def stationary_test(self, test_df, plot=False):
test_obj = test_df['return'][1::]
acf = stattools.acf(test_obj, nlags=10)
pacf = stattools.pacf(test_obj, nlags=10)
ADF = unitroot_adf(test_obj)
if plot:
f = plt.figure(facecolor='white')
ax1 = f.add_subplot(211)
plot_acf(test_obj, lags=10, ax=ax1)
ax2 = f.add_subplot(212)
plot_pacf(test_obj, lags=10, ax=ax2)
plt.show()
#plt.figure(figsize = (10,10))
#plt.stem(acf)
#plt.title('ACF')
#plt.show()
#plt.figure(figsize = (10,10))
#plt.stem(pacf)
#plt.title('PACF')
#plt.show()
return {'acf': acf, 'pacf': pacf, 'adf': ADF}
def baseline(self):
import pandas as pd
import math
import numpy as np
from dev_global.env import TIME_FMT
df = self.mysql.select_values('SH000300', 'trade_date,close_price')
# data cleaning
df.columns = ['trade_date', 'close_price']
pd.to_datetime(df['trade_date'], format=TIME_FMT)
df.set_index('trade_date', inplace=True)
# data constructing
df['shift'] = df['close_price'].shift(1)
df['amplitude'] = df['close_price'] / df['shift']
df['ln_amplitude'] = np.log(df['amplitude'])
df.dropna(inplace=True)
# print(df.head(5))
# plot
input_df = df['ln_amplitude'][-200:]
import statsmodels.tsa.api as smt
acf = smt.stattools.acf(input_df, nlags=40)
pacf = smt.stattools.pacf(input_df, nlags=40)
acf_pacf_plot(input_df, lags=40)
from statsmodels.stats.diagnostic import unitroot_adf
result = unitroot_adf(input_df)
print(result[1])
def adf_(timeseries): # adf_ 检验平稳性
"""
:param timeseries: time series that aims to analyse
:return: the values of the adfuller test and critical test, in order to determine whether the time series is stable or not
"""
adf_test = unitroot_adf(timeseries)
adf_test_value = adf_test[0]
adfuller_value = pd.DataFrame(
{key: value
for key, value in adf_test[4].items()}, index=[0])
adfuller_value = pd.DataFrame(adfuller_value)
adfuller_critical_value = adfuller_value['10%'][0]
return adf_test_value, adfuller_critical_value
def seasonality_price_decomp(self, test_df, f=0, plot=False):
if f == 0:
raise ValueError('\nError freqency input!! \n')
obv = test_df['close']
if test_df['timestamp'][1] - test_df['timestamp'][0] == 60:
raise ValueError('\n Wrong using minute data! \n')
#since 'return[0]' is nan
decomposition = seasonal_decompose(obv, freq=f, model='additive')
trend = decomposition.trend
seasonal = decomposition.seasonal
residual = decomposition.resid
ADF = unitroot_adf(test_df['return'][1::])
if plot:
plt.figure(figsize=(15, 10))
plt.subplot(411)
plt.plot(obv, label='obv', lw=0.7)
plt.legend(loc='best')
plt.subplot(412)
plt.plot(trend, label='trend', lw=0.7)
plt.legend(loc='best')
plt.subplot(413)
plt.plot(seasonal, label='seasonal', lw=0.7)
plt.legend(loc='best')
plt.subplot(414)
plt.plot(residual, label='residual', lw=0.7)
plt.legend(loc='best')
plt.show()
print('stats =', ADF[0], 'Alpha =', ADF[4])
if ADF[0] < ADF[4]['1%']:
print('\nResidual is stable in 99% confid. interval')
return {
'trend': trend,
'seasonal': seasonal,
'residual': residual,
'adf': ADF
}
## 平稳性检验
# Method 1: time series plot
fig, ax = plt.subplots()
fig.set_size_inches(9, 3.5)
ax.plot(df['Date'], df['Close'])
ax.xaxis.set_major_locator(matplotlib.ticker.MultipleLocator(90))
plt.xticks(rotation=90)
ax.tick_params(labelsize=7)
plt.show()
# Method 2: calculate and plot ACF and PACF
n_lag = 100 ###lag
acf = stattools.acf(df['Close'], nlags=n_lag) #Autocorrelation Coefficient
pacf = stattools.pacf(df['Close'], nlags=n_lag)
print('Autocorrelation Coefficient (ACF): \n{}'.format(acf))
print('Partial Autocorrelation Coefficient (PACF): \n{}'.format(pacf))
sm.graphics.tsa.plot_acf(df['Close'], lags=n_lag)
plt.show() # 阴影部分是置信区间。默认情况下，置信区间被设置为95%。
sm.graphics.tsa.plot_pacf(df['Close'], lags=n_lag)
plt.show()
# 单位根检验（这里采用ADF检验，分别用两种方法进行，第二种的输出效果较好)
from statsmodels.stats.diagnostic import unitroot_adf
adf_method1 = unitroot_adf(df['Close'])
print('ADF method 1: \n{}'.format(adf_method1))
from arch.unitroot import ADF
adf_method2 = ADF(df['Close'])
print('ADF method 2: \n{}'.format(adf_method2))
plt.figure(figsize=(8, 5))
plt.errorbar(s.index, means, yerr=sigmas, alpha=0.5)
plt.plot(s.index, means, 'g', linewidth=4)
plt.show()
# 平稳性检测
df = pd.concat([pd.DataFrame(date_range[days:],columns=['date']),
pd.DataFrame(score,columns=['score'])],
axis=1)
from statsmodels.tsa import stattools
from statsmodels.stats.diagnostic import unitroot_adf
print(unitroot_adf(df.score))
#plt.stem(stattools.acf(df.score));
k = stattools.pacf(df.score)
k[1] = 0.003
plt.stem(k);
# 画出波峰波谷
data = means.reshape((len(means),))
doublediff = np.diff(np.sign(np.diff(data)))
peak_locations = np.where(doublediff == -2)[0] + 1
peak_locations = peak_locations[:(len(peak_locations)-1)]
doublediff2 = np.diff(np.sign(np.diff(-1*data)))
trough_locations = np.where(doublediff2 == -2)[0] + 1
for j in codelist:
datas_filename = para.path_data + '%s' % j + '_' + '%s' % i + '.xlsx'
datas = pd.read_excel(datas_filename, index_col=0, parse_dates=True)
datas_minus_signal = pd.DataFrame(datas[datas.signal == c],columns = datas.columns)
datas_minus_signal = pd.DataFrame(datas.iloc[:,2:])
# step 1:描述性统计分析报告
profile = datas_minus_signal.profile_report(title='%s' % j + '_' + '%s' % i + '_' + '%s' % stage+ ' Exploratory Data Analysis')
profile.to_file(
output_file=para.path_results + '%s' % j + '_' + '%s' % i + '_' + '%s' % stage+ 'Exploratory Data Analysis.html')
period = 15
dict_roll = {}
for k in range(1, datas_minus_signal.columns.size):
columns_name = datas_minus_signal.columns[k]
# step 2: 时间序列检验：单位根检验
adf_result = unitroot_adf(datas_minus_signal.iloc[:, k].dropna())
adf_result_diff = unitroot_adf(datas_minus_signal.iloc[:, k].diff().dropna())
print('时间序列检验：单位根检验: %s' % j + '_%s' % i+ '_' + '%s' % stage, columns_name,
round(adf_result[0], 4), round(adf_result[4]['5%'], 4))
print('时间序列检验：单位根检验: %s' % j + '_%s' % i+ '_' + '%s' % stage, columns_name, '同比',
round(adf_result_diff[0], 4), round(adf_result_diff[4]['5%'], 4))
# step 3: 画图
y = pd.DataFrame(datas_minus_signal.iloc[:, 0]).apply(
lambda x: (x - np.min(x)) / (np.max(x) - np.min(x)))
X = pd.DataFrame(datas_minus_signal.iloc[:, k]).apply(
lambda x: (x - np.min(x)) / (np.max(x) - np.min(x)))
figure_count = 1
plt.figure(figure_count)
figure_count += 1
plt.plot(y, 'k-', label='%s' % j+ '_' + '%s' % stage)
plt.plot(X, 'b-', label='%s' % i + '_' + '%s' % stage+ '%s' % columns_name)
plt.title("上证指数的收益率序列")
plt.savefig("plot_images/上证收益率序列.png")
plt.show()
#得到统计量，这部分写入函数mean,max,min,std,skewness,
data_analyse(shdf['ratio'].values)
#绘制收益率的分布直方图
import seaborn
plt.figure(figsize=(8, 3))
seaborn.distplot(shdf['ratio'].values, bins=50, kde=False)
plt.title("收益率分布直方图")
plt.savefig("plot_images/收益率分布直方图.png")
plt.show()
plt.close()
#平稳性检验，单位根检验。ADF
from statsmodels.stats.diagnostic import unitroot_adf
unitroot_adf(shdf['ratio'].values)
#对序列做Ljung-box检验。
import statsmodels as sm
Q, P = sm.stats.diagnostic.acorr_ljungbox(shdf['ratio'].values, lags=20)
box_test = pd.DataFrame({
"Lags": np.arange(1, 21),
"Q_statistic": Q,
"p_value": P
})
print(box_test)
box_test.to_csv("intermediate/Ljung_box.csv") # doctest: +SKIP