start_month = last_date.month + 1
else:
start_year = last_date.year - year_delta + 1
start_month = 1
# 基础设定
history_data = History_Data[asset_list][
str(start_year) + '-' + str(start_month): last_date]
predict_data = Predict_Data[asset_list][
str(start_year) + '-' + str(start_month): last_date]
#cov_mat = history_data[asset_list].cov() * 12.0
cov_mat = pd.ewmcov(history_data, alpha=0.2).iloc[-3:] * 12.0
# print cov_mat
omega = np.matrix(cov_mat.values)
mkt_wgt = Risk_Parity_Weight(cov_mat)
print mkt_wgt
P = np.diag([1] * len(mkt_wgt))
conf_list = list()
for each in asset_list:
conf_temp = ((history_data[each][str(start_year) + '-' + str(start_month):] -
predict_data[each][str(start_year) + '-' + str(start_month):])**2).mean() * 12.0
conf_list.append(conf_temp)
conf_mat = np.matrix(np.diag(conf_list))
Q = np.matrix(Predict_Data[asset_list].loc[next_date])
com_ret, com_cov_mat = Combined_Return_Distribution(
2, cov_mat, tau, mkt_wgt, P, Q, conf_mat)
def Ms_RP(return_frame, switch_map):
temp_columns = list(return_frame.columns)
temp_Ms_list = []
for each in temp_columns:
if switch_map[each]:
temp_std, temp_Coef, temp_transMat, temp_prob_smo = Ms_R(
list(return_frame[each]))
else:
temp_std = np.array([np.std(list(return_frame[each]))] * 2)
temp_Coef = np.array([np.mean(list(return_frame[each]))] * 2)
temp_transMat = np.array([[1, 0], [0, 1]]).reshape(2, 2)
temp_prob_smo = np.array([[0.5] * len(return_frame[each]),
[0.5] * len(return_frame[each])]).T
temp_Ms_list.append(
[temp_std, temp_Coef, temp_transMat, temp_prob_smo])
Ms_frame = pd.DataFrame(temp_Ms_list,
index=temp_columns,
columns=['std', 'Coef', 'transMat', 'prob_smo']).T
temp_cov_list = []
for each_i in temp_columns:
for each_j in temp_columns[temp_columns.index(each_i) + 1:]:
temp_cov_mat = Cross_Cov(return_frame[each_i],
return_frame[each_j],
Ms_frame[each_i]['Coef'],
Ms_frame[each_j]['Coef'],
Ms_frame[each_i]['prob_smo'],
Ms_frame[each_j]['prob_smo'])
temp_cov_list.append(temp_cov_mat)
Tree = Tree_Gen(len(temp_columns))
rp_wgt_list = []
for each_leaf in Tree:
cov_mat_list = []
for i in range(len(temp_columns)):
for j in range(len(temp_columns)):
if i == j:
cov_mat_list.append((Ms_frame[temp_columns[i]]['std'][int(
each_leaf[i])])**2)
else:
if i < j:
location = len(temp_columns) * (i + 1) - sum(
range(i + 2)) - (len(temp_columns) - j)
cov_mat_list.append(
temp_cov_list[location][int(each_leaf[i]),
int(each_leaf[j])])
else:
location = len(temp_columns) * (j + 1) - sum(
range(j + 2)) - (len(temp_columns) - i)
cov_mat_list.append(
temp_cov_list[location][int(each_leaf[j]),
int(each_leaf[i])])
cov_mat = np.array(cov_mat_list).reshape(len(temp_columns),
len(temp_columns))
cov_mat = pd.DataFrame(cov_mat,
columns=temp_columns,
index=temp_columns)
rp_wgt = Risk_Parity_Weight(cov_mat)
rp_wgt_list.append(rp_wgt)
prob_list = []
for each_leaf in Tree:
prob_leaf = 1
for i in range(len(temp_columns)):
stat = int(each_leaf[i])
trans_mat = Ms_frame[temp_columns[i]]['transMat']
temp_prob = Ms_frame[temp_columns[i]]['prob_smo'][
-1, 0] * trans_mat[0, stat] + Ms_frame[
temp_columns[i]]['prob_smo'][-1, 1] * trans_mat[1, stat]
prob_leaf = prob_leaf * temp_prob
prob_list.append(prob_leaf)
'''
filt_prob_list = []
for each_prob in prob_list:
if each_prob == max(prob_list):
filt_prob_list.append(1.0)
else:
filt_prob_list.append(0.0)
prob_list = filt_prob_list
'''
prob_wgt_list = []
for i in range(len(Tree)):
prob_wgt_list.append(rp_wgt_list[i] * prob_list[i])
return sum(prob_wgt_list)
rp_wgt = Ms_RP(data_frame, {'SP500':True, 'Barclays_US_bond':False})
'''
multi_wgt = Ms_Multi(data_frame, {
'SP500': True,
'London_gold': True,
'Barclays_US_bond': False
}, 2)
mu_wgt = multi_wgt["mu_wgt"]
rp_wgt = multi_wgt["rp_wgt"]
print mu_wgt
print rp_wgt
#data_frame = data[data.index[each-120]:data.index[each]]
rp_wgt_bm = Risk_Parity_Weight(data_frame.cov()).round(3)
mu_wgt_bm = Max_Utility_Weight_new(pd.DataFrame(data_frame.mean()),
data_frame.cov(), 2,
[(0.0, None)] * 3).round(3)
'''
mu_wgt = Ms_MU(data_frame, {'SP500':True, 'London_gold':True, 'Barclays_US_bond':False})
rp_wgt = Ms_RP(data_frame, {'SP500':True, 'London_gold':True, 'Barclays_US_bond':False})
#data_frame = data[data.index[each-120]:data.index[each]]
rp_wgt_bm = Risk_Parity_Weight(data_frame.cov())
mu_wgt_bm = Max_Utility_Weight_new(pd.DataFrame(data_frame.mean()), data_frame.cov(), 2, [(0.0,None)]*3)
'''
mu_ms_return = np.sum(mu_wgt * data.loc[data.index[each + 1]])
mu_bm_return = np.sum(mu_wgt_bm * data.loc[data.index[each + 1]])
rp_ms_return = np.sum(rp_wgt * data.loc[data.index[each + 1]])
rp_bm_return = np.sum(rp_wgt_bm * data.loc[data.index[each + 1]])
def Ms_Multi(return_frame, switch_map, lam):
temp_columns = list(return_frame.columns)
temp_Ms_list = []
for each in temp_columns:
if switch_map[each] != 1:
temp_std, temp_Coef, temp_transMat, temp_prob_smo = Ms_R(
list(return_frame[each]), switch_map[each])
#print temp_Coef
#print temp_std
else:
temp_std = np.array([np.std(list(return_frame[each]))] * 2)
temp_Coef = np.array([np.mean(list(return_frame[each]))] * 2)
temp_transMat = np.array([[1, 0], [0, 1]]).reshape(2, 2)
temp_prob_smo = np.array([[0.5] * len(return_frame[each]),
[0.5] * len(return_frame[each])]).T
temp_Ms_list.append(
[temp_std, temp_Coef, temp_transMat, temp_prob_smo])
Ms_frame = pd.DataFrame(temp_Ms_list,
index=temp_columns,
columns=['std', 'Coef', 'transMat', 'prob_smo']).T
temp_cov_list = []
for each_i in temp_columns:
for each_j in temp_columns[temp_columns.index(each_i) + 1:]:
temp_cov_mat = Cross_Cov(return_frame[each_i],
return_frame[each_j],
Ms_frame[each_i]['Coef'],
Ms_frame[each_j]['Coef'],
Ms_frame[each_i]['prob_smo'],
Ms_frame[each_j]['prob_smo'])
temp_cov_list.append(temp_cov_mat)
#print temp_cov_list
Tree = Tree_Gen(switch_map, temp_columns)
rp_wgt_list = []
mu_wgt_list = []
for each_leaf in Tree:
cov_mat_temp = []
exp_ret_temp = []
for i in range(len(temp_columns)):
for j in range(len(temp_columns)):
if i == j:
cov_mat_temp.append((Ms_frame[temp_columns[i]]['std'][int(
each_leaf[i])])**2)
exp_ret_temp.append(Ms_frame[temp_columns[i]]['Coef'][int(
each_leaf[i])])
else:
if i < j:
location = len(temp_columns) * (i + 1) - sum(
range(i + 2)) - (len(temp_columns) - j)
cov_mat_temp.append(
temp_cov_list[location][int(each_leaf[i]),
int(each_leaf[j])])
else:
location = len(temp_columns) * (j + 1) - sum(
range(j + 2)) - (len(temp_columns) - i)
cov_mat_temp.append(
temp_cov_list[location][int(each_leaf[j]),
int(each_leaf[i])])
exp_ret = pd.DataFrame(exp_ret_temp, index=temp_columns)
cov_mat = np.array(cov_mat_temp).reshape(len(temp_columns),
len(temp_columns))
cov_mat = pd.DataFrame(cov_mat,
columns=temp_columns,
index=temp_columns)
rp_wgt = Risk_Parity_Weight(cov_mat)
mu_wgt = Max_Utility_Weight(exp_ret, cov_mat, lam,
[(0.0, None)] * len(temp_columns))
#print mu_wgt
#print "----"
rp_wgt_list.append(rp_wgt)
mu_wgt_list.append(mu_wgt)
prob_list = []
for each_leaf in Tree:
prob_leaf = 1
for i in range(len(temp_columns)):
stat = int(each_leaf[i])
trans_mat = Ms_frame[temp_columns[i]]['transMat']
temp_prob = sum(Ms_frame[temp_columns[i]]['prob_smo'][-1, :] *
trans_mat[:, stat])
#temp_prob = Ms_frame[temp_columns[i]]['prob_smo'][-1,0]*trans_mat[0,stat] + Ms_frame[temp_columns[i]]['prob_smo'][-1,1]*trans_mat[1,stat]
prob_leaf = prob_leaf * temp_prob
prob_list.append(prob_leaf)
#print prob_list
'''
filt_prob_list = []
for each_prob in prob_list:
if each_prob == max(prob_list):
filt_prob_list.append(1.0)
else:
filt_prob_list.append(0.0)
prob_list = filt_prob_list
'''
prob_rp_wgt_list = []
prob_mu_wgt_list = []
prob_rp_list = []
prob_mu_list = []
for i in range(len(Tree)):
if any(np.isnan(rp_wgt_list[i])):
pass
else:
prob_rp_wgt_list.append(rp_wgt_list[i] * prob_list[i])
prob_rp_list.append(prob_list[i])
if any(np.isnan(mu_wgt_list[i])):
pass
else:
prob_mu_wgt_list.append(mu_wgt_list[i] * prob_list[i])
prob_mu_list.append(prob_list[i])
#print np.sum(prob_mu_wgt_list)
return {
"rp_wgt": (sum(prob_rp_wgt_list) / sum(prob_rp_list)).round(3),
"mu_wgt": (sum(prob_mu_wgt_list) / sum(prob_mu_list)).round(3)
}
def Backtest_CPPI_BL_step(History_Data, Predict_Data, History_Data_D,
asset_list, risk_list, bnds, asset_level_1,
asset_level_2, year_delta, portfolio_name,
money_weight, up_per, target_ret, multiplier,
max_risk_weight):
tau = 1.0
if portfolio_name == "wenjian":
lam = 2.3 #进取-1.9 平衡-2.0 稳健-2.3
elif portfolio_name == "pingheng":
lam = 1.9
elif portfolio_name == "jinqu":
lam = 1.7
else:
raise Exception("Wrong portfolio_name!")
pct_list = []
weight_list = []
date_list = []
asset_drawdown = pd.Series([0.0] * len(asset_list), index=asset_list)
asset_position = pd.Series([1.0] * len(asset_list), index=asset_list)
for each_date in Predict_Data.index[60:-1]:
last_date = History_Data.index[list(
Predict_Data.index).index(each_date) - 1] # 当前月份日期
next_date = each_date # 下一月份日期
if last_date.month <= 11:
start_year = last_date.year - year_delta
start_month = last_date.month + 1
else:
start_year = last_date.year - year_delta + 1
start_month = 1
# 基础设定
history_data = History_Data[str(start_year) + '-' +
str(start_month):last_date]
history_data_d = History_Data_D[str(start_year) + '-' +
str(start_month):last_date]
predict_data = Predict_Data[str(start_year) + '-' +
str(start_month):last_date]
cov_mat = history_data_d[risk_list].cov() * (len(history_data_d) /
year_delta)
#cov_mat = history_data[asset_list].cov() * 12.0
#for each_asset in risk_list:
for each_asset in risk_list:
temp_drawdown = (asset_drawdown[each_asset] +
1.0) * (history_data[each_asset][-1] + 1.0) - 1
if temp_drawdown >= 0:
temp_drawdown = 0
else:
pass
asset_drawdown[each_asset] = temp_drawdown
# print cov_mat
omega = np.matrix(cov_mat.values)
mkt_wgt = Risk_Parity_Weight(cov_mat)
#print mkt_wgt
P = np.diag([1] * len(mkt_wgt))
conf_list = list()
for each in risk_list:
conf_temp = ((
history_data[each][str(start_year) + '-' + str(start_month):] -
predict_data[each][str(start_year) + '-' + str(start_month):])
**2).mean() * 12.0
conf_list.append(conf_temp)
conf_mat = np.matrix(np.diag(conf_list))
Q = np.matrix(Predict_Data[risk_list].loc[next_date])
com_ret, com_cov_mat = Combined_Return_Distribution(
2, cov_mat, tau, mkt_wgt, P, Q, conf_mat)
#print com_ret
weight_bl = Max_Utility_Weight(com_ret, com_cov_mat, lam, bnds)
#print weight_bl
rf_ret = history_data[list(set(asset_list) -
set(risk_list))[0]].mean() * 12
if len(pct_list) == 0:
current_nv = 1
nv_threshold = 1
else:
current_nv = (np.array(pct_list) + 1).prod()
#print "current_nv"
#print current_nv
if current_nv > (1 + up_per) * nv_threshold:
nv_threshold = (1 + up_per) * nv_threshold
else:
pass
cushion_per = Cushion_Cal(nv=current_nv,
nv_threshold=nv_threshold,
rf_ret=rf_ret,
target_ret=target_ret)
#print "cushion"
#print cushion_per
para_m = 0.01 / (-VaR_Cal(0.99, [0] * len(risk_list),
history_data[risk_list].cov() * 12.0, 1 / 12,
weight_bl[risk_list]))
#print "para_m"
#print para_m
#print "--------------"
risk_weight = min(cushion_per * multiplier, max_risk_weight)
weight_risk = weight_bl[risk_list] * risk_weight
#weight_risk = weight_bl[risk_list]
#print weight_risk
weight_bl[risk_list] = weight_risk
#weight_bl[risk_list] = [0.0]*len(risk_list)
weight_bl[list(set(asset_list) -
set(risk_list))[0]] = money_weight - sum(
weight_bl[risk_list])
#weight_bl[list(set(asset_list)-set(risk_list))[0]] = 0.0
for each_asset in asset_list:
if asset_position[each_asset] == 1:
if (asset_drawdown[each_asset] <= asset_level_1[each_asset]
) and (asset_drawdown[each_asset] >
asset_level_2[each_asset]):
asset_position[each_asset] = 0.5
elif asset_drawdown[each_asset] <= asset_level_2[each_asset]:
asset_position[each_asset] = 0.0
else:
pass
elif asset_position[each_asset] == 0.5:
if asset_position[each_asset] <= asset_level_2[each_asset]:
asset_position[each_asset] = 0.0
elif (predict_data[each_asset][-1] >
0) and (history_data[each_asset][-1] > 0):
asset_position[each_asset] = 1.0
asset_drawdown[each_asset] = 0.0
else:
pass
elif asset_position[each_asset] == 0.0:
if (predict_data[each_asset][-1] >
0) and (history_data[each_asset][-1] > 0):
asset_position[each_asset] = 0.5
asset_drawdown[each_asset] = 0.0
else:
pass
weight_bl = (weight_bl * asset_position).round(2)
#print weight_bl
#print next_date
#print sum(weight_bl)
port_ret = sum(weight_bl * History_Data[asset_list].loc[next_date]) + (
1.0 - sum(weight_bl)) * History_Data["money"].loc[next_date]
#print sum(weight_bl*History_Data[asset_list].loc[next_date])*money_weight + money_weight*History_Data["money"].loc[next_date]
pct_list.append(port_ret)
weight_list.append(list(weight_bl))
date_list.append(next_date)
#pd.Series(np.array(pct_list), index=date_list).to_csv("/Users/WangBin-Mac/FOF/Asset Allocation/backtest_%s.csv"%portfolio_name)
#pd.DataFrame(np.array(weight_list), index=date_list, columns=asset_list).to_excel("/Users/WangBin-Mac/FOF/Asset Allocation/backtest_%s_weight.xlsx"%portfolio_name)
return pd.Series(np.array(pct_list), index=date_list)
rho_matrix_dcc[i][j] = temp_rho
cov_matrix_garch_dcc = np.matrix(
np.diag(std_list_garch)) * np.matrix(rho_matrix_dcc) * np.matrix(
np.diag(std_list_garch))
cov_matrix_garch = np.matrix(np.diag(std_list_garch)) * np.matrix(
return_frame.corr()) * np.matrix(np.diag(std_list_garch))
cov_matrix_all_garch_dcc = return_frame_all.cov()
cov_matrix_all_garch_dcc.loc[asset_col,
asset_col] = cov_matrix_garch_dcc
cov_matrix_all_garch = return_frame_all.cov()
cov_matrix_all_garch.loc[asset_col, asset_col] = cov_matrix_garch
rw = Risk_Parity_Weight(return_frame_all.cov()).round(3)
rw_gd = Risk_Parity_Weight(cov_matrix_all_garch_dcc).round(3)
rw_g = Risk_Parity_Weight(cov_matrix_all_garch).round(3)
'''
rw = Max_Utility_Weight_new(pd.DataFrame(return_frame_all.mean()), return_frame_all.cov(), 5, [(0.0,None)]*len(asset_col_all)).round(3)
rw_gd = Max_Utility_Weight_new(pd.DataFrame(return_frame_all.mean()), cov_matrix_all_garch_dcc, 5, [(0.0,None)]*len(asset_col_all)).round(3)
rw_g = Max_Utility_Weight_new(pd.DataFrame(return_frame_all.mean()), cov_matrix_all_garch, 5, [(0.0,None)]*len(asset_col_all)).round(3)
'''
mw = Max_Utility_Weight_new(pd.DataFrame(return_frame_all.mean()),
return_frame_all.cov(), 2, [(0.0, None)] *
len(asset_col_all)).round(3)
mw_gd = Max_Utility_Weight_new(
pd.DataFrame(return_frame_all.mean()), cov_matrix_all_garch_dcc, 2,
[(0.0, None)] * len(asset_col_all)).round(3)
mw_g = Max_Utility_Weight_new(pd.DataFrame(return_frame_all.mean()),
cov_matrix_all_garch, 2, [(0.0, None)] *