def rolling_regress_1(y, x, window=5):
try:
rolling_ys = rolling_windows(y, window)
rolling_xs = rolling_windows(x, window)
except Exception as e:
print('debug')
bet = pd.Series()
# enumerate 形成带 i 的一个迭代器
for i, (rolling_x, rolling_y) in enumerate(zip(rolling_xs, rolling_ys)):
tmp_index = y.index[i + window - 1]
try:
model = sm.OLS(rolling_y, rolling_x)
result = model.fit()
params = result.params
b_v = params[0]
# print(result.params)
# print(result.summary())
except:
print(i)
raise
b = pd.Series(index=[tmp_index], data=b_v)
bet = pd.concat([bet, b])
return bet
def __init__(self, y, x=None, window=None, has_const=False,
use_const=True):
self.x, self.y, self.k = _clean_xy(y=y, x=x, has_const=has_const,
use_const=use_const)
self.window = self.n = window
self.xwins = utils.rolling_windows(self.x, window=window)
self.ywins = utils.rolling_windows(self.y, window=window)
self.solution = _rolling_lstsq(self.xwins, self.ywins)
self.has_const = has_const
self.use_const = use_const
def __init__(self, y, x=None, window=None, hasconst=False, names=None):
if window is None:
raise ValueError('must specify `window`')
OLS.__init__(self, y=y, x=x, hasconst=hasconst, names=names)
self.xwins = utils.rolling_windows(self.x, window=window)
self.ywins = utils.rolling_windows(self.y, window=window)
# TODO: iterator?
self.models = [
OLS(y=ywin, x=xwin) for ywin, xwin in zip(self.ywins, self.xwins)
]
def __init__(self, r, proxies, window, sumto=1.):
self.r = r
self.proxies = proxies
self.window = int(window)
self.sumto = sumto
self.newidx = r.index[window - 1:]
self.cols = proxies.columns
self.n = proxies.shape[1]
# TODO: constrain
self._r = utils.rolling_windows(r, window=window)
self._proxies = utils.rolling_windows(proxies, window=window)
def _wrap_multidx(self, stat, name=None):
if name is None:
name = stat[1:]
outer = np.repeat(self.ridx, self.window)
inner = np.ravel(utils.rolling_windows(self.index.values,
window=self.window))
return Series(getattr(self, stat).flatten(), index=[outer, inner],
name=name).rename_axis(['end', 'subperiod'])
def _rolling_regress(self, y, x, window=5, half_life=None,
intercept=True, verbose=False, fill_na=0):
fill_args = {'method': fill_na} if isinstance(fill_na, str) else {'value': fill_na}
x, y = self._align(x, y)
stocks = y.columns
if half_life:
weight = self._get_exp_weight(window, half_life)
else:
weight = 1
start_idx = x.loc[pd.notnull(x).values.flatten()].index[0]
x, y = x.loc[start_idx:], y.loc[start_idx:,:]
rolling_ys = rolling_windows(y, window)
rolling_xs = rolling_windows(x, window)
beta = pd.DataFrame()
alpha = pd.DataFrame()
sigma = pd.DataFrame()
for i, (rolling_x, rolling_y) in enumerate(zip(rolling_xs, rolling_ys)):
rolling_y = pd.DataFrame(rolling_y, columns=y.columns,
index=y.index[i:i+window])
window_sdate, window_edate = rolling_y.index[0], rolling_y.index[-1]
stks_to_regress = sorted(set(self._get_codes_listed(stocks, window_sdate)) & \
set(self._get_codes_not_delisted(stocks, window_edate)))
rolling_y = rolling_y[stks_to_regress].fillna(**fill_args)
try:
b, a, resid = self._regress(rolling_y.values, rolling_x,
intercept=True, weight=weight, verbose=True)
except:
print(i)
raise
vol = np.std(resid, axis=0)
vol.index = a.index = b.columns = stks_to_regress
b.index = [window_edate]
vol.name = a.name = window_edate
beta = pd.concat([beta, b], axis=0)
alpha = pd.concat([alpha, a], axis=1)
sigma = pd.concat([sigma, vol], axis=1)
beta = beta.T
beta = beta.reindex(y.index, axis=1).reindex(y.columns, axis=0)
alpha = alpha.reindex(y.index, axis=1).reindex(y.columns, axis=0)
sigma = sigma.reindex(y.index, axis=1).reindex(y.columns, axis=0)
return beta, alpha, sigma
def REVSU(self):
netprofit = self.totaloperatingrevenueps
# 得到单季度的数据。
sig_season_va = get_signal_season_value(netprofit)
cols = pd.DataFrame([i for i in sig_season_va.columns])
revsu = pd.DataFrame()
rolling_cols = rolling_windows(cols, 6)
for roll in rolling_cols:
res = _calculate_su_simple(sig_season_va[roll])
res = pd.DataFrame(res.values, index=res.index, columns=[roll[-1]])
revsu = pd.concat([revsu, res], axis=1)
revsu.dropna(how='all', axis=0, inplace=True)
revsu = adjust_months(revsu)
revsu = append_df(revsu)
return revsu
def SUE(self):
# 使用原始的财务数据
eps = self.basiceps
# 得到单季度的数据。
sig_season_va = get_signal_season_value(eps)
cols = pd.DataFrame([i for i in sig_season_va.columns])
sue = pd.DataFrame()
rolling_cols = rolling_windows(cols, 6)
for roll in rolling_cols:
res = _calculate_su_simple(sig_season_va[roll])
res = pd.DataFrame(res.values, index=res.index, columns=[roll[-1]])
sue = pd.concat([sue, res], axis=1)
sue.dropna(how='all', axis=0, inplace=True)
sue = adjust_months(sue)
sue = append_df(sue)
return sue
def test__rolling_lstsq(x, y, window, out):
xwins = utils.rolling_windows(x, window)
ywins = utils.rolling_windows(y, window)
assert np.allclose(ols._rolling_lstsq(xwins, ywins), out)
