菜单
Evaluate cross-sectional factors with IC, Rank IC, ICIR, decay, and turnover in open-xquant; use for stock selection and multi-asset ranking factors.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when open-xquant artifacts must be checked for version/run/final traceability before comparison, report review, migration, or final selection.
Compile an open-xquant strategy spec preview and audit that material SPEC execution semantics are preserved in compiled_plan.json before backtests.
Audit strategy_idea_brief.json and the brainstorm conversation before strategy_spec.yaml construction begins.
Audit strategy_spec.yaml field provenance before backtests.
Author workspace-local OpenXQuant custom components from component_request.json, with tests, manifest hashing, catalog refresh, and no global SDK mutation.
Guide a user through the open-xquant pre-spec strategy idea workflow and produce strategy_idea_brief.json before any strategy_spec.yaml work begins.
| name | evaluate-cross-sectional |
| description | Evaluate cross-sectional factors with IC, Rank IC, ICIR, decay, and turnover in open-xquant; use for stock selection and multi-asset ranking factors. |
Use this when the factor ranks assets on the same date.
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)
Use thresholds as heuristics, not rules:
0.03: promising0.01 and 0.03: weak but worth inspecting0.1: unstableInclude: