Methodology

Eight public sources, each used for what it actually publishes rather than blended into one series:

The per-bank data covers 38 banks, of which 38 have audited filings extracted and loaded, current through 2026Q1. Figures are extracted from the filed PDFs by deterministic code — anchored to the labels and coordinates the regulator’s own template uses — not by a language model, so the same PDF always yields the same number. Each bank carries the quarters it has filed since 2022.

One bank is carried but excluded from every peer statistic. Takasbank is BDDK-licensed and files the same quarterly reports, but it is the central clearing and settlement institution — no deposits, and most of its balance sheet is member cash and collateral it merely holds. Ranking it against lenders would make loan/deposit, NPL and margin ratios meaningless and would seat it near the top of an asset league on money it does not own. It keeps its own page; it is absent from ranks, market shares, the sector HHI and every audited sector ratio. Peer statistics are therefore drawn from 37 banks.

This is the single largest source of apparent disagreement in Turkish banking data, and the site handles it by naming the basis rather than picking a winner. Sector capital adequacy is one number in the BDDK monthly bulletin and a slightly different one summed from the audited filings; loan-to-deposit exists as a published all-currency monthly ratio, as a TL-only weekly ratio by ownership group, and as a bank’s own audited quarterly figure.

All are correct and they answer different questions. So: a figure prints the basis it was computed on, a comparison is never made across two bases, and where a reader will meet the sibling figure on another page, the page says where and why it differs.

• Real terms use the exact Fisher form, (1+g)/(1+π) − 1, never the g − π shortcut. At a ~30% CPI the shortcut is more than a point adrift. Growth rates are deflated by year-on-year CPI; a return earned across the year is deflated by the twelve-month average, and the page states which.
• Income-statement figures are year-to-date at source and are de-cumulated before being read as a quarter. Return on equity is trailing-twelve-month profit over average equity across the window, not an annualised snapshot.
• Sector aggregates are summed, not averaged, from the banks reporting that quarter — numerator and denominator over the same population, so a bank missing one component cannot drag a ratio down while still contributing to its denominator. Where a ratio has no stored numerator (liquidity coverage, for instance), the sector view is an asset-weighted average and is labelled as one.
• Growth is paired by date, not by row offset. A source that skips a week must render as a gap; measuring 65 weeks and calling it a year is the failure mode this rule exists to prevent.

Three figures here are computed differently from the way a Turkish bank publishes them or a sell-side model carries them. None is an error — each is a definition — but a reader reconciling against their own numbers will find a gap, and ought to know which gap it is rather than conclude the data is wrong.

• Net interest margin is not swap-adjusted, and is struck on total assets. Turkish banks fund lira assets by swapping foreign currency, and under TFRS the cost of that swap lands in the trading and FX line, not in interest expense. Banks therefore publish a swap maliyetine göre düzeltilmiş marj computed from their own treasury books — a figure that is not a line in the filing and cannot be reproduced from it. Ours is the unadjusted margin over average total assets rather than average interest-earning assets, so it reads lower than a bank-published or broker NIM, and the gap is widest for the banks that swap most.
• Cost / income uses BRSA gross operating profit(the statement’s own subtotal), which includes other operating income — provision reversals among it. Most brokers strip that out. Ours therefore reads structurally low against a sell-side cost/income, and the difference is largest in quarters with big reversals.
• Unconsolidated is the default everywhere. Every audited figure is the bank-only (solo) filing unless a page says otherwise; per-bank pages carry a toggle where the consolidated report exists, but every sector aggregate is solo. Brokers generally model consolidated, which for a bank with material subsidiaries is a different company.

The extraction is validated against the filings themselves — statements must foot, parents must equal the sum of their children, and totals must reconcile to the figure the bank printed. A partition that fails is held back rather than published; the coverage matrix behind the site tracks every bank-quarter-statement cell.

On the presentation side, three checks run in continuous integration on every change: a sentence asserting a direction, a level or a ranking must be computed from the series beside it or it does not print; every colour used as text must clear the WCAG AA contrast floor; and the documentation must name every scheduled job and every secret it reads. These are gates, not conventions — a change that breaks one does not ship.

• Not investment advice, and no forecasts. Where a scenario is sized, its assumptions are printed next to it.
• Not real-time. Each page carries the period of the data it shows and the date it was last refreshed. Market prices are delayed.
• Not a substitute for the filings. Where a figure matters, the source document is the authority — this site is a faster way to read it, and its extraction is auditable in the repository.
• Not complete. Some disclosures are filed as images rather than text and resist extraction; some banks disclose an item others do not. A gap prints as a gap.

If a figure here disagrees with a filing, that is a bug and worth reporting: incesalim10@gmail.com. Fixes land in the public repository with the reasoning attached, and the dated change history is part of it.