aszer/etf
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(strategy): 取消数据不足标的剔除逻辑，保留所有标的以暴露策略问题

Browse Source 
- compute_factors: 不剔除数据不足/缺失率高的标的
- 改为警告并保留，因子值NaN时信号生成自动跳过
- 目的：暴露策略自身问题，后续支持更多大类资产
- 回测配置改为start_date=2000-01-01以测试更长历史
This commit is contained in:
aszerW
2026-05-15 23:18:44 +08:00
parent 763713213c
commit 85c20b4626
2 changed files with 25 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
strategies/rotation/config.yaml
View File

@@ -68,7 +68,7 @@ benchmark:
name: "沪深300"
# ==================== 回测参数 ====================
start_date: "2019-01-01"
start_date: "2000-01-01"
# ==================== 因子参数 ====================
# 动量/趋势窗口期（天数）

46
strategies/rotation/strategy.py
View File

@@ -277,7 +277,10 @@ class RotationStrategy(StrategyBase):
}
def compute_factors(self, data: dict) -> pd.DataFrame:
"""计算因子值（匹配原引擎：先计算因子再对齐到A股交易日历）"""
"""计算因子值（匹配原引擎：先计算因子再对齐到A股交易日历）
注意：不剔除数据不足的标的，保留所有标的以暴露策略问题
"""
index_data = data['index_data']
valid_codes = data['valid_codes']
@@ -299,53 +302,52 @@ class RotationStrategy(StrategyBase):
for code in valid_codes:
df = index_data[code].copy()
# 原引擎剔除逻辑：如果有OHLCV列，整行dropna()后再检查长度
# 这会剔除国债等只有close数据的标的（open/high/low全空）
# 检查是否有OHLCV数据
ohlcv_cols = ['open', 'high', 'low', 'close', 'volume']
has_ohlcv = all(col in df.columns for col in ['open', 'high', 'low', 'close'])
if has_ohlcv:
# 原引擎逻辑：整行dropna()后检查数据是否足够
# 如果有完整OHLCV，整行dropna()后提取close
df_clean = df[ohlcv_cols].dropna()
if len(df_clean) < self.n_days + 1:
print(f" ⚠ 剔除 {code}: OHLCV数据不足 ({len(df_clean)} < {self.n_days + 1})")
continue
close_series = df_clean['close']
close_series = df_clean['close'] if len(df_clean) > 0 else pd.Series(dtype=float)
else:
# 只有close列的情况
if 'close' in df.columns:
close_series = df['close'].dropna()
else:
close_series = df.dropna()
if len(close_series) < self.n_days + 1:
print(f" ⚠ 剔除 {code}: close数据不足 ({len(close_series)} < {self.n_days + 1})")
continue
# 检查数据长度并警告，但不剔除
if len(close_series) < self.n_days + 1:
print(f" ⚠ {code}: 数据不足 ({len(close_series)} < {self.n_days + 1})，保留但因子值可能为NaN")
# 原引擎逻辑：先在原始交易日历上计算因子
# rolling窗口使用的是原始交易日数据，不包含ffill填充的重复值
close_df = pd.DataFrame({'close': close_series})
factor_series = self._factor.compute(close_df)
# 然后对齐因子序列到A股交易日历（匹配原引擎逻辑）
factor_aligned = factor_series.reindex(a_share_dates, method='ffill')
if len(close_series) > 0:
close_df = pd.DataFrame({'close': close_series})
factor_series = self._factor.compute(close_df)
# 然后对齐因子序列到A股交易日历（匹配原引擎逻辑）
factor_aligned = factor_series.reindex(a_share_dates, method='ffill')
else:
# 没有数据，创建空的因子序列
factor_aligned = pd.Series(index=a_share_dates, dtype=float)
factor_values[code] = factor_aligned
final_valid_codes.append(code)
factor_df = pd.DataFrame(factor_values)
# 过滤缺失率过高的标的
# 检查缺失率并警告，但不剔除（保留所有标的以暴露策略问题）
total_rows = len(factor_df)
for code in final_valid_codes:
if code in factor_df.columns:
null_pct = factor_df[code].isnull().sum() / total_rows
if null_pct > 0.5:
print(f" ⚠ 剔除 {code}: 缺失率 {null_pct:.1%} 过高")
factor_df = factor_df.drop(columns=[code])
print(f" ⚠ {code}: 缺失率 {null_pct:.1%} 较高，保留但信号生成时可能跳过")
# 更新有效代码列表
data['valid_codes'] = [c for c in final_valid_codes if c in factor_df.columns]
# 不更新有效代码列表，保留所有原始代码
data['valid_codes'] = final_valid_codes
return factor_df