def weighted_mean_return(factor_data: pd.DataFrame, grouper: list):
"""
计算加权平均收益/标准差
"""
forward_returns_columns = get_forward_returns_columns(factor_data.columns)
def agg(values, weights):
count = len(values)
average = np.average(values, weights=weights, axis=0)
variance = (np.average(
(values - average)**2, weights=weights, axis=0) * count / max(
(count - 1), 1))
return pd.Series([average, np.sqrt(variance), count],
index=["mean", "std", "count"])
group_stats = factor_data.groupby(grouper)[forward_returns_columns.append(
pd.Index(["weights"
]))].apply(lambda x: x[forward_returns_columns].apply(
agg, weights=x["weights"].fillna(0.0).values))
mean_ret = group_stats.xs("mean", level=-1)
std_error_ret = group_stats.xs("std", level=-1) / np.sqrt(
group_stats.xs("count", level=-1))
return mean_ret, std_error_ret
def demean_forward_returns(factor_data: pd.DataFrame,
grouper: list = None) -> pd.DataFrame:
"""
按照分组对因子远期收益进行去均值
参数
---
:param factor_data: 因子远期收益, 索引为 ['日期' '股票'] 的 MultiIndex,
columns 为因子远期收益
:param grouper: 分组信息，如果为 None, 则默认按日期进行去均值
返回值
---
:return adjust_forward_returns: 去均值后的因子远期收益
"""
factor_data = factor_data.copy()
if not grouper:
grouper = ["datetime"]
cols = get_forward_returns_columns(factor_data.columns)
factor_data[cols] = factor_data.groupby(
grouper, as_index=False)[cols.append(pd.Index(
["weights"]))].apply(lambda x: x[cols].subtract(np.average(
x[cols], axis=0, weights=x["weights"].fillna(0.0).values),
axis=1))
return factor_data
def factor_returns(
factor_data: pd.DataFrame,
demeaned: bool = True,
group_adjust: bool = False,
equal_weight: bool = False,
by_asset: bool = False,
):
"""
计算按因子值加权的投资组合收益
参数
---
:param factor_data: 因子数据
:param demeaned: 是否构建多空组合
:param group_adjust: 是否按分组进行多空组合
:param equal_weight: 针对因子中位数分别构建多空组合
:param by_asset: 按股票展示组合收益, 默认为 False
返回值
---
"""
weights = factor_weights(factor_data, demeaned, group_adjust, equal_weight)
weighted_returns = factor_data[get_forward_returns_columns(
factor_data.columns)].multiply(weights, axis=0)
if by_asset:
returns = weighted_returns
else:
returns = weighted_returns.groupby(level="datetime").sum()
return returns
def src_ic(group):
f = group["factor"]
_ic = group[get_forward_returns_columns(
factor_data.columns
)].apply(lambda x: stats.spearmanr(x,
f)[0])
return _ic
def factor_alpha_beta(
factor_data: pd.DataFrame,
returns: pd.DataFrame = None,
demeaned: bool = True,
group_adjust: bool = False,
equal_weight: bool = False,
):
"""
计算因子的 alpha (超额收益), alpha 的 t-统计量 以及 beta 值
参数
---
:param factor_data: 索引为 ['日期' '股票'] 的 MultiIndex, values 包括因子值，远期收益，因子分位，因子分组 [可选]
:param returns: 因子远期收益，默认为 None, 如果为 None 的时候，会通过调用 `factor_returns` 来计算相应的收益
:param demeaned: 是否基于一个多空组合
:param group_adjust: 是否进行行业中性处理
:param equal_weight:
返回
---
"""
if returns is None:
returns = factor_returns(
factor_data,
demeaned,
group_adjust,
equal_weight
)
universe_ret = (
factor_data.groupby(level="datetime")[get_forward_returns_columns(
factor_data.columns
)].mean().loc[returns.index]
)
if isinstance(returns, pd.Series):
returns.name = universe_ret.columns.values[0]
returns = pd.DataFrame(returns)
alpha_beta = pd.DataFrame()
for period in returns.columns.values:
x = universe_ret[period].values
y = returns[period].values
x = add_constant(x)
reg_fit = OLS(y, x).fit()
try:
alpha, beta = reg_fit.params
except ValueError:
alpha_beta.loc["Ann. alpha", period] = np.nan
alpha_beta.loc["beta", period] = np.nan
else:
freq_adjust = pd.Timedelta(days=DAYS_PER_YEAR) / pd.Timedelta(
utils.get_period(period.replace("period_",
""))
)
alpha_beta.loc["Ann. alpha",
period] = (1 + alpha)**freq_adjust - 1.0
alpha_beta.loc["beta", period] = beta
return alpha_beta
def create_summary_tear_sheet(factor_data: pd.DataFrame,
long_short: bool = True,
group_neutral: bool = False):
"""
"""
# Return Analysis
mean_quant_ret, std_quantiles = perf.mean_return_by_quantile(
factor_data,
by_group=False,
demeaned=long_short,
group_adjust=group_neutral)
mean_quant_rateret = mean_quant_ret.apply(
utils.rate_of_return, axis=0, base_period=mean_quant_ret.columns[0])
mean_quant_ret_bydatetime, std_quant_bydatetime = perf.mean_return_by_quantile(
factor_data,
by_datetime=True,
by_group=False,
demeaned=long_short,
group_adjust=group_neutral,
)
mean_quant_rateret_bydatetime = mean_quant_ret_bydatetime.apply(
utils.rate_of_return,
axis=0,
base_period=mean_quant_ret_bydatetime.columns[0])
std_quant_bydatetime = std_quant_bydatetime.apply(
utils.std_conversion,
axis=0,
base_period=std_quant_bydatetime.columns[0])
alpha_beta = perf.factor_alpha_beta(factor_data,
demeaned=long_short,
group_adjust=group_neutral)
mean_ret_spread_quant, std_spread_quant = perf.mean_returns_spread(
mean_quant_rateret_bydatetime,
factor_data["factor_quantile"].max(),
factor_data["factor_quantile"].min(),
std_err=std_quant_bydatetime,
)
periods = utils.get_forward_returns_columns(factor_data.columns)
fr_cols = len(periods)
vertical_sections = 2 + fr_cols * 3
gr = GridFigure(rows=vertical_sections, cols=1)
plot_quantile_statistics_table(factor_data)
plot_quantile_returns_bar(mean_quant_rateret,
by_group=False,
ylim_percentiles=None,
ax=gf.next_row())
def create_summary_tear_sheet(
self,
by_datetime=True,
by_group: bool = False,
long_short: bool = True,
group_neutral: bool = False,
):
"""
创建一个小型的汇总表格，包括因子的收益率分析，IC 值，换手率等分析
参数
---
:param factor_data: 因子数据
:param long_short: 是否构建多空组合，在该组合上进行进行分析。
:param group_neutral: 是否进行行业中性
"""
# Returns Analysis
mean_quant_ret, std_quant = self.calc_mean_return_by_quantile(
by_group=by_group, demeaned=long_short, group_adjust=group_neutral)
mean_quant_rateret = mean_quant_ret.apply(
utils.rate_of_return,
axis=0,
base_period=mean_quant_ret.columns[0]
)
mean_quant_ret_bydatetime, std_quant_bydatetime = self.calc_mean_return_by_quantile(
by_datetime=True, demeaned=long_short, group_adjust=group_neutral)
mean_quant_rateret_bydatetime = mean_quant_ret_bydatetime.apply(
utils.rate_of_return,
axis=0,
base_period=mean_quant_ret_bydatetime.columns[0],
)
std_quant_rate_bydatetime = std_quant_bydatetime.apply(
utils.rate_of_return,
axis=0,
base_period=std_quant_bydatetime.columns[0]
)
alpha_beta = self.calc_factor_alpha_beta(
demeaned=long_short,
group_adjust=group_neutral
)
mean_ret_spread_quant, std_spread_quant = self.calc_mean_returns_spread(
)
fr_cols = utils.get_forward_returns_columns(
self._clean_factor_data.columns
)
vertical_sections = 2 + len(fr_cols) * 3
gf = GridFigure(rows=vertical_sections, cols=1)
plotting.plot_quantile_statistics_table(self._clean_factor_data)
plotting.plot_returns_table(
alpha_beta,
mean_quant_rateret,
mean_ret_spread_quant
)
plotting.plot_quantile_returns_bar(
mean_quant_rateret,
by_group=False,
ylim_percentiles=None,
ax=gf.next_row()
)
# Information Analysis
ic = perf.factor_information_coefficient(self._clean_factor_data)
plotting.plot_information_table(ic)
# Turnover Analysis
# FIXME: 股票是 T+1，意味着频率只能是 Day 及以上频率
quantile_factor = self._clean_factor_data["factor_quantile"]
quantile_turnover = {
p: pd.concat(
[
perf.quantile_turnover(quantile_factor,
q,
p)
for q in range(1,
int(quantile_factor.max()) + 1)
],
axis=1,
)
for p in self.periods
}
autocorrelation = pd.concat(
[
perf.factor_rank_autocorrelation(
self._clean_factor_data,
period
) for period in self.periods
],
axis=1,
)
plotting.plot_turnover_table(autocorrelation, quantile_turnover)
plt.show()
gf.close()
def create_summary_tear_sheet(
factor_data: pd.DataFrame,
by_datetime: bool = True, # 按日期计算
by_group: bool = False, # 按分组计算
long_short: bool = True, # 多空组合
group_neutral: bool = False, # 分组中性
periods: Union[int, Tuple[int], List[int]] = 1,
frequence: str = '1d'):
"""
创建一个小型的汇总表格，包括因子的收益率分析，IC 值，换手率等分析
参数
---
:param factor_data: 因子数据
:param long_short: 是否构建多空组合，在该组合上进行进行分析。
:param group_neutral: 是否进行行业中性
"""
if isinstance(periods, int):
periods = [
periods,
]
# 收益分析
mean_quant_ret, std_quant = perf.mean_return_by_quantile(
factor_data,
by_group=by_group,
demeaned=long_short,
group_adjust=group_neutral)
mean_quant_rateret = mean_quant_ret.apply(
utils.rate_of_return, axis=0, base_period=mean_quant_ret.columns[0])
std_quant_rate = std_quant.apply(utils.std_conversion,
axis=0,
base_period=std_quant.columns[0])
mean_quant_ret_bydatetime, std_quant_bydatetime = perf.mean_return_by_quantile(
factor_data,
by_datetime=by_datetime,
by_group=by_group,
demeaned=long_short,
group_adjust=group_neutral)
mean_quant_rateret_bydatetime = mean_quant_ret_bydatetime.apply(
utils.rate_of_return,
axis=0,
base_period=mean_quant_ret_bydatetime.columns[0])
std_quant_rate_bydatetime = std_quant_bydatetime.apply(
utils.std_conversion,
axis=0,
base_period=std_quant_bydatetime.columns[0])
alpha_beta = perf.factor_alpha_beta(factor_data=factor_data,
demeaned=long_short,
group_adjust=group_neutral)
mean_ret_spread_quant, std_spread_quant = perf.mean_returns_spread(
mean_quant_rateret,
upper_quant=factor_data.factor_quantile.max(),
lower_quant=factor_data.factor_quantile.min(),
std_err=std_quant_rate)
fr_cols = utils.get_forward_returns_columns(factor_data.columns)
vertical_sections = 2 + len(fr_cols) * 3
gf = GridFigure(rows=vertical_sections, cols=1)
plotting.plot_quantile_statistics_table(factor_data)
plotting.plot_returns_table(alpha_beta, mean_quant_rateret,
mean_ret_spread_quant)
plotting.plot_quantile_returns_bar(mean_quant_rateret,
by_group=by_group,
ylim_percentiles=None,
ax=gf.next_row())
# Information Analysis
ic = perf.factor_information_coefficient(factor_data)
plotting.plot_information_table(ic)
# Turnover Analysis
# FIXME: 股票是 T+1，意味着频率只能是 Day 及以上频率
quantile_factor = factor_data["factor_quantile"]
quantile_turnover = {
p: pd.concat(
[
perf.quantile_turnover(quantile_factor, q, p)
for q in range(1,
int(quantile_factor.max()) + 1)
],
axis=1,
)
for p in periods
}
autocorrelation = pd.concat(
[
perf.factor_rank_autocorrelation(factor_data, period)
for period in periods
],
axis=1,
)
plotting.plot_turnover_table(autocorrelation, quantile_turnover)
plt.show()
gf.close()
