【日経225先物】売買システム開発に向けた統計的分析①

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
# 日経225先物データを読み込む
N225_daytime = pd.read_excel('./data/N225f_2020.xlsx', sheet_name='日中日足', index_col='日付')
N225_night = pd.read_excel('./data/N225f_2020.xlsx', sheet_name='ナイト場足', index_col='日付')
# 結合
df = pd.merge(N225_daytime, N225_night, on='日付', suffixes=['_D', '_N']) # D:daytime, N:night
# 各パターンの価格差を求める
daytime = df['終値_D'] - df['始値_D'] 
d_to_n = df['始値_N'] - df['終値_D'] 
night = df['終値_N'] - df['始値_N']
n_to_d = df['始値_D'] - df['終値_N'].shift(1)
# 価格差の累積和をグラフにする
plt.figure(figsize=(8, 6))
daytime.cumsum().plot(label='daytime')
d_to_n.cumsum().plot(label='d_to_n')
night.cumsum().plot(label='night')
n_to_d.cumsum().plot(label='n_to_d')
plt.legend(loc='upper left')

# ヒストグラムを書く
plt.figure(figsize=(12, 4))
ax1 = plt.subplot(1, 2, 1)
daytime.hist(label='daytime', histtype='step', linewidth=3, ax=ax1)
d_to_n.hist(label='d_to_n', color='lightgreen', rwidth=0.7, ax=ax1)
plt.legend()
ax2 = plt.subplot(1, 2, 2)
night.hist(label='night', histtype='step', linewidth=3, ax=ax2)
n_to_d.hist(label='n_to_d', color='lightgreen', rwidth=0.7, ax=ax2)
plt.legend()
# 記述統計を算出
stats_desc = pd.concat([daytime.describe(), d_to_n.describe(), night.describe(), n_to_d.describe()],axis=1)
stats_desc.columns = ['daytime', 'd_to_n', 'night', 'n_to_d']　
pd.options.display.precision = 4 # dfの桁数を調整
stats_desc

alpha = 0.15
patterns = {'daytime': daytime,
            'd_to_n' : d_to_n,
            'night'  : night,
            'n_to_d' : n_to_d}
results = []
for key in patterns.keys():
    target = patterns[key].dropna()
    # ADF検定
    adf_nc = sm.tsa.adfuller(target, regression='nc')[1] #p値：ドリフト項無し
    adf_c  = sm.tsa.adfuller(target, regression='c')[1] #p値：ドリフト項付き
    adf_ct = sm.tsa.adfuller(target, regression='ct')[1] #p値：ドリフト項＋時間項付き
    # t検定
    t_value = target.mean() * np.sqrt(target.count()) / target.std()
    t0 = t.ppf(1-alpha, target.count()-1)
    if t_value > t0:
        t_result = 'μ>0'
    elif t_value < -t0:
        t_result = 'μ<0'
    else:
        t_result = 'μ=0'
    # JB検定
    JB = sms.jarque_bera(target) # JB, p値, 歪度, 尖度
    
    # 結果を保存
    results.append([adf_nc, adf_c, adf_ct, t_result, JB[1], JB[2], JB[3]])
# dfにまとめる
df_result = pd.DataFrame(results, index=patterns.keys())
df_result.columns = ['ADF nc', 'ADF c', 'ADF ct', 't', 'JB', '歪度', '尖度']
df_result.T