def showtable5():
df1 = beta_equal_annual.mean()
df2 = pd.Series(nwttest_1samp(beta_equal_annual, 0)[0], index=df1.index)
df3 = pd.Series(nwttest_1samp(beta_equal_annual, 0)[1], index=df1.index)
table = pd.concat([df1, df2, df3], axis=1).T
del table['year']
table.index = ['Average', 't-statistic', 'p-value']
table = table.applymap(lambda x: round(x, 2))
print(table)
def trasecfee(feerate):
'''
:param feerate: 交易费率
:return:不输出只打印结果
'''
for i in range(len(factorlist)):
temp = pd.read_csv('../Database/returnseries/12/' + factorlist[i])
long_short = np.array(temp['long-short'])
afterLS = list(long_short - feerate*2)
length = 12
name = factorlist[i]
T_value = []
Mean = []
p_value = []
sharpratio = []
Std = []
TO = [afterLS]
for l in TO:
t_test = nwttest_1samp(l, 0, L=1)
mean = np.average(l) * 12- rf12.mean().tolist()[0] * 12 / 100
STD = np.std(l) * np.sqrt(12)
sharp = (mean) / STD
T_value.append(t_test.statistic)
p_value.append(t_test.pvalue)
Mean.append(mean)
Std.append(STD)
sharpratio.append(sharp)
print(name, 'long-short')
print('mean', Mean[0] / 12)
print('t-statistic', '('+str(round(T_value[0],4))+')')
ff3 = pd.read_csv('../Database/ff3.csv')
ff5 = pd.read_csv('../Database/ff5.csv')
A = pd.DataFrame(afterLS, columns=['long-short'])
M = pd.concat([A], axis=1)
alpha3 = []
t3 = []
t5 = []
alpha5 = []
for i in range(1):
X1 = ff3.iloc[length:, 1:]
X2 = ff5.iloc[length:, 1:]
Y = M.iloc[:-2, i]
Y.index = X1.index
Y = Y - rf12.RF[:-1] / 100
used1 = {'X': X1, 'Y': Y}
reg = smf.ols(formula='Y~1+X', data=used1).fit(cov_type='HAC', cov_kwds={'maxlags': 1})
t3.append(reg.tvalues[0])
alpha3.append(reg.params[0] * 12)
used2 = {'X': X2, 'Y': Y}
reg = smf.ols(formula='Y~1+X', data=used2).fit(cov_type='HAC', cov_kwds={'maxlags': 1})
t5.append(reg.tvalues[0])
alpha5.append(reg.params[0] * 12)
print('alpha-FF3', alpha3[0]/12)
print('t-statistic', '('+str(round(t3[0],4))+')')
print('alpha-FF5', alpha5[0]/12,)
print('t-statistic','('+str(round(t5[0],4))+')')
print('sharpe', sharpratio[0])
gc.collect()
print('*'*30)#分隔开不同收益序列
def returnseriestest(length):
path = r'..\DataBase\returnseries' + '\\' + str(length)
file = glob.glob(os.path.join(path, "*.csv"))
ols = pd.read_csv(path + '\\OLS' + str(length) + '.csv')
dfn = pd.read_csv(path + '\\DFN' + str(length) + '.csv')
k = [] # 每个算法一个df
for i in range(len(file)):
k.append(pd.read_csv(file[i]))
#OLS与其他算法区别
for i in range(len(k)):
t = []
t1 = nwttest_1samp(k[i].iloc[:, 1] - ols['long-short'], 0)
t.append(t1.statistic)
t2 = nwttest_1samp(k[i].iloc[:, 2] - ols['long'], 0)
t.append(t2.statistic)
t3 = nwttest_1samp(k[i].iloc[:, 3] - ols['short'], 0)
t.append(t3.statistic)
print('ols-' + file[i][27:-4], t)
#DFN与其他算法区别
for i in range(len(k)):
t = []
t1 = nwttest_1samp(-k[i].iloc[:, 1] + dfn['long-short'], 0)
t.append(t1.statistic)
t2 = nwttest_1samp(-k[i].iloc[:, 2] + dfn['long'], 0)
t.append(t2.statistic)
t3 = nwttest_1samp(-k[i].iloc[:, 3] + dfn['short'], 0)
t.append(t3.statistic)
print('dfn-' + file[i][27:-4], t)
return
def showtable8():
table = pd.DataFrame()
datalist = [beta_equal_annual, mktcap_equal_annual, bm_equal_annual]
namelist = ['beta', 'MktCap', 'BM']
for i in range(3):
df1 = datalist[i].mean()
df2 = pd.Series(nwttest_1samp(datalist[i], 0)[0], index=df1.index)
x = pd.concat([df1, df2], axis=1).T
del x['year']
x.columns = [1, 2, 3, 4, 5, 6, 7, '7-1']
x.index = [namelist[i], ' ']
table = pd.concat([table, x])
table = table.applymap(lambda x: round(x, 2))
print(table)
def beta_mktcap_excess(data):
#df = pd.DataFrame(columns = data.columns[2:])
df = pd.DataFrame()
for i in data['X2_group'][:5]:
X = pd.DataFrame(columns=data.columns[2:])
x = pd.DataFrame(columns=data.columns)
for year in range(1988, 2012):
temp = data[(data['year'] == year) & (data['X2_group'] == i)]
x = x.append(temp, ignore_index=True)
x.index = x['year']
x = x[x.columns[2:]]
x = x.astype(np.float64)
X = X.append(x.mean(), ignore_index=True)
tv1 = pd.Series(nwttest_1samp(x, 0)[0], index=x.columns)
X = X.append(tv1, ignore_index=True)
#接下来回归FFC_alpha,用到前面的四因子表格ff
temp_alpha = []
temp_tv = []
for j in x.columns:
temp_data = pd.concat([x[j], ff], axis=1)
model = smf.ols(str(j) + '~mktrf+smb+hml+umd',
temp_data).fit(cov_type='HAC',
cov_kwds={'maxlags': 6})
temp_alpha.append(model.params[0])
temp_tv.append(model.tvalues[0])
temp_alpha = pd.Series(temp_alpha, index=x.columns)
temp_tv = pd.Series(temp_tv, index=x.columns)
X = X.append(temp_alpha, ignore_index=True)
X = X.append(temp_tv, ignore_index=True)
X.index = [[i, i, i, i],
[
'Excess_return', 't_value_er', 'FFC_alpha',
't_value_alpha'
]]
X.index.name = ['X2_group', 'Coefficient']
df = pd.concat([df, X])
return df
def showtable7():
data = all_beta_group.copy()
col = [1, 2, 3, 4, 5, 6, 7]
df1 = data.groupby('group')['beta'].mean()
df2 = data.groupby('group')['mktcap'].mean()
df3 = data.groupby('group')['bm'].mean()
df1.index = col
df2.index = col
df3.index = col
X = []
for i in ['beta', 'mktcap', 'bm']:
temp = data.groupby(['year', 'group'])[i].mean().reset_index()
temp = pd.pivot_table(temp, index='year',
columns='group')[i].reset_index()
name = temp.columns
temp['diff'] = temp[name[-1]] - temp[name[1]]
x = nwttest_1samp(temp['diff'], 0)[0]
X.append(x)
X = pd.Series(X, name='7-1 t-statistic', index=['beta', 'MktCap', 'BM'])
table = pd.concat([df1, df2, df3], axis=1).T
table.index = ['beta', 'MktCap', 'BM']
table['7-1'] = table[7] - table[1]
table['7-1 t-statistic'] = X
print(table)
def FC(length,rf, timeseries, lenn=96, na='FC'):
#length为滑动窗口长度：取值{3,12,24,36}
#na为输出文件名称
#rf为无风险利率，取值与length对应{rf3,rf12,rf24,rf36}
Long_Short = []
Long = []
Short = []
for i in range(len(timeseries) - length):
print(i)
FINALm = pd.concat(timeseries[i:i + (length +1)], axis=0)
FINALm = FINALm.fillna(0)
FINAL_X = FINALm.iloc[:, :-2]
FINAL_x = scale(FINAL_X)
final = pd.concat(timeseries[i:i + length], axis=0)
x_train = FINAL_x[:len(final)]
x_test = FINAL_x[len(final):]
y_train = final.iloc[:, -1]
test = timeseries[i + length]
clf = LinearRegression()
k = []
for i in range(lenn):
x = x_train[:, i].reshape(-1, 1)
clf.fit(x, y_train)
k.append(clf.coef_[0])
PREDICTION = []
for i in range(len(x_test)):
test0 = np.array(x_test[i])
y = 0
for j in range(lenn):
y = y + test0[j] * k[j]
PREDICTION.append(y)
y_test = test.iloc[:, -1]
# 构建投资组合
r_predict = pd.DataFrame(PREDICTION, columns=['predict'])
r_ture = pd.DataFrame(y_test)
r_ture.columns = ['ture']
r_ture.index = r_predict.index
FINAL = pd.concat([r_predict, r_ture], axis=1)
FINAL_sort = FINAL.sort_values(by='predict', axis=0)
r_final = np.array(FINAL_sort['ture'])
m = int(len(r_final) * 0.1) + 1
r_final = r_final.tolist()
long = r_final[-m:]
short = r_final[:m]
r_end = (np.sum(long) - np.sum(short)) / m
Long_Short.append(r_end)
Long.append(np.average(long))
Short.append(np.average(short))
T_value = []
Mean = []
p_value = []
sharpratio = []
Std = []
TO = [Long_Short, Long, Short]
for l in TO:
t_test = nwttest_1samp(l, 0)
mean = np.average(l) * 12
STD = np.std(l) * np.sqrt(12)
sharp = (mean - rf.mean().tolist()[0] * 12 / 100) / STD
T_value.append(t_test.statistic)
p_value.append(t_test.pvalue)
Mean.append(mean)
Std.append(STD)
sharpratio.append(sharp)
name = na
length = length
print(name, 'long-short', 'long', 'short')
print('mean', Mean[0] / 12, Mean[1] / 12, Mean[2] / 12)
print('t-statistic', '(' + str(round(T_value[0], 4)) + ')', '(' + str(round(T_value[1], 4)) + ')',
'(' + str(round(T_value[2], 4)) + ')')
A = pd.DataFrame(Long_Short, columns=['long-short'])
B = pd.DataFrame(Long, columns=['long'])
C = pd.DataFrame(Short, columns=['short'])
M = pd.concat([A, B, C], axis=1)
M.to_csv('..\output\\'+name + '.csv')
ff3 = pd.read_csv('..\DataBase\\ff3.csv')
ff5 = pd.read_csv('..\DataBase\\ff5.csv')
alpha3 = []
t3 = []
t5 = []
alpha5 = []
for i in range(3):
X1 = ff3.iloc[length:, 1:]
X2 = ff5.iloc[length:, 1:]
Y = M.iloc[:-2, i]
Y.index = X1.index
Y = Y - rf.RF[:-1] / 100
x1 = sm.add_constant(X1)
reg = sm.OLS(Y, x1).fit()
t3.append(reg.tvalues[0])
alpha3.append(reg.params[0] * 12)
x2 = sm.add_constant(X2)
reg = sm.OLS(Y, x2).fit()
t5.append(reg.tvalues[0])
alpha5.append(reg.params[0] * 12)
print('alpha-FF3', alpha3[0] / 12, alpha3[1] / 12, alpha3[2] / 12)
print('t-statistic', '(' + str(round(t3[0], 4)) + ')', '(' + str(round(t3[1], 4)) + ')',
'(' + str(round(t3[2], 4)) + ')')
print('alpha-FF5', alpha5[0] / 12, alpha5[1] / 12, alpha5[2] / 12)
print('t-statistic', '(' + str(round(t5[0], 4)) + ')', '(' + str(round(t5[1], 4)) + ')',
'(' + str(round(t5[2], 4)) + ')')
print('sharpe', sharpratio[0], sharpratio[1], sharpratio[2])
def comboutput(length, clf, name, rf,timeseries2, index):
Long_Short = []
Long = []
Short = []
for i in range(len(timeseries2) - length):
print(i)
# LSTM数据
FINALm = pd.concat(timeseries2[i:(i + length + 1)], axis=0)
FINALm[~FINALm['ret'].isin(['null'])] = FINALm[~FINALm['ret'].isin(['null'])].fillna(0)
FINAL_X = FINALm.iloc[:, :-2]
FINAL_x = FINAL_X
FINAL_x[~FINALm['ret'].isin(['null'])] = scale(FINAL_X[~FINALm['ret'].isin(['null'])])
FINAL_x[FINALm['ret'].isin(['null'])] = 0
FINALm[FINALm['ret'].isin(['null'])] = 0
Nx_train = [FINAL_x.iloc[j * 3571:(j + 1) * 3571, :].values for j in range(length)]
Ny_train = [FINALm.iloc[j * 3571:(j + 1) * 3571, -1].values for j in range(length)]
Nx_test = [FINAL_x.iloc[j * 3571:(j + 1) * 3571, :].values for j in range(1, length + 1)]
## 传统数据
dl = timeseries2[i:i + (length + 1)]
p = [dl[j][~index[i+j]] for j in range(len(dl))]
TTX = pd.concat(p, axis=0)
TTX = TTX.fillna(0)
TXX = TTX.iloc[:, :-2]
TXx = scale(TXX)
final = pd.concat(p[:length], axis=0)
Tx_train = TXx[:len(final)]
Tx_test = TXx[len(final):]
Ty_train = final.iloc[:, -1]
test = p[-1]
Ty_test = test.iloc[:, -1]
# 基准-linear
clf = clf
clf.fit(Tx_train, Ty_train, Nx_train, Ny_train)
PREDICTION = clf.predict(Tx_test, Nx_test, index[i+length], length)
r_predict = pd.DataFrame(PREDICTION, columns=['predict'])
r_ture = pd.DataFrame(Ty_test)
r_ture.columns = ['ture']
r_ture.index = r_predict.index
FINAL = pd.concat([r_predict, r_ture], axis=1)
FINAL_sort = FINAL.sort_values(by='predict', axis=0)
r_final = np.array(FINAL_sort['ture'])
m = int(len(r_final) * 0.1) + 1
r_final = r_final.tolist()
long = r_final[-m:]
short = r_final[:m]
r_end = (np.sum(long) - np.sum(short)) / m
Long_Short.append(r_end)
Long.append(np.average(long))
Short.append(np.average(short))
gc.collect()
T_value = []
Mean = []
p_value = []
sharpratio = []
Std = []
TO = [Long_Short, Long, Short]
for l in TO:
t_test = nwttest_1samp(l, 0, L=1)
mean = np.average(l) * 12 - rf.mean().tolist()[0] * 12 / 100
STD = np.std(l) * np.sqrt(12)
sharp = (mean) / STD
T_value.append(t_test.statistic)
p_value.append(t_test.pvalue)
Mean.append(mean)
Std.append(STD)
sharpratio.append(sharp)
print(name, 'long-short', 'long', 'short')
print('mean', Mean[0] / 12, Mean[1] / 12, Mean[2] / 12)
print('t-statistic', '(' + str(round(T_value[0], 4)) + ')', '(' + str(round(T_value[1], 4)) + ')',
'(' + str(round(T_value[2], 4)) + ')')
A = pd.DataFrame(Long_Short, columns=['long-short'])
B = pd.DataFrame(Long, columns=['long'])
C = pd.DataFrame(Short, columns=['short'])
M = pd.concat([A, B, C], axis=1)
M.to_csv('..\output\\' + name + '.csv')
ff3 = pd.read_csv('..\DataBase\\ff3.csv')
ff5 = pd.read_csv('..\DataBase\\ff5.csv')
alpha3 = []
t3 = []
t5 = []
alpha5 = []
for i in range(3):
X1 = ff3.iloc[length:, 1:]
X2 = ff5.iloc[length:, 1:]
Y = M.iloc[:-2, i]
Y.index = X1.index
Y = Y - rf.RF[:-1] / 100
x1 = sm.add_constant(X1)
reg = sm.OLS(Y, x1).fit()
t3.append(reg.tvalues[0])
alpha3.append(reg.params[0] * 12)
x2 = sm.add_constant(X2)
reg = sm.OLS(Y, x2).fit()
t5.append(reg.tvalues[0])
alpha5.append(reg.params[0] * 12)
print('alpha-FF3', alpha3[0] / 12, alpha3[1] / 12, alpha3[2] / 12)
print('t-statistic', '(' + str(round(t3[0], 4)) + ')', '(' + str(round(t3[1], 4)) + ')',
'(' + str(round(t3[2], 4)) + ')')
print('alpha-FF5', alpha5[0] / 12, alpha5[1] / 12, alpha5[2] / 12)
print('t-statistic', '(' + str(round(t5[0], 4)) + ')', '(' + str(round(t5[1], 4)) + ')',
'(' + str(round(t5[2], 4)) + ')')
print('sharpe', sharpratio[0], sharpratio[1], sharpratio[2])
def output2(length,CLF,name,rf,timeseries2):
# length为滑动窗口长度：取值{3,12,24,36}
# CLF为预测选取的机器学习模型
# name为输出文件名称（type:string)
# rf为无风险利率，取值与length对应{rf3,rf12,rf24,rf36}
Long_Short = []
Long = []
Short = []
for i in range(len(timeseries2) - length):
FINALm = pd.concat(timeseries2[i:(i + length + 1)], axis=0)
FINALm[~FINALm['ret'].isin(['null'])] = FINALm[~FINALm['ret'].isin(['null'])].fillna(0)
FINAL_X = FINALm.iloc[:, :-2]
FINAL_x = FINAL_X
FINAL_x[~FINALm['ret'].isin(['null'])] = scale(FINAL_X[~FINALm['ret'].isin(['null'])])
FINAL_x[FINALm['ret'].isin(['null'])] = 0
FINALm[FINALm['ret'].isin(['null'])] = 0
x_train = [FINAL_x.iloc[j * 3571:(j + 1) * 3571, :].values for j in range(length)]
y_train = [FINALm.iloc[j * 3571:(j + 1) * 3571, -1].values for j in range(length)]
x_test = np.array([FINAL_x.iloc[j * 3571:(j + 1) * 3571, :].values for j in range(1, length + 1)])
y_test = list(FINALm.iloc[(length) * 3571:(length + 1) * 3571, -1].values)
# 基准-linear
clf = CLF
clf.fit(x_train, y_train)
PREDICTION = clf.predict(x_test)
PREDICTION = [PREDICTION[m][0] for m in range(3571 * (length - 1), 3571 * length)]
# 构建投资组合
r_predict = pd.DataFrame(PREDICTION, columns=['predict'])
r_ture = pd.DataFrame(y_test)
r_ture.columns = ['ture']
r_ture.index = r_predict.index
FINAL = pd.concat([r_predict, r_ture], axis=1)
FINAL = FINAL[~timeseries2[i + length]['ret'].isin(['null'])]
FINAL_sort = FINAL.sort_values(by='predict', axis=0)
r_final = np.array(FINAL_sort['ture'])
m = int(len(r_final) * 0.1) + 1
r_final = r_final.tolist()
long = r_final[-m:]
short = r_final[:m]
r_end = (np.sum(long) - np.sum(short)) / m
Long_Short.append(r_end)
Long.append(np.average(long))
Short.append(np.average(short))
gc.collect()
T_value = []
Mean = []
p_value = []
sharpratio = []
Std = []
TO = [Long_Short, Long, Short]
for l in TO:
t_test = nwttest_1samp(l, 0, L=1)
mean = np.average(l) * 12- rf.mean().tolist()[0] * 12 / 100
STD = np.std(l) * np.sqrt(12)
sharp = (mean ) / STD
T_value.append(t_test.statistic)
p_value.append(t_test.pvalue)
Mean.append(mean)
Std.append(STD)
sharpratio.append(sharp)
print(name, 'long-short', 'long', 'short')
print('mean', Mean[0] / 12 + rf.mean().tolist()[0] / 100, Mean[1] / 12 + rf.mean().tolist()[0] / 100, Mean[2] / 12 + rf.mean().tolist()[0] / 100)
print('t-statistic', '('+str(round(T_value[0],4))+')', '('+str(round(T_value[1],4))+')', '('+str(round(T_value[2],4))+')')
A = pd.DataFrame(Long_Short, columns=['long-short'])
B = pd.DataFrame(Long, columns=['long'])
C = pd.DataFrame(Short, columns=['short'])
M = pd.concat([A, B, C], axis=1)
M.to_csv('..\output\\' + name + '.csv')
ff3 = pd.read_csv('..\DataBase\\ff3.csv')
ff5 = pd.read_csv('..\DataBase\\ff5.csv')
alpha3 = []
t3 = []
t5 = []
alpha5 = []
for i in range(3):
X1 = ff3.iloc[length:, 1:]
X2 = ff5.iloc[length:, 1:]
Y = M.iloc[:-2, i]
Y.index = X1.index
Y = Y - rf.RF[:-1] / 100
x1 = sm.add_constant(X1)
reg = sm.OLS(Y, x1).fit()
t3.append(reg.tvalues[0])
alpha3.append(reg.params[0] * 12)
x2 = sm.add_constant(X2)
reg = sm.OLS(Y, x2).fit()
t5.append(reg.tvalues[0])
alpha5.append(reg.params[0] * 12)
print('alpha-FF3', alpha3[0] / 12, alpha3[1] / 12, alpha3[2] / 12)
print('t-statistic', '(' + str(round(t3[0], 4)) + ')', '(' + str(round(t3[1], 4)) + ')',
'(' + str(round(t3[2], 4)) + ')')
print('alpha-FF5', alpha5[0] / 12, alpha5[1] / 12, alpha5[2] / 12)
print('t-statistic', '(' + str(round(t5[0], 4)) + ')', '(' + str(round(t5[1], 4)) + ')',
'(' + str(round(t5[2], 4)) + ')')
print('sharpe', sharpratio[0], sharpratio[1], sharpratio[2])
x.columns = [1, 2, 3, 4, 5, 6, 7, '7-1']
x.index = [namelist[i], ' ']
table = pd.concat([table, x])
table = table.applymap(lambda x: round(x, 2))
print(table)
showtable8()
###生成表格9
beta_equal_annual = annual_equal_mean(beta_group)
beta_equal_annual.index = beta_equal_annual['year']
del beta_equal_annual['year']
#excessreturn部分结果
df1 = beta_equal_annual.mean()
df2 = pd.Series(nwttest_1samp(beta_equal_annual, 0)[0], index=df1.index)
excess_return = pd.concat([df1, df2], axis=1).T
excess_return.index = ['excess return', ' ']
excess_return.index.name = None
excess_return.columns = [1, 2, 3, 4, 5, 6, 7, 8]
#制作capm部分结果
CAPM = pd.DataFrame(index=['alpha', ' ', 'beta_mkt', ' '])
for i in range(8):
name = beta_equal_annual.columns
data = pd.concat([beta_equal_annual[name[i]], ff['mktrf']], axis=1)
capm = smf.ols(name[i] + '~mktrf', data).fit(cov_type='HAC',
cov_kwds={'maxlags': 6})
X = [capm.params[0], capm.tvalues[0], capm.params[1], capm.tvalues[1]]
X = pd.Series(X, name=i + 1, index=['alpha', ' ', 'beta_mkt', ' '])
CAPM = pd.concat([CAPM, X], axis=1)
#制作FF部分结果
't', 'ff30ret', 't'
])
rf = pd.read_csv('RF.csv')
for i in range(len(factoreq.columns) - 1):
if factoreq.iloc[:, i + 1].mean() > 0:
faceq = factoreq.iloc[:, i + 1]
else:
faceq = -factoreq.iloc[:, i + 1]
if factorvw.iloc[:, i + 1].mean() > 0:
facvw = factorvw.iloc[:, i + 1]
else:
facvw = -factorvw.iloc[:, i + 1]
####return
reteq0 = faceq.dropna()
resultew.iloc[0, i] = reteq0.mean()
ttest = nwttest_1samp(reteq0, 0)
resultew.iloc[1, i] = ttest.statistic
retvw0 = facvw.dropna()
resultvw.iloc[0, i] = retvw0.mean()
ttest = nwttest_1samp(retvw0, 0)
resultvw.iloc[1, i] = ttest.statistic
##excess return
exeq = faceq - rf.RF / 100
exvw = facvw - rf.RF / 100
exeq0 = exeq.dropna()
resultew.iloc[2, i] = exeq0.mean()
#ttest=stats.ttest_1samp(exeq0,0)
ttest = nwttest_1samp(exeq0, 0)
resultew.iloc[3, i] = ttest.statistic
exvw0 = exvw.dropna()
