Cross-Sectional Analysis

q.cross_section owns characteristics calculated across a universe, including ranks, neutralization, and relative rating models such as Elo.

Group-weighted returns

q.cross_section.group_weighted_return aggregates long-form asset returns by date and a required group column, which defaults to sectorid. It uses a return column when available, otherwise calculates simple returns from close. Other price or return columns must be supplied with an explicit field_type.

Equal, market-cap, volume, inverse-volatility, and custom-column weights are supported. Non-equal weights are lagged by one symbol observation by default so the return ending at time \(t\) does not use information first observed at \(t\). Missing returns or non-positive weights are excluded, and valid weights are renormalized within each date and group.

The result contains one row per date and group with weighted_return, asset_count, and the raw weight_sum. A tutorial will be added when a multi-symbol dataset with real sector classifications is available.

Elo ratings

q.cross_section.compute_elo compares each symbol with nearby-rated symbols in the same sector on each date. Call it once with the complete long-form universe; do not group by symbol before calling it.

The input requires date, symbol, close, and sectorid. The sector column is not optional because matches are constrained to peers in the same sector. Missing sector values are placed in one "UnknownSector" bucket.

rated = q.cross_section.compute_elo(
    daily_prices,  # date, symbol, close, sectorid
    daily_vol_window=20,
    matches_per_stock=1,
    initial_elo=1500,
    rf_data=daily_risk_free_rates,  # optional: date, rate
)

The result includes prepared return/volatility columns, the current elo, the day’s elo_change, and cumulative matches_played. If rf_data is supplied and use_excess_returns=True, matches use log return minus the aligned daily risk-free rate; otherwise they use log returns.

This is a migration of the legacy model, not a claim that every modeling choice has been settled. See the cross-section roadmap for the known match scheduling, K-factor, sample-data, and missing-price issues.

A factor or group return measures cross-sectional characteristics; portfolio selection and position sizing belong in q.signal and q.portfolio.