q.cross_section Roadmap

• Cross-sectional ranks with explicit date and symbol columns
• Group and sector neutralization
• Cross-sectional z-scores and winsorization
• Add a group-weighted-return tutorial when a multi-symbol demo dataset with real sector classifications is available

Elo follow-up issues

q.cross_section.compute_elo is migrated with symbol replacing insref. sectorid remains a required input because the current algorithm only matches symbols inside the same sector.

• Add sector classifications to a multi-symbol demo dataset before adding an Elo tutorial. The bundled AAPL/SPY/BTC-USD datasets contain OHLCV only and do not have sectorid; inventing sectors in an example would conceal a real required input and produce a misleading demonstration.

• Redesign match scheduling as deterministic simultaneous matches. The migrated algorithm independently selects each symbol’s nearest-rated peers, resolves equal-distance candidates in snapshot order, and updates only the current symbol during that pass. A pair can therefore be selected in both directions or only one direction, and multiple opponents are applied sequentially. This is not a conventional paired simultaneous Elo update and need not conserve total rating, especially when K-factors differ.

• Resolve the K-factor threshold conflict. The active legacy _assign_k compares the current Elo rating directly with low_qt and high_qt, using K=40 for the first 60 days, K=10 above the high threshold, K=30 below the low threshold, and K=20 otherwise. The public defaults are 0.2 and 0.8, which look like quantiles and came from an alternate volatility-quantile design left commented out in the legacy source. Ratings start near 1500, so almost every non-IPO symbol is above 0.8 and receives K=10. This can suppress rating responsiveness and makes the lower threshold effectively unreachable. Preserve it for migration compatibility until research determines whether the API should use rating levels such as 1300/1800 or actual volatility quantiles.

• Define missing-close and calendar semantics. The migrated code forward-fills at most three observed rows per symbol before computing returns. Those are rows, not calendar or exchange sessions, so an irregular dataset can bridge a much longer elapsed period and assign its eventual move to the wrong date. Decide whether gaps should be rejected, aligned to an explicit exchange calendar, or filled under a documented session-based policy.