Minimum Covariance Determinant or MLEで求めた共分散行列による異常探知比較

飛んだ値に思いっ切り引きずられるのを避けるためにも、マーケットデータと相性良いメソッドかと


import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
import warnings
from sklearn.covariance import EmpiricalCovariance, MinCovDet
from sklearn.covariance import EmpiricalCovariance

warnings.filterwarnings('ignore')

plt.style.use('seaborn-darkgrid')
plt.rcParams['axes.xmargin'] = 0.01
plt.rcParams['axes.ymargin'] = 0.01

ReadDFに週次の各種マーケットリターン(為替、株、債券etc)

# ルックバック過去50週間リターンを用いて算出

ts_out = pd.DataFrame()
for date in ReadDF.dropna(axis=0)[50:].index:
    x = ReadDF[:date][-50:]
    x = (x/x.std()).dropna(axis=0)
    
    mcd.fit(x[:-1])
    anomaly_score_mcd = mcd.mahalanobis(x[-1:])
    
    mle.fit(x[:-1])
    anomaly_score_mle = mle.mahalanobis(x[-1:])
    
    out = pd.DataFrame([anomaly_score_mcd, anomaly_score_mle]).T
    
    out.columns = ['mcd', 'mle']
    out.index = [date]
    
    ts_out = pd.concat([ts_out, out], axis=0)

fig = plt.figure(figsize=(15, 10), dpi=80)
out = pd.DataFrame(ts_out/ts_out.std())  # 規格化
sns.set_palette("hls", len(out.columns))
ax1 = fig.add_subplot(1, 1, 1)
ax1.plot(out, alpha=0.6)
plt.legend(out.columns)
plt.show()