def __init__(self, **kwargs):
self._mu = kwargs.get('mu', None)
self._sd = kwargs.get('sd', None)
self._ir = kwargs.get('ir', None)
self._mu_min = np.nan
self._mu_max = np.nan
self._desc = kwargs.get('desc', 'Portfolio')
self._tol = kwargs.get('tol', 1e-10)
self._asset_return = kwargs.get('asset_return', None)
self._asset_mu = kwargs.get('asset_mu', None)
self._asset_cov = kwargs.get('asset_cov', None)
self._asset_weight = kwargs.get('asset_weight', None)
self._asset_name = kwargs.get('asset_name', None)
pyFinWarning(
not (self._asset_return is not None
and self._asset_mu is not None), UserWarning,
"When both asset return and asset mu is given, only asset mu {0} will be used"
.format(self._asset_mu))
if self._asset_name is None:
if isinstance(self._asset_mu, pd.DataFrame) or isinstance(
self._asset_mu, pd.Series):
self._asset_name = self._asset_mu.index
elif isinstance(self._asset_cov, pd.DataFrame):
self._asset_name = self._asset_cov.index
elif isinstance(self._asset_weight, pd.DataFrame) or isinstance(
self._asset_mu, pd.Series):
self._asset_name = self._asset_weight.index
else:
self._asset_name = pd.Index(['Asset_' + str(i) for i in range(len(self._asset_cov))]) \
if self._asset_cov is not None else \
pd.Index(['Asset_' + str(i) for i in range(len(self._asset_return))])
self._nb_asset = len(self._asset_name)
if self._asset_mu is None:
if self._asset_return is None:
self._asset_mu = pd.Series(
{'mu': np.repeat(np.nan, self._nb_asset, 0)},
index=self._asset_name)
else:
mu = np.mean(self._asset_return, axis=0)
self._asset_mu = pd.Series({'mu': mu}, index=self._asset_name)
else:
self._mu_min = np.ceil(
self._asset_mu.min()[0] / self._tol) * self._tol
self._mu_max = np.floor(
self._asset_mu.max()[0] / self._tol) * self._tol
if self._asset_cov is None:
if self._asset_return is None:
tmp = np.empty((self._nb_asset, self._nb_asset))
tmp[:] = np.nan
else:
tmp = np.cov(self._asset_return)
self._asset_cov = pd.DataFrame(tmp,
index=self._asset_name,
columns=self._asset_name)
def neutralize(factors,
industries,
caps=None,
na_handler=FactorNAHandler.ReplaceWithMedian):
"""
:param factors: pd.Series, 原始截面因子
:param industries: pd.Series, value = 行业名称
:param caps: optional, pd.Series, value = caps value
:param na_handler: enum, handler for na values
:return: 中性化后的因子
"""
# 通过concat把数据对齐
pyFinWarning(factors.size == industries.size, Warning,
"size of factors does not equal to that of industries")
if caps is None:
data = pd.concat([factors, industries],
join_axes=[factors.index],
axis=1)
lcap = None
else:
pyFinWarning(factors.size == caps.size, Warning,
"size of factors does not equal to that of caps")
data = pd.concat([factors, industries, caps],
join_axes=[factors.index],
axis=1)
lcap = np.log(data[data.columns[2]])
factors = data[data.columns[0]]
industries = data[data.columns[1]]
# 把没有行业对应的变成'other'
industries = industries.fillna('other')
# 把没有市值的设置成均值
if lcap is not None:
lcap = factor_na_handler(lcap, na_handler)
factors = factor_na_handler(factors, na_handler)
linreg = LinearRegression(fit_intercept=False)
y = factors
x = get_industry_matrix(industries, lcap)
model = linreg.fit(x, y)
coef = np.mat(linreg.coef_)
a = np.dot(x, coef.T)
residues = y.values - a.A1
ret = pd.Series(residues, index=factors.index, name=factors.name).dropna()
return ret
def format_raw_data(raw_data, freq, field, return_type):
ret = pd.DataFrame()
if len(raw_data) > 0:
if freq == FreqType.EOD:
raw_data['tradeDate'] = pd.to_datetime(raw_data['tradeDate'])
if return_type == DfReturnType.DateIndexAndSecIDCol:
pyFinWarning(
len(field) == 1, Warning,
'Only the columns with name=field[0] will be used as values in pivot table'
)
ret = raw_data.pivot(index='tradeDate',
columns='secID',
values=field[0])
else:
ret = raw_data['tradeDate', 'secID', field]
ret = ret.set_index('tradeDate')
return ret
def testPyFinWarning(self):
with warnings.catch_warnings(record=True) as warning_list:
warnings.simplefilter("always")
pyFinWarning(1 == 2, DeprecationWarning)
self.assertTrue(any(item.category == DeprecationWarning for item in warning_list))