xingwudao/open-xquant/evaluate-time-series

Evaluate Time-Series Factor

Use this when the factor predicts direction through time for one asset or a small rotation set.

Minimal SDK Pattern

from pathlib import Path
from oxq.factor_eval.bundle import create_bundle
from oxq.factor_eval.tearsheet import generate_tearsheet
factor_series = factor_df.stack().rename_axis(["date", "asset"]).rename("factor")
bundle = create_bundle(
    factor_values=factor_series,
    prices=prices_df,
    forward_periods=[1, 5, 20],
)
out = Path("/tmp/oxq_tearsheet")
result = generate_tearsheet(
    bundle=bundle,
    forward_periods=[1, 5, 20],
    output_dir=str(out),
)

If matplotlib is missing, install the chart extra:

uv sync --extra chart

Review Checklist

• t-day factor is evaluated against future returns, not same-day returns
• factor timestamps and price timestamps are aligned
• hit rate and P/L ratio are both reviewed
• decay curve is checked across multiple horizons
• cash or no-position periods are counted
• impossible trading days are excluded when that data exists

Interpretation

• hit rate above 55% can be useful only if P/L ratio is acceptable
• high hit rate with low P/L ratio can still lose money
• fast decay implies execution timing matters
• long cash periods reduce capital usage and should be reported

Red Lines

• Do not report hit rate alone.
• Do not skip T+1 alignment.
• Do not treat a tearsheet image as an audit.