How every figure here is built

Isolated quarters, never year-to-date

Q2 2026 revenue = the row spanning 01 Apr → 30 Jun
(not the row 01 Jan → 30 Jun)

In quarterly filings the income statement appears twice for the same period: year-to-date and the quarter on its own. This site always uses the isolated quarter, and shows each row's actual window.

Q4 is computed, and labelled as such

Q4 = full year (annual filing) − first nine months (Q3 filing)

The CVM does not publish a standalone fourth quarter: the annual filing carries the closed year. Every December row here is tagged “computed”, and any ratio using one says so on the same screen.

Price to earnings, without estimating a share count

P/E = last close ÷ trailing twelve-month earnings per share

Earnings per share comes from the filing itself, for the ticker's own share class — PETR4 is preferred stock, so the divisor is the preferred EPS. The multiple therefore depends on no externally estimated share count.

When this site refuses to show a multiple

A company with twelve-month losses (a negative P/E says nothing); a reported EPS that does not reconcile with published net income — PINE3 filed 190,833, which would imply 2,742 shares in the entire company; and negative shareholders' equity, which takes down P/B, ROE and leverage at once, because dividing by negative equity flips the sign — a loss would read as record return.

Market cap, class by class

market cap = common shares × common price + preferred shares × preferred price

The share count comes from the capital stated in the company's FRE filing, broken down by class — never derived from earnings divided by EPS, which fails whenever classes carry different EPS. Since PETR3 and PETR4 close at different prices every day, the sum runs class by class, and both pages show the same market cap — the company's. P/B divides shareholders' equity by total shares: common and preferred therefore share one book value per share, and what differs between them is the price.

EBITDA, debt, and the multiples built from them

EBITDA = operating result + depreciation and amortization (from the cash-flow statement)

D&A has no account of its own: it is an adjustment line in the cash-flow statement, freely named — and the DFC (cash-flow statement) only publishes year-to-date figures, never the quarter. The quarter is the difference between neighbouring accumulations; without the neighbour, that window's EBITDA does not exist and the page says so. Gross debt adds the two loan accounts on the balance sheet (current and long-term), net debt subtracts cash and equivalents, and EV adds net debt to market cap. Gross margin uses gross profit (account 3.03), the only income-statement code that is identical across all charts of accounts. The 5-year CAGR requires both endpoints published and positive — a rate that crosses a loss-making year is not a rate of anything.

Distributions: B3's registry, by share class and record date

12m yield = distributions per share with record date in the last 365 days ÷ current price

Dividends, interest on equity and REIT income come from B3's own public distributions registry — value per share with share class (ON, PN), approval date and record date. Yield is by record date, which belongs to whoever held the share that day, and by CLASS: PETR4's yield sums only PN distributions. Payout multiplies each class's distribution by its share count (from the FRE) and divides by twelve-month earnings — and only shows when the company runs plain ON and PN, because splitting with PNA/PNB or units in the mix would be guessing. For REITs the same registry carries R$ per share with reference month and payment date — which the monthly report does not publish.

REITs: fundamentals come from the monthly report, bridged by ISIN

P/B = B3 price ÷ book value per share as filed

Net assets, book value per share, unitholders, yield and monthly return come from the monthly report each fund files with CVM — published as filed, with the reference month printed on the page. The bridge between the report's CNPJ and B3's ticker is the ISIN, published on both sides; short names stay out, having produced wrong matches before. P/B only shows when the report is under six months old: an old book value against today's price is a multiple that lies.

Consolidated before parent-only

A company with subsidiaries files both; the consolidated statement is the one describing the whole business. A company without subsidiaries files only parent-level accounts — and there, that is the correct figure, not a lesser substitute. The page states which one is on screen.

The filing estimate is measured habit

estimate = next period end + this company's historical median

How many days this company usually takes between quarter end and lodging the filing, measured across every filing it has made. The minimum and maximum sit alongside, because a median alone hides the spread.

Twenty-six years of raw prices — and what that leaves unsaid

N-year return = current month close ÷ same-month close N years earlier − 1

Since 2026-09-02 the archive holds B3's COTAHIST from 2000 to date, cash market only (before the real there are three currencies in the way — a series crossing a redenomination is not a number). The high/low/change window became a true 52 weeks; 1/3/5/10-year returns come from monthly closes, and the long chart shows monthly closes since 2000. All RAW PRICE: no adjustment for distributions or corporate events — B3 does not publish the factor, and inferring it would be our number.

The window's label says what it measures

Since 2 September 2026 the closing series runs from 2000 to date, and the high/low/change window is a true 52 weeks. Before that the series began on 2 January 2026 — and this site called that “the window”, never “12 months” or “52 weeks”, because it was not. A label that promises more than the data holds is the easiest lie to publish; when the data grew, the label grew with it, not before.

Real returns: both ends on the last CPI month

real = (close in Jul-26 ÷ close in Jul-16) ÷ Π(1 + monthly CPI) − 1

The 1, 3, 5 and 10-year returns come in two versions: nominal, to the last close, and real, net of Brazil's CPI (IPCA, central bank series 433). The real figure sits on the same card, with the date it runs to.

Yearly P/E and DY

P/E for a year = close of the fiscal year-end month ÷ that year's EPS
DY for a year = payouts with ex-date in the year ÷ the same close

The per-year table carries each year's multiple since 2010, computed with that year's price — not today's. EPS and payouts are already per share, so neither needs a share count. The EPS scale check (net income ÷ EPS must yield a plausible share count) runs year by year, and a year that fails is refused with the reason.

Corn: one session's curve, from B3's settlement price

curve = settlement price of each corn (CCM) expiry in the same session
US$/bag = settlement ÷ central bank PTAX (sell) of the same day

The curve uses the settlement price — the reference B3 computes at the close — not the last trade. Each expiry carries the sessions left to settlement, the day's trades and the open interest. Only the latest session is published: a per-contract history, ready to download, would be redistributing B3's data. The dollar is the central bank's PTAX sell rate (series 1) for the same day; the contract is in reais, and the conversion is for reading only.

Carry: what the market pays to carry, against what interest costs

spread = settlement of the longer expiry − settlement of the shorter
carry p.a. = (longer ÷ shorter)^(252 ÷ sessions in the leg) − 1
interest = shorter settlement × (DI factor over the leg − 1)
left over = spread − interest

A leg runs from one expiry to the next. Interest comes from the same session's DI1 curve, interpolated flat-forward — the DI market convention — and the annualised carry uses the same 252-business-day base, so both fit in one column. What is left over has to pay for storage, insurance and shrink, which have no market source: the reader types that cost in.