Qinvia Pairs Trading

Clean baseline for reproducible, cost-aware pairs-trading research.

Status: prepared for a next-generation implementation. The main branch intentionally contains no strategy code, empirical result, or performance claim until the new research contract is implemented and reviewed.

Preserved legacy

The 2023–2024 notebook project remains available in two permanent references:

• immutable tag: legacy-2024-03;
• browsable branch: legacy/2024-notebooks.

Those materials are historical artifacts, not the specification for the new implementation. They have not been rewritten or presented as current evidence.

Research contract for the rewrite

The next implementation should make the following choices explicit and auditable:

1. Universe and data: market, instruments, timestamps, corporate actions, delistings, missing observations, and survivorship controls.
2. Pair formation: economic rationale, clustering or screening rules, correlation versus cointegration, stationarity diagnostics, and multiple-testing controls.
3. Validation: separate formation, validation, and test periods with no parameter or universe leakage.
4. Trading model: signal construction, hedge-ratio estimation, entry and exit rules, sizing, rebalance timing, and failure handling.
5. Costs and feasibility: fees, spread, slippage, borrow or funding, turnover, liquidity, and execution assumptions.
6. Robustness: regime stability, sensitivity analysis, concentration, capacity constraints, and negative or inconclusive results.
7. Reproducibility: deterministic fixtures, executable examples, tests, provenance, and versioned result artifacts.

Suggested repository shape

src/qinvia_pairs_trading/   reusable research and validation code
tests/                      deterministic, network-free tests
examples/                   small executable examples and fixtures
notebooks/                  concise analyses built on the public package API
docs/                       methodology, data contract, findings and limits
artifacts/                  lightweight, versioned result summaries

Definition of ready

A public release should include a clean-environment quickstart, a bounded worked example, methodology and limitations, cost-aware evidence, explicit failure modes, and conclusions proportionate to the observed results. Historical performance must never be presented as evidence of future profitability.

GitHub Actions is intentionally disabled by default. GitHub is used as a public distribution and archival surface; routine verification runs locally.

License

The clean baseline and future repository work are licensed under the Apache License 2.0. The previous README's MIT statement was incorrect and has been removed.