def generate_pairs(self, clustering_results: Dict[int,
Tuple[Tuple[Tickers]]],
hurst_exp_threshold: float, current_window: Window):
# run cointegration_analysis on all poss combinations of pairs within the same cluster
current_cointegrated_pairs = []
n_cointegrated = 0
tickers_per_cluster = [i for i in clustering_results.values()]
for cluster in tickers_per_cluster:
for pair in itertools.combinations(list(cluster), 2):
t1 = current_window.get_data(tickers=[pair[0]],
features=[Features.CLOSE])
t2 = current_window.get_data(tickers=[pair[1]],
features=[Features.CLOSE])
try:
# sometimes there are no price data, in which case, skip
residuals, beta, reg_output = self.__logged_lin_reg(t1, t2)
except ValueError:
continue
adf_test_statistic, adf_critical_values = self.__adf(
residuals.flatten())
hl_test = self.__hl(residuals)
he_test = self.__hurst_exponent_test(residuals, current_window)
ou_mean, ou_std, ou_diffusion_v, \
recent_dev, recent_dev_scaled = self.__ou_params(residuals)
ols_stdzed_residuals = (residuals - ou_mean) / ou_std
is_cointegrated = self.__acceptance_rule(
adf_test_statistic,
adf_critical_values,
self.adf_confidence_level,
hl_test,
self.max_mean_rev_time,
he_test,
hurst_exp_threshold,
ols_stdzed_residuals,
at_least=int(current_window.window_length.days / 6))
if is_cointegrated:
#a = pd.concat([t1, t2], axis=1).iplot(asFigure=True)
#b = pd.concat([np.log(t1), np.log(t2)], axis=1).iplot(asFigure=True)
#a.show()
#b.show()
n_cointegrated += 1
t1_most_recent = float(t1.iloc[-1, :])
t2_most_recent = float(t2.iloc[-1, :])
hedge_ratio = beta * t1_most_recent / t2_most_recent
scaled_beta = hedge_ratio / (hedge_ratio - 1)
recent_dev_scaled_hist = [recent_dev_scaled]
cointegration_rank = self.__score_coint(
adf_test_statistic, self.adf_confidence_level,
adf_critical_values, he_test, hurst_exp_threshold, 10)
#a = pd.DataFrame(ols_stdzed_residuals).iplot(asFigure=True)
#a.show()
position = Position(pair[0], pair[1])
current_cointegrated_pairs.append(
CointegratedPair(pair, reg_output, scaled_beta,
hl_test, ou_mean, ou_std,
ou_diffusion_v, recent_dev,
recent_dev_scaled,
recent_dev_scaled_hist,
cointegration_rank,
ols_stdzed_residuals, position))
if n_cointegrated == self.target_number_of_coint_pairs:
current_cointegrated_pairs = sorted(
current_cointegrated_pairs,
key=lambda coint_pair: coint_pair.
cointegration_rank,
reverse=True)
self.previous_cointegrated_pairs = current_cointegrated_pairs
return current_cointegrated_pairs
self.previous_cointegrated_pairs = current_cointegrated_pairs
return current_cointegrated_pairs
def generate_pairs(self, clustering_results: Dict[int,
Tuple[Tuple[Tickers]]],
hurst_exp_threshold: float, current_window: Window):
# run cointegration_analysis on all poss combinations of pairs
current_cointegrated_pairs = []
n_cointegrated = 0
list_of_lists = [i for i in clustering_results.values()]
flattened = [pair for x in list_of_lists for pair in x]
sorted_cluster_results = sorted(flattened, key=lambda x: x[0].value)
for pair in sorted_cluster_results:
t1 = current_window.get_data(universe=Universes.SNP,
tickers=[pair[0]],
features=[Features.CLOSE])
t2 = current_window.get_data(universe=Universes.SNP,
tickers=[pair[1]],
features=[Features.CLOSE])
try:
# sometimes there are no price data
residuals, beta, reg_output = self.__logged_lin_reg(t1, t2)
except:
continue
# for some reason residuals is a (60,1) array not (60,) array when i run the code so have changed input to residuals.flatten
adf_test_statistic, adf_critical_values = self.__adf(
residuals.flatten())
hl_test = self.__hl(residuals)
he_test = self.__hurst_exponent_test(residuals, current_window)
is_cointegrated = self.__acceptance_rule(
adf_test_statistic, adf_critical_values,
self.adf_confidence_level, hl_test, self.max_mean_rev_time,
he_test, hurst_exp_threshold)
target_coint = 300
if is_cointegrated:
n_cointegrated += 1
print(n_cointegrated, " / ", target_coint, "cointegrated")
r_x = self.__log_returner(t1)
mu_x_ann = float(250 * np.mean(r_x))
sigma_x_ann = float(250**0.5 * np.std(r_x))
ou_mean, ou_std, ou_diffusion_v, recent_dev, recent_dev_scaled = self.__ou_params(
residuals)
scaled_beta = beta / (beta - 1)
recent_dev_scaled_hist = [recent_dev_scaled]
cointegration_rank = self.__score_coint(
adf_test_statistic, self.adf_confidence_level,
adf_critical_values)
current_cointegrated_pairs.append(
CointegratedPair(pair, mu_x_ann, sigma_x_ann, reg_output,
scaled_beta, hl_test, ou_mean, ou_std,
ou_diffusion_v, recent_dev,
recent_dev_scaled, recent_dev_scaled_hist,
cointegration_rank))
if n_cointegrated == target_coint:
# logic to be fixed and made more efficient by: 1) having proper
# clustering algorithm; 2) not running clustering and cointegration
# everyday 3) taking best 10 pairs according to some score
current_cointegrated_pairs = sorted(
current_cointegrated_pairs,
key=lambda coint_pair: coint_pair.cointegration_rank,
reverse=True)
self.previous_cointegrated_pairs = current_cointegrated_pairs
return current_cointegrated_pairs
self.previous_cointegrated_pairs = current_cointegrated_pairs
return current_cointegrated_pairs