Connect Your AI to PairSync: Nightly Pairs Trading Statistics, On Tap
PairSync Research Team · Published 2026-11-03 · 9 min
Nightly pairs trading statistics, on tap
PairSync Research — November 2026
You ask your AI: "What are the strongest cointegration relationships in US equities from last night's run?"
It returns a ranked list with locked parameters, half-lives, calibration records, and the exact refit date. No improvisation. No generic cointegration lecture. Just the statistics from that night's refit — dated, hashed, and verifiable.
That is what the PairSync connector does. This article explains the layer it occupies, how to connect in three steps, and five worked examples you can run tonight.
The Missing Layer
Connector directories today offer three layers:
- Raw data and research feeds — prices, filings, news.
- Broker and custody connections — order entry, account state.
- Portfolio aggregation — what you hold, across venues.
None of them occupy the derived-statistics layer: a connector that takes raw prices and turns them into ranked, calibrated, refit-tracked statistics an AI can reason over.
PairSync is that layer. Each night it refits cointegration statistics across six markets — US equities, US ETFs, EU, India, Australia, and crypto — and publishes ranked, calibrated pair statistics: spread z-scores, locked parameters, unit ratios, calibration records, and backtest-tracked spread states.
Your AI can already read prices. Your broker connector can already place orders. PairSync sits between them and supplies the statistical composite the AI can cite.
The difference is concrete. An AI without calibrated statistics improvises. An AI with refit output cites locked parameters and calibration records. And the nightly cadence gives every conversation a shared, dated fact base: "tonight's refit" means the same thing to you and the AI.
How to Connect
- Sign in at terminal.pair-sync.com.
- Mint your API key from the terminal account page, or approve the PairSync connector at
https://mcp.pair-sync.com/mcp. - Ask your first question.
Full connection details and the methodology library live at pair-sync.com/ai and terminal.pair-sync.com/api.
The security model is deliberately boring: one API key, minted by you. Tokens are sealed short-lived envelopes. Nothing is persisted server-side. Every tool is read-only.
Worked Examples
Five questions, asked against real refit output. Every number below comes from the dated refits named in each example — you can reproduce them by pinning the same refit date.
1. Show me tonight's leaders in US equities
Show me tonight's leaders in US equities.
From the 2026-10-30 US refit (content hash sha256:1d2bb63a…), the top five by statistical composite:
| Pair | pair_score | half_life_days | hurst | entry_z | state | z_at_fit |
|---|---|---|---|---|---|---|
| NMIH / WSBC | 0.794 | 9.36 | 0.239 | 2.61 | closed | 1.04 |
| ACM / JBTM | 0.792 | 14.72 | 0.419 | 1.76 | closed | 1.73 |
| CTOS / RXO | 0.790 | 10.24 | 0.197 | 2.53 | closed | 0.21 |
| VKTX / WRBY | 0.783 | 9.33 | 0.251 | 2.66 | closed | −0.04 |
| FIG / TTAN | 0.778 | 11.95 | 0.258 | 2.32 | closed | −1.69 |
A calibration record is the set of locked parameters frozen at fit time — spread_mean, spread_vol, portfolio_beta, entry_z, exit_z, stop_z — together with the z_at_fit value and the content hash. It lets you verify later that the numbers your AI is citing match the exact night's solve.
Why this step matters: it replaces an improvised ranking with a dated, filter-qualified list the research workflow can start from.
2. Screen crypto pairs and explain the top one
Screen crypto pairs from the latest refit and explain the top one.
From the 2026-11-02 CRYPTO refit (content hash sha256:cefb22f7…), the top pair is WLFI-PERP / XPL-PERP.
Locked parameters: portfolio_beta 0.725, spread_mean −0.00767, spread_vol 0.00690, entry_z 2.30, exit_z 2.08, stop_z 4, half_life_days 14.98, hurst 0.249, pair_score 0.771, state closed, z_at_fit 0.847. The unit ratio follows from the locked portfolio_beta. The calibration self-check expects z ≈ 0.847 (±0.15) when recomputed from your own prices on the fitted_at close.
Why this step matters: the AI walks the pair card rather than inventing a relationship. It surfaces the locked parameters, the backtest-tracked state, and the calibration record so you can decide whether the statistics warrant further research.
3. Explain the methodology — where do I start?
Explain the PairSync methodology — where do I start?
The methodology library describes a plain-English staircase. Start with the nightly refit: the engine rebuilds the candidate universe after each market close, applies hard filters — Engle-Granger cointegration, Hurst, half-life, completed backtest — then ranks survivors on the seven-factor statistical composite. Parameters freeze at fit. You research the leaders.
The library covers the research workflow, the meaning of each locked parameter, and the distinction between floating values that update with the next refit and frozen recipes that stay on your side.
Why this step matters: it gives the AI a shared vocabulary, so subsequent questions stay inside the same research frame instead of drifting into generic pairs-trading explanations.
4. Freeze a recipe, then check it against tomorrow's refit
Freeze this pair's recipe for NMIH/WSBC, then check it against tomorrow's refit.
The AI calls prepare_arm with freeze: true. It returns the locked parameters from the 2026-10-30 US refit — portfolio_beta 1.100, spread_mean −0.809, spread_vol 1.099, entry_z 2.61, exit_z 2.66, stop_z 4 — plus the fit pin (refit_date and content hash). The recipe lives on your side: in agent memory, a local file, or venue notes. PairSync stores nothing and never auto-updates a frozen recipe.
On the next night you re-query the same pair. The AI compares the new locked parameters and z_at_fit against the frozen values and reports any drift.
Why this step matters: it turns a one-time research snapshot into a persistent, user-controlled reference that can be checked for change — without the server retaining state.
5. Build a chart template with the locked parameters
Build me a chart template for WLFI-PERP/XPL-PERP with the locked parameters.
The AI calls get_connector (venue generic_python or json_recipe, freeze: true). It returns a ready-to-paste template that embeds the locked parameters from the 2026-11-02 CRYPTO refit, the calibration self-check (expected z 0.847 ±0.15 on the fitted_at close), and the fit pin. The template is research-only: it draws the spread with the locked mean and bands; it does not place orders.
Why this step matters: the calibration self-check is baked in. Paste the template into your own charting or notebook environment and immediately verify that the parameters match the refit output before you do any further work.
What PairSync Deliberately Does Not Do
PairSync has no order path, no custody, no execution, and never manages money. Prices, decisions, and execution stay entirely on your side.
The tools are read-only. Frozen recipes are persisted by you; the server keeps nothing. Numeric fields are null where the solver did not solve — absences are honest.
These limits are the feature: the derived-statistics layer stays a research layer.
Connect and Run Tonight's Refit
Sign in at terminal.pair-sync.com, mint your key or approve the connector, and run tonight's refit output through your own questions. The statistics are dated, the parameters are locked, and the calibration records are there for you to verify.
PairSync provides statistical research and educational content only. Nothing here is financial advice, a recommendation, or an inducement to trade. All figures are historical refit output; past calibration does not predict future outcomes.