def compute_covariance_matrix(self, bool_print=False):
# compute variance-covariance matrix by pairwise covariances
scale = 252 # annualised
size = len(self.rics)
mtx_covar = np.zeros([size, size])
mtx_correl = np.zeros([size, size])
vec_returns = np.zeros([size, 1])
vec_volatilities = np.zeros([size, 1])
returns = []
for i in range(size):
ric1 = self.rics[i]
temp_ret = []
for j in range(i + 1):
ric2 = self.rics[j]
ret1, ret2, t = stream_functions.synchronise_timeseries(
ric1, ric2)
returns = [ret1, ret2]
# covariances
temp_mtx = np.cov(returns)
temp_covar = scale * temp_mtx[0][1]
temp_covar = np.round(temp_covar, self.nb_decimals)
mtx_covar[i][j] = temp_covar
mtx_covar[j][i] = temp_covar
# correlations
temp_mtx = np.corrcoef(returns)
temp_correl = temp_mtx[0][1]
temp_correl = np.round(temp_correl, self.nb_decimals)
mtx_correl[i][j] = temp_correl
mtx_correl[j][i] = temp_correl
if j == 0:
temp_ret = ret1
# returns
temp_mean = np.round(scale * np.mean(temp_ret), self.nb_decimals)
vec_returns[i] = temp_mean
# volatilities
temp_volatility = np.round(
np.sqrt(scale) * np.std(temp_ret), self.nb_decimals)
vec_volatilities[i] = temp_volatility
self.covariance_matrix = mtx_covar
self.correlation_matrix = mtx_correl
self.returns = vec_returns
self.volatilities = vec_volatilities
if bool_print:
print('-----')
print('Portfolio Manager details:')
print('Securities:')
print(self.rics)
print('Returns (annualised):')
print(self.returns)
print('Volatilities (annualised):')
print(self.volatilities)
print('Variance-covariance matrix (annualised):')
print(self.covariance_matrix)
print('Correlation matrix:')
print(self.correlation_matrix)
import bollinger_bands
importlib.reload(bollinger_bands)
# inputs
backtest = bollinger_bands.backtest()
backtest.ric_long = 'SGREN.MC'
backtest.ric_short = 'VWS.CO'
backtest.rolling_days = 20
backtest.level_1 = 1.
backtest.level_2 = 2.
backtest.data_cut = 0.7
backtest.data_type = 'in-sample' # in-sample out-of-sample
# load data
_, _, t = stream_functions.synchronise_timeseries(backtest.ric_long,
backtest.ric_short)
cut = int(backtest.data_cut * t.shape[0])
if backtest.data_type == 'in-sample':
df1 = t.head(cut)
elif backtest.data_type == 'out-of-sample':
df1 = t.tail(t.shape[0] - cut)
df1 = df1.reset_index(drop=True)
# spread at current close
df1['spread'] = df1['price_1'] / df1['price_2']
base = df1['spread'][0]
df1['spread'] = df1['spread'] / base
# spread at previous close
df1['spread_previous'] = df1['price_1_previous'] / df1['price_2_previous']
df1['spread_previous'] = df1['spread_previous'] / base
def load_timeseries(self):
self.returns_benchmark, self.returns_ric, self.dataframe\
= stream_functions.synchronise_timeseries(self.benchmark, self.ric)
def load_timeseries(self):
# load timeseries and synchronise them
self.x, self.y, self.t = stream_functions.synchronise_timeseries(
self.ric, self.benchmark)
# x, x_str, t = stream_functions.load_timeseries(ric)
# returns.append(x)
# mtx_covar = np.cov(returns) # cov = covariance
# mtx_correl = np.corrcoef(returns) # corrcoef = correlation
# # this did not work, we need to synchronise the timeseries before computing the covar
# compute variance-covariance matrix by pairwise covariances
size = len(rics)
mtx_covar = np.zeros([size, size])
mtx_correl = np.zeros([size, size])
returns = []
for i in range(size):
ric1 = rics[i]
for j in range(size):
ric2 = rics[j]
ret1, ret2, t = stream_functions.synchronise_timeseries(ric1, ric2)
returns = [ret1, ret2]
# covariances
temp_mtx = np.cov(returns)
temp_covar = scale * temp_mtx[0][1]
mtx_covar[i][j] = np.round(temp_covar, nb_decimals)
# correlations
temp_mtx = np.corrcoef(returns)
temp_correl = temp_mtx[0][1]
mtx_correl[i][j] = np.round(temp_correl, nb_decimals)
# # compute eigenvalues and eigenvectors
# eigenvalues, eigenvectors = LA.eig(mtx_covar)
# compute eigenvalues and eigenvectors for symmetric matrices
eigenvalues, eigenvectors = LA.eigh(mtx_covar)