Basis Momentum

A compact, public research package for a Goldman-style commodity basis-momentum replication and a liquidity-ranked contract-curve extension.

Scope

Included:

• Unscaled F0/F6 cross-sectional baseline using F3 same-contract close-to-close returns.
• A separate 10% target-volatility overlay.
• Aligned L0-L1, L0-L2, and a 50/50 combination of separately constructed liquidity books.
• Compact committed daily results, annual tables, summaries, figures, checksums, tests, executed notebook, and a static documentation site.

Excluded:

• Licensed raw data, local SQLite database, proprietary PDFs, exploratory output trees, capacity studies, unfinished signals, and unrelated experiments.

Quick start

python -m pip install -e '.[test]'
pytest -q
python scripts/run_goldman_style.py --dry-run
python scripts/run_liquidity_extension.py --dry-run

The two wrappers require a caller-supplied SQLite database only for local regeneration. See the data contract for its schema. The committed notebook and result snapshots work without it.

Method

The F0/F6 signal is the difference between 252-day compounded returns of the nearest and seventh listed contracts. Weekly centered-rank weights are dollar neutral; the return leg is F3 same-contract close-to-close, with 5 bp one-way turnover cost.

The liquidity extension ranks eligible contracts by maturity after the liquidity screen. Its L0-L1 and L0-L2 signals sum 120 days of log-return spreads. The displayed strategies share the same daily performance window, 2019-01-07 to 2026-07-10. The 50/50 strategy combines separate weekly books before backtesting, so its gross return is exactly the half-and-half gross-return identity; costs can differ after position netting.

Results and caveats

Open the executed notebook, published results, provenance, or the static research site. This is not an exact Bloomberg index replication and not investment advice. The extension uses a main-contract return proxy rather than a locked-contract execution simulation; inverse-volatility weighting, capacity, integer lots, margin, and limit-move handling are out of scope.

All result files are listed in results/checksums.sha256.