evaluate-cross-sectional

Evaluate Cross-Sectional Factor

Use this when the factor ranks assets on the same date.

Minimal SDK Pattern

import pandas as pd

from oxq.data.market import LocalMarketDataProvider
from oxq.factor_eval.metrics import (
    compute_decay,
    compute_ic,
    compute_icir,
    compute_rank_ic,
    compute_turnover,
)

symbols = ["SPY", "QQQ", "IWM"]
market = LocalMarketDataProvider(data_dir="/path/to/parquet")

factor_df = pd.DataFrame()
prices_df = pd.DataFrame()
for sym in symbols:
    bars = market.get_bars(sym, "2018-01-01", "2024-12-31")
    prices_df[sym] = bars["close"]
    factor_df[sym] = bars["close"].pct_change(20)

factor_df = factor_df.dropna()
prices_df = prices_df.loc[factor_df.index]
forward_returns = prices_df.pct_change(1).shift(-1).dropna()

common = factor_df.index.intersection(forward_returns.index)
factor_df = factor_df.loc[common]
forward_returns = forward_returns.loc[common]

ic = compute_ic(factor=factor_df, forward_returns=forward_returns)
rank_ic = compute_rank_ic(factor=factor_df, forward_returns=forward_returns)
icir = compute_icir(ic["mean"], ic["std"])
decay = compute_decay(factor=factor_df, prices=prices_df, horizons=[1, 3, 5, 10, 20])
turnover = compute_turnover(factor=factor_df)

Review Checklist

• enough symbols for cross-sectional claims
• no forward-return leakage
• factor and returns aligned on common dates
• NaN rows removed deliberately
• multiple horizons checked
• turnover considered with likely trading cost

Interpretation

Use thresholds as heuristics, not rules:

• IC mean above 0.03: promising
• IC mean between 0.01 and 0.03: weak but worth inspecting
• ICIR below 0.1: unstable
• high turnover: likely cost-sensitive even with positive IC
• fast decay: signal may require execution assumptions the backtest cannot meet

Report Shape

Include:

• factor name and formula
• symbols and date range
• number of dates and symbols
• horizon results
• IC mean, IC std, ICIR, Rank IC
• turnover
• decay pattern
• pass, weak, or fail conclusion with caveats