def computeClipped(self, train):
'''
compute the clipped estimate covariance
Parameters
-----
train: ndarray
training data
Return
-----
empirical estimate covariance
'''
clipped = pyRMT.clipped(train, return_covariance=True)
return clipped * self._standarized_factor
def getClippedCovariance(self):
return pyRMT.clipped(np.transpose(self.stockReturns),
return_covariance=True)
local_df = df[start_date_i: start_portfolio_i] local_df = local_df.replace(0, np.nan).dropna(axis=1) columns = list(local_df.columns) drop = list(set(cryptos) - set(columns)) local_df = returns[start_date_i: start_portfolio_i] local_df = local_df.iloc[1:] # remove first row local_df = local_df.drop(drop, axis=1) columns = list(local_df.columns) mu = expected_returns.mean_historical_return(local_df,returns_data=True,frequency=365) rf = 0.00 #risk free rate - 0.02 (in this case = 0) mu_excess = mu - rf Sigma = risk_models.sample_cov(local_df,returns_data=True,frequency=365) Sigma = rmt.clipped(Sigma,return_covariance=True) Sigma = pd.DataFrame(data=Sigma,index=columns,columns=columns) ones = [1] * (len(columns)) arr_ones = np.array(ones) arr_ones = pd.DataFrame([arr_ones], columns=columns) Sigma_inv = pd.DataFrame(np.linalg.pinv(Sigma.values), Sigma.columns, Sigma.index) x = arr_ones.dot(Sigma_inv) y = x.dot(arr_ones.T) y = y ** -1 y=y[0] weights = Sigma_inv.multiply(y[0]) print(weights) weights = weights.dot(arr_ones.T)
def __init__(self, _ret, has_mean):
self._ret = _ret
if has_mean:
self._ret -= self._ret.mean(axis=0)
self._clip_cov = clipped(self._ret, return_covariance=True)