Each build is one file that opens in any browser, no account, nothing installed, nothing phoning home, and it screens roughly 1,900 liquid US names, 1.7 million pairs: Engle-Granger in both directions at finite-sample critical values, walk-forward hedge ratios, a 108-cell parameter sweep, per-leg dollar volume with capital conversion, market beta per spread, deal and reporting-window flags from SEC filings, and a build stamp on every number. What it sells is computation, freshness and correctness. Not forward information. That distinction was measured, on this product, and published with the free launch build: with acceptance recomputed honestly inside each test year, accepted pairs do not beat correlation-matched rejected ones, and no sortable column ranks the year ahead. Every screener in this niche shares that property. One of them says so on its own front page.
Read the 21 August 2026 build: open the desk, complete and self-contained: 12,000 register rows, 700 candidates with charts, every column explained on hover, the methodology inside.
This is not a teaser cut. It is the whole desk, aged. The list it carries is still mostly true; the standings are mostly gone. The figure below says exactly how much of each, which is the entire argument for the subscription in two curves.
Two goods age at two speeds. The list (which pairs pass) moves slowly, because a few sessions against a 2,500-session window is nothing. The standings (which spreads sit beyond two sigma today) move with daily prices, and a standing that faded mostly faded because the spread converged: the opportunity played out on someone else's screen.
| desk age | list valid | standings standing |
|---|---|---|
| 1 week | 95% | 76% |
| 1 month | 87% | 52% |
| 2 months | 80% | 35% |
| 3 months | 73% | 28% |
Standings pooled across 12 month-end reference dates on 3,000 accepted pairs, about 3,200 observations per cell; the list measured by full re-test at the build end and corroborated by build-to-build turnover of about 3.3 percent. A fresh two-sigma standing has even odds of still standing a month later. The measurement re-runs with the builds, and this table moves with it. Method and caveats in the public repository.
| tier | delivery | price |
|---|---|---|
| base | monthly desk and brief: list-grade, standings honestly stale | $29 / month |
| standard | bi-weekly desk and brief: 76% of standings readings current on average | $39 / month |
| pro | weekly desk and brief: standings-grade, 85% of readings current on average | $49 / month |
The cadences are read off the curve, not chosen: by one month standings validity has fallen through two-thirds, so the monthly desk is honestly a list product, and weekly is the only cadence that keeps standings validity at 85 percent on average.
The first ten subscribers take the weekly desk at the bi-weekly price, $39 a month, locked for as long as they stay.
The tiers above are one seat for personal use. A firm forwards the file, feeds the register into its own systems and needs a vendor on paper, so a firm is priced as a firm.
| tier | delivery | price |
|---|---|---|
| firm license | weekly desk and brief for up to five named seats, the register as a file for your own systems, commercial use, vendor paperwork, data questions answered within one business day | $199 / month |
| desk build | your universe, your data, your delivery shape: scoped in writing and quoted before anything is built | from $6,000 |
The first three firms take the firm license at $149 a month, locked for as long as they stay.
ayan@financebroski.com. The button copies the address. Billing is a monthly PayPal invoice; the desk and brief arrive by email on schedule. No platform, no login, no lock-in beyond the month you paid for.
The full desk as a single self-contained file, nothing to install and nothing phoning home, plus a one-page brief: what entered, what left, what crossed two sigma since the last build, and which decay flags rose. The register carries acceptance statistics with their true false-positive size printed beside them, half-lives, walk-forward standings, borrow breakevens, deal and reporting-window flags, and survivorship costs, quantified. The price source drops most names that stopped trading; that cost is measured on the page today, and a survivorship-correct US panel is the next build, the one that removes it rather than measures it. The caveats are on the page, not in a footnote, because the reader this desk is for checks.
What is PairDesk? A weekly pairs trading screener for US stocks and ETFs. Each build tests about 1.7 million pairs for cointegration and delivers a register of accepted pairs with half-lives, current z-scores, hedge ratios, borrow breakevens and out-of-sample checks, as one self-contained file plus a one-page brief.
How are the pairs tested? Engle-Granger in both orientations against MacKinnon finite-sample critical values, hedge ratios by total least squares, half-lives with the Kendall small-sample correction, a held-back year for out-of-sample checks, and a false-discovery q-value beside every test statistic.
Which markets? US stocks and ETFs, about 1,900 liquid names per build. Not India, not crypto, not FX.
Is there a free version? Yes, the complete 21 August 2026 build, aged on purpose: 12,000 register rows, 700 candidates with charts, the methodology inside. The decay table above shows what the aging costs.
Does it predict which pairs will make money? No, and this page says so. With acceptance recomputed inside each test year, accepted pairs do not beat correlation-matched rejected ones, and no column ranks the year ahead. The desk sells computation, freshness and correctness, not forward information.
What does it cost, and how does it arrive? $29, $39 or $49 a month by cadence, monthly, bi-weekly or weekly, delivered by email as a file that opens in any browser with nothing to install. The first ten subscribers take the weekly desk at $39 a month, locked. Firms take a five-seat license (five named people) with the register export at $199 a month, the first three at $149; custom desk builds from $6,000, quoted before anything is built.