def industry_neutral(factor,data_source, config):
"""行业中性化方式
每一个时间截面按照不同行业进行分层,每一层再进行打分
"""
N = config.total_groups
industry_name = config.neutral_industry_used
industry_axe = parse_industry(industry_name)
data = factor.data
is_ascending = True if factor.direction == -1 else False
all_ids = data.index.levels[1].unique().tolist()
all_dates = data.index.levels[0].tolist()
industry_info = data_source.sector.get_stock_industry_info(ids=all_ids,industry=industry_name,
dates=all_dates)
common = data.merge(industry_info, left_index=True, right_index=True, how='left')
group_info_list = []
for date in all_dates:
section_factor = common.ix[date]
group_id = section_factor.groupby(
industry_axe,group_keys=False).apply(_add_group_id,factor_name=factor.axe,
n_groups=N,ascending=is_ascending)
idx = pd.MultiIndex.from_product([[date], group_id.index])
group_id.index = idx
group_info_list.append(group_id)
grouping_info = pd.concat(group_info_list)
grouping_info.index.names = ['date','IDs']
factor.grouping_info['industry_neutral'] = grouping_info
def get_stock_industry_info(self,
ids,
industry='中信一级',
start_date=None,
end_date=None,
dates=None):
"""股票行业信息"""
dates = self.trade_calendar.get_trade_days(
start_date, end_date) if dates is None else dates
if not isinstance(dates, list):
dates = [dates]
symbol = parse_industry(industry)
industry_info = self.h5DB.load_factor(symbol,
'/indexes/',
ids=ids,
dates=dates)
return get_industry_names(symbol, industry_info)
def score_by_industry(factor_data, industry_name, factor_names=None, **kwargs):
factor_names = factor_names if factor_names is not None else list(
factor_data.columns)
industry_str = parse_industry(industry_name)
all_ids = factor_data.index.get_level_values(1).unique().tolist()
all_dates = factor_data.index.get_level_values(0).unique().tolist()
# 个股的行业信息与因子数据匹配
industry_info = data_source.sector.get_stock_industry_info(
all_ids, industry=industry_name, dates=all_dates).reset_index()
factor_data = factor_data.reset_index()
factor_data = pd.merge(factor_data, industry_info, how='left')
score = factor_data.set_index(['date', 'IDs']).groupby(
industry_str, group_keys=False).apply(ScoringFactors,
factors=factor_names,
**kwargs)
return score
def get_index_industry_weight(self,
ids,
industry_name='中信一级',
start_date=None,
end_date=None,
dates=None):
"""获取指数的行业权重"""
symbol = parse_industry(industry_name)
dates = self.trade_calendar.get_trade_days(
start_date, end_date) if dates is None else dates
index_weight = self.get_index_weight(ids, dates=dates)
all_stocks = list(index_weight.index.levels[1].unique())
industry_data = self.get_stock_industry_info(all_stocks,
industry=industry_name,
dates=dates)
common = pd.concat([index_weight, industry_data], join='inner', axis=1)
index_industry_weight = common.reset_index().groupby(
['date', symbol])[index_weight.columns[0]].sum()
return index_industry_weight
def industry_neutral(factor, **kwargs):
"""行业中性化方式
每一个时间截面按照不同行业进行分层,每一层再进行打分
"""
N = kwargs['total_groups']
industry_name = kwargs['industry_name']
industry_axe = parse_industry(industry_name)
data = factor.data
is_ascending = True if factor.direction == -1 else False
data_source = kwargs['env']._data_source
all_ids = data.index.get_level_values(1).unique().tolist()
all_dates = data.index.get_level_values(0).unique().tolist()
industry_info = data_source.sector.get_stock_industry_info(
ids=all_ids, industry=industry_name, dates=all_dates)
common = data.merge(industry_info,
left_index=True,
right_index=True,
how='left')
group_info_list = []
for date in all_dates:
section_factor = common.ix[date]
group_id = section_factor.groupby(
industry_axe, group_keys=False).apply(_add_group_id,
factor_name=factor.name,
n_groups=N,
ascending=is_ascending)
idx = pd.MultiIndex.from_product([[date], group_id.index])
group_id.index = idx
group_info_list.append(group_id)
grouping_info = pd.concat(group_info_list)
grouping_info.index.names = ['date', 'IDs']
grouping_info.columns = pd.MultiIndex.from_product([['industry_neutral'],
grouping_info.columns])
factor.group_info.from_frame(grouping_info)
def typical(factor,
name,
direction=None,
industry_neutral=True,
benchmark=None,
industry_name='中信一级',
prc=0.05,
**kwargs):
factor_data = factor.reset_index()
prc = 1 - prc if (direction == 1 or direction is None) else prc
if industry_neutral:
industry_str = parse_industry(industry_name)
all_ids = factor_data['IDs'].unique().tolist()
all_dates = factor_data['date'].unique().tolist()
# 个股的行业信息与因子数据匹配
industry_info = data_source.sector.get_stock_industry_info(
all_ids, industry=industry_name, dates=all_dates).reset_index()
factor_data = pd.merge(factor_data, industry_info, how='left')
quantile_per_industry = factor_data.groupby(['date', industry_str
])[name].quantile(prc)
quantile_per_industry.name = 'quantile_value'
factor_data = factor_data.join(quantile_per_industry,
on=['date', industry_str],
how='left')
# 股票选择，stocks=DataFrame[日期 IDs 因子值 行业 行业分位数]
if direction == 1:
stocks = factor_data[
factor_data[name] >= factor_data['quantile_value']]
else:
stocks = factor_data[
factor_data[name] <= factor_data['quantile_value']]
# 配置权重
benchmark_weight = data_source.sector.get_index_industry_weight(
benchmark, industry_name=industry_name,
dates=all_dates) # 基准指数的行业权重
stock_counts_per_industry = stocks.groupby(['date', industry_str
])['IDs'].count()
weight = (benchmark_weight /
stock_counts_per_industry).rename('Weight')
stocks = stocks.join(weight, on=['date', industry_str],
how='left').set_index(['date', 'IDs'])[['Weight']]
stocks.dropna(inplace=True)
sum_weight = stocks.groupby(level=0)['Weight'].sum()
stocks['Weight'] = stocks['Weight'] / sum_weight
else:
quantile_per_date = factor_data.groupby('date')[name].quantile(prc)
quantile_per_date.name = 'quantile_value'
factor_data = factor_data.join(quantile_per_date,
on='date',
how='left')
# 股票选择，stocks=DataFrame[日期 IDs 因子值 分位数]
if direction == 1:
stocks = factor_data[
factor_data[name] >= factor_data['quantile_value']]
else:
stocks = factor_data[
factor_data[name] <= factor_data['quantile_value']]
stock_counts_per_date = stocks.groupby('date')['IDs'].count()
stocks = stocks.join(1 / stock_counts_per_date.rename('Weight'),
on='date').set_index(['date', 'IDs'])
return stocks[['Weight']]
def typical(factor,
industry_neutral=True,
industry_name='中信一级',
prc=0.05,
**kwargs):
factor_data = factor.data.reset_index()
prc = 1 - prc if factor.direction == 1 else prc
if industry_neutral:
industry_str = parse_industry(industry_name)
all_ids = factor_data['IDs'].unique().tolist()
all_dates = factor_data['date'].unique().tolist()
benchmark = kwargs['env']._config.benchmark
# 个股的行业信息与因子数据匹配
industry_info = kwargs[
'env']._data_source.sector.get_stock_industry_info(
all_ids, industry=industry_name,
dates=all_dates).reset_index()
factor_data = pd.merge(factor_data, industry_info, how='left')
quantile_per_industry = factor_data.groupby(
['date', industry_str])[factor.name].quantile(prc)
quantile_per_industry.name = 'quantile_value'
factor_data = factor_data.join(quantile_per_industry,
on=['date', industry_str],
how='left')
# 股票选择，stocks=DataFrame[日期 IDs 因子值 行业 行业分位数]
if factor.direction == 1:
stocks = factor_data[
factor_data[factor.name] >= factor_data['quantile_value']]
else:
stocks = factor_data[
factor_data[factor.name] <= factor_data['quantile_value']]
# 配置权重
benchmark_weight = kwargs[
'env']._data_source.sector.get_index_industry_weight(
benchmark, industry_name=industry_name,
dates=all_dates) # 基准指数的行业权重
stock_counts_per_industry = stocks.groupby(['date', industry_str
])['IDs'].count()
weight = (benchmark_weight /
stock_counts_per_industry).rename('Weight')
stocks = stocks.join(weight, on=['date', industry_str],
how='left').set_index(['date', 'IDs'])[['Weight']]
stocks.dropna(inplace=True)
sum_weight = stocks.groupby(level=0)['Weight'].sum()
stocks['Weight'] = stocks['Weight'] / sum_weight
else:
quantile_per_date = factor_data.groupby('date')[factor.name].quantile(
prc)
quantile_per_date.name = 'quantile_value'
factor_data = factor_data.join(quantile_per_date,
on='date',
how='left')
# 股票选择，stocks=DataFrame[日期 IDs 因子值 分位数]
if factor.direction == 1:
stocks = factor_data[
factor_data[factor.name] >= factor_data['quantile_value']]
else:
stocks = factor_data[
factor_data[factor.name] <= factor_data['quantile_value']]
stock_counts_per_date = stocks.groupby('date')['IDs'].count()
stocks = stocks.set_index(['date', 'IDs'])
stocks['Weight'] = 1 / stock_counts_per_date
if 'typical' not in factor.stock_list:
factor.stock_list['typical'] = pd.DataFrame()
factor_stock_list = factor.stock_list['typical'].append(stocks[['Weight']])
factor_stock_list = factor_stock_list[~factor_stock_list.index.duplicated(
keep='last')]
factor.stock_list['typical'] = factor_stock_list