# How Standard Equity's reports get made What data we use, how the AI analyst works, where the projections come from, and what they can and can't tell you. Snapshot: Sun 12:00 UTC Max age: 7 days Free tier: SE-52 Part 1 ## The data Regulatory filings, normalised, and rebuilt once a week. ### Where the numbers come from Financial data is sourced from regulatory filings (SEC EDGAR, ASX announcements, and equivalent regulators worldwide), plus licensed third-party financial data APIs that aggregate and normalise these sources. Price data comes from market data providers. We don't publicly name our specific upstream providers — they change as quality and pricing dictate, and naming them would imply a warranty on data that ultimately originates from filings. For US operating companies and investment holding companies, years older than our main statement source's come from a second source, line by line, only where its figures have matched the main source closely on the years both cover. They extend the table's history (the 10-year growth rate's starting year) and nothing else: projections, the Reference Value and the SE Grade use the main source's years. Years from before the company listed are left out, including a blank-cheque company's years before its merger and a predecessor's years before a reverse merger. ### A weekly snapshot Every report is a weekly snapshot, true as of 12:00 UTC on the most recent Sunday. Reports are rebuilt once a week, so a report can be up to seven days old by design. In each snapshot: * **Annual & quarterly financials** are whatever has been filed by the snapshot; a filing made mid-week appears in the next week's report. * **Stock prices** are the last close at or before the snapshot, for free and Pro alike. * **Company Notes** are written once per snapshot week. * **Analyst targets and estimates** are taken with the weekly snapshot where consensus data is available. Part 2 ## The metrics A bank and a retailer are not measured alike: each report shows the lines that fit its business. ### Sector-aware metrics One size doesn't fit all. A bank doesn't meaningfully have free cash flow margin; a retailer doesn't meaningfully have net interest income. Reports detect the company's sector and surface the metrics that actually matter for that business model: **Banks.** Net interest income (NII) and NII growth, net interest margin (NIM), efficiency ratio, return on tangible equity (ROTE), tangible book value per share, P/B and P/TBV, equity-to-assets, total assets; valued on excess returns over book. **Lenders.** Card issuers, consumer and student lenders and mortgage lenders (S&P's Financial Services statement layout, for example American Express): the bank layout when interest is at least 15% of revenue; net interest margin is replaced by the loan yield (net interest income over average loans), because only part of a lender's assets are loans. A lender whose revenue is fees or gains on sale, such as an asset manager or a mortgage originator, keeps the operating layout. **Brokers and asset managers.** The bank layout, with interest income as the top line; valued on excess returns over book. **Insurance.** No gross margin, free cash flow, EV multiples or ROIC (premiums and invested float distort them); combined ratio (benefit ratio for life insurers and composites, which hold many years of premiums as invested assets, and for any insurer without an underwriting cost line), premium growth, investment yield, book value per share growth, P/B, P/TBV, tangible book value per share and total assets instead; valued on excess returns over book. **Holding companies.** Investment holding companies such as Exor, found from the statements: most of the assets are investments, funded mostly by equity. The top line is the net investment result (income and fair-value gains on the stakes), which can be negative; margins and growth are shown only on a positive result. No gross margin, free cash flow, EV multiples or ROIC; book value per share (Book/sh), book value per share growth, P/B and total assets instead. The EV/EBIT and FCF per share rows are dashes: the P/TBV and TBV/sh that banks show there would repeat P/B and Book/sh. We make no projections of our own, only analyst consensus. Why book value and not net asset value: for a fair-value investment entity book value is its net asset value; for a holder that carries its stakes by the equity method it is not (it is their carrying value, not their market value), and the data cannot tell the two apart. **REITs.** The general metrics, with FFO per share, FFO growth, P/FFO and the FFO payout in place of EPS, EPS growth, P/E and the payout (the header, the Peers column and SE Grade use P/FFO too); valued on dividends. Why: GAAP earnings deduct depreciation on buildings that mostly hold their value, so a REIT's EPS and P/E say little about its rents. We use the company's reported FFO first; only when none is reported do we compute net income + real-estate depreciation − gains on property sales. **General companies.** Revenue and EPS growth, gross/operating/net/FCF margins, ROIC, ROE, ROA, EV/EBIT, FCF yield, debt-to-equity; valued on discounted free cash flow. Full definitions for every metric are in the [glossary](https://standard-equity.report/glossary). ### The tables under the Company Note Three small tables sit under the Company Note, each three rows with a Trend column of up to eleven annual bars (ten in the Reinvestment table), drawn like the main table's (each row on its own scale). Each heading states its basis (_Top 3 87%_, _7Y CAGR_, _10Y median_, _1Y_), a dash means the figure is not reported and _n/m_ means it exists but would not be meaningful. Numbers are in the company's reporting currency. The tables follow the company's report type, so a bank, an insurer and a landlord each get their own lines. Definitions are in the [glossary](https://standard-equity.report/glossary). **Driver.** What drives the business, always as an amount or share, its _Trend_, a third figure and one comparison with the company (1.00x is in line). Where the company reports earnings by segment and they match at least 80% of segment revenue, it is _Profit by segment_: share of profit, its Trend, share of revenue (over the same profit-making segments) and _Rel margin_, the segment's margin against the margin of the profit-making segments together. In the Trend each year is measured against that year's profit-making segments, so a segment that lost money in a year draws a negative bar. Where only segment revenue is reported, _Revenue by segment_: revenue share, its Trend, growth and _Rel growth_, its growth against the company's total revenue growth. With neither, _Growth vs revenue_: general companies and REITs show revenue, cost of revenue and operating expenses; banks and brokers total assets, revenue and non-interest expense (derived from revenue when not reported); each with its latest amount, Trend, growth and _Rel growth_. Insurers show revenue, claims and benefits, and total costs (revenue less pre-tax profit). Subtotals are rolled into one row so a group is not counted twice, loss-making segments are not shown as rows, and corporate and elimination lines are never shown. A segment table for an earlier year than the main table's newest (never more than one fiscal year earlier) names it in the title, for example _Revenue by segment FY25_. **Revenue.** Where the money comes from. _Revenue by region_ shows the top three regions with their combined coverage in the heading; when the regions are reported only up to an earlier year than the main table's newest (never more than one fiscal year earlier), the title names that year, _Revenue by region FY25_. Without regions: _Revenue mix_ for banks and brokers (net interest income, non-interest income), insurers (premiums, investment income, other) and REITs (rent, management fees, other); for everyone else, _Asset mix_, the biggest asset lines as a share of total assets. **Reinvestment.** What the company does with its earnings. _Reinvestment returns_ shows the reinvestment rate (the median share of operating profit put back into the business over up to ten years; banks and insurers show the retention rate instead, REITs measure against FFO), the return on capital (the median over up to ten years of each year's profit over the capital in use) and its spread over the hurdle the Reference Value used: the WACC for a free cash flow model, the cost of equity otherwise. With fewer than three usable years a cell is n/m; banks, insurers and companies with a captive finance arm use net income, book equity and the cost of equity. A company with a loss or a cash burn gets _Financing_ (_1Y_): burn, equity raised and debt raised. ### Data quality indicators When a metric is missing, partial, or stale for a given period, we mark it visually rather than silently leaving a blank or showing zero. Symbols you'll see across reports: ### The marks | ○ | Partial data (some quarters within the period are missing) | | --------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | — | Not applicable for this company type (e.g., FCF margin for a bank) | | Italics | Estimated or projected, not yet reported | | ○ with a figure | The figure is our estimate, or is on another definition than the row, because the company's own is not available (for example a bank's net interest margin before the years it has reported); hover for the reason | Part 3 ## The AI analyst Claude writes the narrative from the structured data, and sees nothing else. ### What it receives The narrative analysis on each report is generated by Anthropic's Claude. The AI receives: * The company's last 10+ years of financial statements (income, balance sheet, cash flow) * Recent management commentary where available (10-K MD&A, earnings call quotes) * Peer comparison data: how the company compares with the companies nearest it in size * The methodology and prompt template that guides what to look for and how to frame uncertainty ### What it does not see The AI does _not_ have access to: * Your portfolio, watchlist, or any user-specific data * Real-time market sentiment, news flow, or social media chatter * Insider information of any kind * Anything outside the structured data we explicitly feed it ### How it writes Reports take positions where the data supports one and explicitly hedge where it doesn't. Mealy-mouthed AI summaries that avoid every conclusion are a failure mode we actively prompt against. Part 4 ## The projections Revenue first, then earnings, then analyst consensus, then the range. ### How a projection is built Forward-looking projections (revenue, earnings, cash flow) are built in this order: 1. **Revenue and margins from the company's own record.** Revenue follows a trend fitted to the company's multi-year growth in which only growth above the risk-free rate of its reporting currency fades, so a steady grower keeps growing instead of sliding toward zero; margins hold close to their recent level. The fair value model's first years use the same revenue. 2. **Earnings built from them.** For every company but a holding company (which shows analyst figures only), net income is revenue times the projected net margin, EPS is that net income over the share count, and free cash flow is revenue times the projected cash-flow margin (banks, insurers and broker-dealers show no free cash flow), so the rows always agree with each other. The share count is carried flat at the last diluted figure (marked ○, with no confidence dots) because a trend through share counts forecast worse than leaving them alone; dividends keep their own trend, since deriving them from earnings forecast worse. 3. **Analyst consensus where it is broad.** Shown as is where four or more analysts cover a year. With fewer, their figure is mixed with ours in proportion to their number: one analyst counts for nothing, two for a third, three for two thirds. The mix is a stated rule, not a measured one: we do not yet hold enough history of past estimates to score it. The Reference Value weighs thin analyst years the same way. 4. **Confidence intervals.** Wide for volatile companies, narrow for stable trends. ### Reading the projection zone The projection zone on each chart shows the model's best estimate plus an upper/lower band. If those bands are wide, that's the model telling you it doesn't have high conviction. ### Reported and adjusted earnings per share The table's history and "PE TTM" use reported (GAAP) EPS. Analysts forecast _adjusted_ EPS, which leaves out items such as acquisition amortisation, one-off charges and often stock-based pay, so the forecast columns can sit on a different basis from the years beside them. Where the two differ by more than 10%, the forecast cell carries an "adj" tag (its tooltip gives the reported estimate) and the header's forward P/E reads "PE (Fwd Adj)". The adjusted figures are S&P Global's consensus; other data providers' consensus can differ, especially for companies that report no adjusted EPS. Part 5 ## The Reference Value A range from thousands of runs of a cash flow model, not a single point. The Reference Value is built from a discounted cash flow model, run thousands of times with its main inputs varied within the analysts' range or the company's own history. Three numbers come out of it: ### The three numbers | Expected value | The probability-weighted average of the simulated values, with the discount rate at its estimate (the range also lets the rate vary, but the headline does not take credit for that). This is the headline. | | -------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Median | The middle simulated value. A few large outcomes pull the expected value above it. | | Base case | One run with every input at its estimate (Damodaran's method, with no randomness). | ### Two groups of outcomes When the simulated values form two separate groups, the expected value can sit between them, where few outcomes land. We show a warning in that case. ### The Reference Range The **Reference Range** is the central 50%: the 25th to 75th percentile of the simulated values (the wider 10th to 90th range is in the workings). It is model-implied, and it is not a confidence interval: it shows how far our inputs can move the value, not how often the real outcome falls inside it. We have not yet tested that coverage, and the workings say so (calibration: none). We will test it once enough past estimates and outcomes have built up. Until then, treat the range and its width (narrow, moderate or wide) as a statement about the model's inputs. ### What the price assumes The **What the price assumes** block runs the model backwards from today's price: the growth or margin the price needs, and the revenue it implies in ten years. It is compared with round judgement base rates for companies of similar size and growth. Those rates come from survivors and are not measured, so they flatter what companies achieve. The block is a market-expectations check and never changes the Reference Value. It shows for operating companies (growth, margin, revenue multiple) and, as dividend growth only, for REITs. Banks, capital markets firms and insurers use an excess-return model with no such solve, so they show nothing. ### When the cash flow model does not fit Companies where the cash flow model is not meaningful (heavy net debt, negative value) get an Indicated Value instead: the average of simulated pricing values (each the median of how the market prices their peers, moved to the company's own growth for the price-to-earnings and EV/EBITDA multiples, and the company's own history, with the multiples and next year's earnings drawn), labelled as such and shown with the same central 50% range as the cash flow model. Investment holding companies get no cash flow model at all (their value is in their stakes): their Indicated Value is book value per share at the company's own 10-year median price-to-book, and there is none without that history. Part 6 ## The Grade and the prices The SE Grade, the Consensus Price and the Year-Ahead Price, each with its rule stated. ### The SE Grade The stars say how good and durable the business is, from 0 to 3, and are meant to be rare, as Michelin's are. They are not a forecast: on a sample of 606 stocks early Moat stars ranked later returns on capital well but not later share-price returns, which is why price may move the result, but only a little. **The business.** Quality (is it good and safe today) and Moat (will it stay good) each earn 0 to 3. A pillar at 0 leaves the Grade blank: a dash, never a bad mark. Together they set the business tier. **The price, one notch at most.** The Reference Value and the stock's multiple against its own 10-year history place the price between cheap and rich, in a band that widens with the Reference Value's uncertainty. A cheap price lifts the tier by one notch and a rich price lowers it by one; a lift never makes 3★, and a business with a 1★ Moat cannot climb above 1★. A business with the top mark in both Quality and Moat is never below 2★, whatever the price. The notch moves smoothly with the price, and the workings show it. **The half star.** (½☆☆) — in the guide, below one star: a 1★ business lowered one notch by a rich price. **The rank cap.** 3★ also needs the 5-year return above its cost of capital in the top 5% of the \~31,000 companies ranked, 2★ the top 10%. The rank can lower the stars, never raise them. A stock the screener has not ranked is placed by its own 5-year spread; one still unranked stops at 2★. **By report type.** Operating companies and REITs are judged on ROIC, cash conversion, debt and interest cover, with the stability of gross and operating margin for the Moat (REITs price on P/FFO). Banks and capital-markets firms are judged on ROE over the cost of equity, ROA and capital, with book value per share growth for durability and P/B for the price. Insurers are judged on ROE alone, so none reaches 3★. Investment holding companies are not rated yet: the overall, Quality and Moat show a dash until the Grade has been tested on them, and the Value row still shows. Each Quality signal is a straight scale rather than a set of steps. The weights were checked against what about 27,000 companies went on to earn over 2019 to 2025: the star measures how good and durable the business is, not what the share price will do. That test has limits (it holds companies that survived to today, and one market cycle). The workings open from the SE Grade in the sidebar. ### The Consensus Price and the Year-Ahead Price The **Consensus** figure on the price chart (the Consensus Price) is our analyst consensus: the average of each analyst's own price target, weighted by their track record (better-ranked analysts count more) and by how recent the rating is (a rating loses half its weight every three months and is dropped after a year). It is used alone when five or more analysts cover the stock. With fewer, the consensus is thin, so it is mixed with our Modelled figure in proportion to the number of analysts (one analyst carries a fifth, four carry four fifths, none carries nothing) and the label changes to **1Y Price** (the Year-Ahead Price). Hover the label for the mix. The dots beside the figure grade the range on the same scale as the table's estimates: the range's half-width as a share of the figure, within 15% three dots, within 30% two, within 60% one. Analysts alone give a narrow range, so a well-covered stock will often show three dots; they say how tight the opinions are, not how likely they are to be right. ### The Consensus Rating The rating beside it (Strong Buy, Buy, Hold, Sell, Strong Sell, then the five counts) is our consensus too. It starts from the whole pool of analysts S&P Global counts, scored the way S&P scores them: a Strong Buy counts fully, a Buy half, and the same on the sell side, so the score runs from −10 to +10 and the labels change at ±2.5 and ±7.5. Where we also have the individual ratings, each firm's latest rating is weighted like the targets (track record and recency) and the score is moved toward what those firms say, in proportion to the share of the analysts they cover. Without them the score is the pool's own. Where S&P Global has no ratings at all, Yahoo Finance's analyst counts are scored by the same rule, without any weighting; Yahoo's numbers are never mixed with S&P Global's. The five counts are always every analyst counted, unweighted, so the label can differ from the counts when the better-ranked, more recent analysts lean the other way. A stock only reads Strong Buy or Strong Sell when the analysts themselves graded it that way. ### How the Year-Ahead Price is built Our figure is deliberately plain. The Modelled price is today's price grown at the stock's cost of equity (the rate the Valuation Model discounts at), less the dividends it pays. It is what a fairly priced stock returns, with no opinion on whether the price is cheap or dear: the Reference Value is for that. With fewer than five analysts the range around it is an 80% range built from a weighted mix of the analysts (each by their own target, weighted by track record and how recent the rating is) and the model, and it is never narrower than the stock's own price volatility over the last three years, which widens with the square root of time: analysts' targets cluster more tightly than prices move. Where analysts and the model disagree, the range widens. In a backtest on about 4,900 stock-years the volatility range alone held the next year's price 79% of the time; we do not yet have the history to test the weighting. It is a model figure, not a prediction or a recommendation. Where a stock has no analyst ratings, the ratings line instead says where the price sits against our Reference Value (below, near or above, within 10%). Part 7 ## The free list 52 stocks, chosen by rule rather than by hand. The free tier covers 52 stocks, chosen by rule rather than by hand: * The largest companies by market value in each region, with fixed slots (US 20, Europe including the UK 15, Australia 6, Asia ex-Japan 5, Japan 2, Latin America 2, rest of world 2); in a region of several countries no country takes more than a third of the slots (at least two) * A company needs at least 15% of its shares trading freely, as in the major indexes, at least $50 million traded on an average day, and ten years of reported statements with at most two loss years and none in the last three * A company with two share classes takes one slot; Berkshire Hathaway appears as both classes, each with its own report * A member near the cut keeps its place within a small buffer of ranks, so the list doesn't churn on small moves The list is rebuilt every four weeks — see the [SE-52 page](https://standard-equity.report/se-52) and the [changelog](https://standard-equity.report/changelog) for additions and removals. Part 8 ## Limits and corrections What we don't do, and what happens when a number is wrong. ### What we explicitly don't do | Recommendations | No buy/sell/hold ratings. The reports describe; you decide. | | ------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Real-time alerts | No price triggers, no earnings-surprise notifications. | | Portfolio tracking | We don't store your holdings. | | Manual hand-tuning | No analyst overrides on individual reports. If the data shows a number, the report shows it. If it's wrong upstream, we either fix the source or flag the gap. | ### Errors and corrections Found a number that looks wrong? Email [hello@standard-equity.report](mailto:hello@standard-equity.report). We investigate, document fixes in the [changelog](https://standard-equity.report/changelog), and where the issue is upstream we file with the data provider. Transparency about errors is part of the job. ## At a glance | Snapshot | Sun 12:00 UTC | | ------------- | ---------------- | | Max age | 7 days | | Company Notes | Once per week | | Free tier | SE-52 | | SE-52 rebuilt | Every four weeks | ## In this report 1. [The data](#data) 2. [The metrics](#metrics) 3. [The AI analyst](#analyst) 4. [The projections](#projections) 5. [The Reference Value](#fair-value) 6. [The Grade and the prices](#scores) 7. [The free list](#free-list) 8. [Limits and corrections](#limits)