops.
- Witnesses — behind a number you can open a sample of the real observations: company, entry date and what followed. They pass the same eligibility as the cohort and a data-quality filter; the sample stays the same during the day and, for rare outcomes, may bring fewer cases than requested.
- This is not a forecast: every reading describes what already happened.
What we don't do
HornScout is not financial advice and not a recommendation to buy or sell any security. It is a data and context tool for your own analysis. Investment decisions are yours and your responsibility.
FAQ
- Does HornScout tell me what to buy or sell? — No. We don't give recommendations or investment advice. We show you data and context for your own analysis; the decision is yours.
- Do you predict price or returns? — No. The reads, including percentiles, describe the current state and its history; they don't anticipate what will happen.
- What does 'zero flags' on a stock mean? — That our data found no alerts, not that the stock is free of risk. Coverage is limited to the data we hold.
- What does data health mean? — It indicates whether the information is fresh and complete. It affects the confidence of the read, not the market direction.
- Which markets do you have data for? — Mainly the US and Europe. The universe is our own selection, not a full index.
- Why do some figures change these days? — Because we are adding companies that stopped trading back into the history. It affects Research Lab cohorts and precedents with long histories; daily prices and each morning's Context do not change because of it.
Questions
Questions about how we read something?
Write to us if a reading isn't clear, or see the method applied to a real company, step by step.