def cal_factor_return(self, sf_ids):
sfs = []
for sf_id in sf_ids:
sfs.append(
StockFactor.subclass(sf_id, StockFactor.stock_factors[sf_id]))
close = StockAsset.all_stock_nav()
ret = close.pct_change()
ret = ret[StockAsset.all_stock_info().index]
dates = ret.index
dates = dates[dates > '2000-01-01']
df_ret = pd.DataFrame(columns=sf_ids)
df_sret = pd.DataFrame(columns=StockAsset.all_stock_info().index)
for date, next_date in zip(dates[:-1], dates[1:]):
print 'cal_factor_return:', date
tmp_exposure = {}
tmp_ret = ret.loc[next_date].values
for sf in sfs:
tmp_exposure[sf.factor_id] = sf.exposure.loc[date]
tmp_exposure_df = pd.DataFrame(tmp_exposure)
tmp_exposure_df = tmp_exposure_df[sf_ids].fillna(0.0)
tmp_exposure_df = tmp_exposure_df.loc[
StockAsset.all_stock_info().index]
mod = sm.OLS(tmp_ret, tmp_exposure_df.values, missing='drop').fit()
df_ret.loc[next_date] = mod.params
df_sret.loc[next_date] = tmp_ret - np.dot(tmp_exposure_df.values,
mod.params)
return df_ret, df_sret
def cal_factor_return(self, sf_ids):
period = 21
sfs = []
for sf_id in sf_ids:
sfs.append(
StockFactor.subclass(sf_id, StockFactor.stock_factors[sf_id]))
close = StockAsset.all_stock_nav()
ret = close.pct_change(period).iloc[period:]
ret = ret[StockAsset.all_stock_info().index]
dates = ret.index
dates = dates[dates >= '2005-01-01']
df_ret = pd.DataFrame(columns=sf_ids)
df_sret = pd.DataFrame(columns=StockAsset.all_stock_info().index)
pool = Pool(len(sfs))
sfs = pool.map(multiprocess_load_factor_exposure, sfs)
pool.close()
pool.join()
for date, next_date in zip(dates[:-period], dates[period:]):
tmp_exposure = {}
tmp_ret = ret.loc[next_date].values
for sf in sfs:
tmp_exposure[sf.factor_id] = sf.exposure.loc[date]
#tmp_exposure[sf.factor_id] = fed[sf.factor_id].loc[date]
tmp_exposure_df = pd.DataFrame(tmp_exposure)
tmp_exposure_df = tmp_exposure_df[sf_ids].fillna(0.0)
tmp_exposure_df = tmp_exposure_df.loc[
StockAsset.all_stock_info().index]
mod = sm.OLS(tmp_ret, tmp_exposure_df.values, missing='drop').fit()
# mod = sm.WLS(tmp_ret, tmp_exposure_df.values, weights = tmp_amount, missing = 'drop').fit()
# print(mod.summary())
df_ret.loc[next_date] = mod.params
df_sret.loc[next_date] = tmp_ret - np.dot(tmp_exposure_df.values,
mod.params)
return df_ret, df_sret
def cal_indexposure(self, stock_id):
stock_quote = StockAsset.get_stock(stock_id).quote
stock_info = StockAsset.all_stock_info()
sf = pd.DataFrame(index=stock_quote.index)
sf_ind = stock_info.loc[stock_id].sk_swlevel1code
if sf_ind == self.sf_ind:
sf_exposure = 1
else:
sf_exposure = 0
sf['exposure'] = sf_exposure
return sf.exposure
def cal_factor_exposure(self):
all_stocks = StockAsset.all_stock_info()
factor_exposure = []
for desc_method in self.desc_methods:
stock_exposure = {}
for stock_id in all_stocks.index:
stock_exposure[stock_id] = desc_method(stock_id)
stock_exposure_df = pd.DataFrame(stock_exposure)
stock_exposure_df = StockFactor.stock_factor_filter(
stock_exposure_df)
stock_exposure_df = StockFactor.normalized(stock_exposure_df)
factor_exposure.append(stock_exposure_df)
factor_exposure_df = reduce(lambda x, y: x + y,
factor_exposure) / len(factor_exposure)
factor_exposure_df = factor_exposure_df[all_stocks.index]
self.exposure = factor_exposure_df
return factor_exposure_df
def valid_stock_table():
all_stocks = StockAsset.all_stock_info()
all_stocks = all_stocks.reset_index()
all_stocks = all_stocks.set_index(['sk_secode'])
st_stocks = StockAsset.stock_st()
all_stocks.sk_listdate = all_stocks.sk_listdate + timedelta(365)
engine = database.connection('caihui')
Session = sessionmaker(bind=engine)
session = Session()
sql = session.query(asset_stock.tq_qt_skdailyprice.tradedate,
asset_stock.tq_qt_skdailyprice.secode,
asset_stock.tq_qt_skdailyprice.tclose,
asset_stock.tq_qt_skdailyprice.amount).filter(
asset_stock.tq_qt_skdailyprice.secode.in_(
all_stocks.index)).statement
#过滤停牌股票
quotation_amount = pd.read_sql(sql,
session.bind,
index_col=['tradedate', 'secode'],
parse_dates=['tradedate'])
quotation = quotation_amount[['tclose']]
quotation = quotation.replace(0.0, np.nan)
quotation = quotation.unstack()
quotation.columns = quotation.columns.droplevel(0)
#60个交易日内需要有25个交易日未停牌
quotation_count = quotation.rolling(60).count()
quotation[quotation_count < 25] = np.nan
#过滤掉过去一年日均成交额排名后20%的股票
amount = quotation_amount[['amount']]
amount = amount.unstack()
amount.columns = amount.columns.droplevel(0)
year_amount = amount.rolling(252, min_periods=100).mean()
def percentile20nan(x):
x[x <= np.percentile(x, 20)] = np.nan
return x
year_amount = year_amount.apply(percentile20nan, axis=1)
quotation[year_amount.isnull()] = np.nan
session.commit()
session.close()
#过滤st股票
for i in range(0, len(st_stocks)):
secode = st_stocks.index[i]
record = st_stocks.iloc[i]
selecteddate = record.selecteddate
outdate = record.outdate
if secode in set(quotation.columns):
#print secode, selecteddate, outdate
quotation.loc[selecteddate:outdate, secode] = np.nan
#过滤上市未满一年股票
for secode in all_stocks.index:
if secode in set(quotation.columns):
quotation.loc[:all_stocks.loc[secode, 'sk_listdate'],
secode] = np.nan
quotation = quotation.rename(
columns=dict(zip(all_stocks.index, all_stocks.globalid)))
asset_stock_factor.update_valid_stock_table(quotation)