def find_more_accurate_beta(y, x):
_yvar = med_abs_dev(y)
_xvar = med_abs_dev(x)
_yxrat = _yvar / _xvar
_yxcors = smart_spearman(y, x)
_yxcork = smart_kendall(y, x)
_collects = _yxrat * _yxcors *x
_collectk = _yxrat * _yxcork * x
_errs = y - _collects
_errk = y - _collectk
_excors = smart_kendall(_errs, _collects) # _yxcors -> _excors
_excork = smart_kendall(_errk, _collectk) # _yxcork -> _excork
_excors_ = smart_spearman(_errs, _collects) # _yxcors -> _excors
_excork_ = smart_spearman(_errk, _collectk) # _yxcork -> _excork
if abs(_excork - _excors) > 1e-8:
# adjust 1 (low kendall)
_yxcorn = _yxcors - _excors*(_yxcork-_yxcors)/(_excork-_excors)
_collectn = _yxrat * _yxcorn * x
_errn = y - _collectn
_excorn = smart_kendall(_errn, _collectn)
# adjust 2 (low spearman)
_yxcorn2 = _yxcors - _excors_*(_yxcork-_yxcors)/(_excork_-_excors_)
_collectn2 = _yxrat * _yxcorn2 * x
_errn2 = y - _collectn2
_excorn2 = smart_kendall(_errn2, _collectn2)
if abs(_excorn-_excorn2) > 1e-8:
# adjust 3 (low kendall2)
_yxcorn3 = _yxcorn - _excorn*(_yxcorn-_yxcorn2)/(_excorn-_excorn2)
_collectn3 = _yxrat * _yxcorn3 * x
_errn3 = y - _collectn3
_excorn3 = smart_kendall(_errn3, _collectn3)
# interpolate between adjust 1 and adjust 3
if abs(_excorn-_excorn3) > 1e-8:
_yxcorn4 = _yxcorn - _excorn*(_yxcorn-_yxcorn3)/(_excorn-_excorn3)
_collectn4 = _yxrat * _yxcorn4 * x
_errn4 = y - _collectn4
_excorn4 = smart_kendall(_errn4, _collectn4)
if abs(_excorn) < abs(_excorn3):
_excorn1 = _excorn
_yxcorn1 = _yxcorn
else:
_excorn1 = _excorn3
_yxcorn1 = _yxcorn3
if abs(_excorn1-_excorn4) > 1e-8:
# interpolate between smaller of 1 and 3
_yxcorn5 = _yxcorn4 - _excorn4*(_yxcorn4-_yxcorn1)/(_excorn4-_excorn1)
# _collectn5 = _yxrat * _yxcorn5 * x
# _errn5 = y - _collectn5
# _excorn5 = smart_kendall(_errn5, _collectn5)
return _yxrat * _yxcorn5
else:
return _yxrat * _yxcorn4
else:
return _yxrat * _yxcorn3
else:
return _yxrat * _yxcorn
else:
return _yxrat * _excork
def get_smart_measures():
univ_ib_data = cr_cret.retrieve(univ_ib_eqidx_ext + 'GBM')
univ_ib_cl = univ_ib_data['Close'].values
univ_ib_vl = univ_ib_data['Volatility'].values
del univ_ib_data['Close']
del univ_ib_data['Volatility']
new_col_names = list(univ_ib_data.columns)
univ_ib_data = univ_ib_data.values
num_col = len(new_col_names)
# change data to appropriate lags
for i in range(0, num_col):
univ_ib_data[:, i] = filt.lag(univ_ib_data[:, i], 2)
univ_ib_vl = filt.lag(univ_ib_vl, 2)
univ_ib_cl1 = univ_ib_cl.copy()
univ_ib_cl2 = (univ_ib_cl+filt.lag(univ_ib_cl))/2
# divide by the volatility
univ_ib_cl1 = univ_ib_cl1/univ_ib_vl
univ_ib_cl2 = univ_ib_cl2/univ_ib_vl
for i in range(0, num_col):
univ_ib_data[:, i] = univ_ib_data[:, i]/univ_ib_vl
# standardize them
for i in range(0, num_col):
univ_ib_data[:, i] = univ_ib_data[:, i]/med_abs_dev(univ_ib_data[5220:, i])
univ_ib_cl1 = univ_ib_cl1/med_abs_dev(univ_ib_cl1[5220:])
univ_ib_cl2 = univ_ib_cl2/med_abs_dev(univ_ib_cl2[5220:])
# get the correlations with univ_ib_cl
univ_ib_correl1 = np.zeros(num_col)
univ_ib_correl2 = np.zeros(num_col)
for i in range(0, num_col):
univ_ib_correl1[i] = smart_kendall(univ_ib_data[5220:, i], univ_ib_cl1[5220:])
univ_ib_correl2[i] = smart_kendall(univ_ib_data[5220:, i], univ_ib_cl2[5220:])
univ_ib_correl1a = np.array([(j if 'M_' not in new_col_names[i] else np.nan) for i, j in enumerate(univ_ib_correl1)])
univ_ib_correl2a = np.array([(j if 'M_' not in new_col_names[i] else np.nan) for i, j in enumerate(univ_ib_correl2)])
univ_ib_correl1b = np.array([(j if (('M_' not in new_col_names[i]) and ('W_' not in new_col_names[i])) else np.nan)
for i, j in enumerate(univ_ib_correl1)])
univ_ib_correl2b = np.array([(j if (('M_' not in new_col_names[i]) and ('W_' not in new_col_names[i])) else np.nan)
for i, j in enumerate(univ_ib_correl2)])
for k in range(0, 3):
if k == 0:
univ_ib_correl1_ = univ_ib_correl1
univ_ib_correl2_ = univ_ib_correl2
elif k == 1:
univ_ib_correl1_ = univ_ib_correl1a
univ_ib_correl2_ = univ_ib_correl2a
else:
univ_ib_correl1_ = univ_ib_correl1b
univ_ib_correl2_ = univ_ib_correl2b
_correl = univ_ib_correl1_ + univ_ib_correl2_
_mom_sig = []
_mom_sig_idx = []
_mom_val = []
for i in range(0, 20):
if i == 0:
_idx = argmax_fixed(_correl)
_mom_sig.append(new_col_names[_idx])
_mom_sig_idx.append(_idx)
_mom_val.append(_correl[_idx])
else:
# calculate correlation with the already selected series
_scorrel = np.ones(7200, dtype=bool)
for j1 in _mom_sig_idx:
_scorrel[j1] = False
for j1 in range(0, 7200):
if j1 not in _mom_sig_idx:
for j2 in _mom_sig_idx:
if _scorrel[j1] and (smart_kendall(univ_ib_data[5220:, j1], univ_ib_data[5220:, j2]) > 0.6):
_scorrel[j1] = False
# add constraint to handle the case where _scorrel is True, but the correlation with
_scorrel_idx = np.where(_scorrel)[0]
if _scorrel_idx.shape[0] > 0:
_correl_red = _correl[_scorrel_idx]
_idx = _scorrel_idx[argmax_fixed(_correl_red)]
if _correl[_idx] > 0.015:
_mom_sig.append(new_col_names[_idx])
_mom_sig_idx.append(_idx)
_mom_val.append(_correl[_idx])
else:
break
else:
break
print(i)
print(_mom_sig)
print(_mom_val)
_rev_sig = []
_rev_sig_idx = []
_rev_val = []
for i in range(0, 40):
# i = 0
if i == 0:
_idx = argmin_fixed(_correl)
_rev_sig.append(new_col_names[_idx])
_rev_sig_idx.append(_idx)
_rev_val.append(_correl[_idx])
else:
# calculate correlation with the already selected series
_scorrel = np.ones(7200, dtype=bool)
for j1 in _rev_sig_idx:
_scorrel[j1] = False
for j1 in range(0, 7200):
if j1 not in _rev_sig_idx:
for j2 in _rev_sig_idx:
if _scorrel[j1] and (smart_kendall(univ_ib_data[5220:, j1], univ_ib_data[5220:, j2]) > 0.6):
_scorrel[j1] = False
# add constraint to handle the case where _scorrel is True, but the correlation with
_scorrel_idx = np.where(_scorrel)[0]
if _scorrel_idx.shape[0] > 0:
_correl_red = _correl[_scorrel_idx]
_idx = _scorrel_idx[argmin_fixed(_correl_red)]
if _correl[_idx] < -0.015:
_rev_sig.append(new_col_names[_idx])
_rev_sig_idx.append(_idx)
_rev_val.append(_correl[_idx])
else:
break
else:
break
print(i)
print(_rev_sig)
print(_rev_val)
