TA-Libでテクニカル分析

簡単にテクニカル分析ができるライブラリTA-Libを試しました．TA-LibにはPythonのラッパーがあり，かなり簡単にですが紹介します．

準備

はじめに，TA-Libをインストールする必要があります．Macの場合はHomebrewでインストールできます．

$ brew install ta-lib

Pythonで使えるようにします．

$ pip install TA-Lib

テクニカル分析

今回使用するライブラリをインポートします．

import numpy as np
import pandas as pd
import talib
import quandl

今回はquandlでデータを取得します．

df = quandl.get('NIKKEI/INDEX')  # quandlの場合

TA-LibはDataFrameを引数にできないので，numpyのndarray形式にします．

close = np.array(df['Close Price'])
close[:5]

とにかく，いろいろなテクニカル指標を試してみます．（今回は終値のみで，ローソク足を用いる指標は使っていません．）

output = close.copy()
cols = ['Original']
# 単純移動平均(SMA: Simple Moving Average)
output = np.c_[output, talib.SMA(close)]
cols += ['SMA']
# 加重移動平均(WMA: Weighted Moving Average)
output = np.c_[output, talib.WMA(close)]
cols += ['WMA']
# 指数移動平均(EMA: Exponential Moving Average)
output = np.c_[output, talib.EMA(close)]
cols += ['EMA']
# ２重指数移動平均(DEMA: Double Exponential Moving Average)
output = np.c_[output, talib.DEMA(close)]
cols += ['DEMA']
# ３重指数移動平均(TEMA: Triple Exponential Moving Average)
output = np.c_[output, talib.T3(close)]
cols += ['TEMA']
# 三角移動平均(TMA: Triangular Moving Average)
output = np.c_[output, talib.TRIMA(close)]
cols += ['TMA']
# Kaufmanの適応型移動平均(KAMA: Kaufman Adaptive Moving Average)
output = np.c_[output, talib.KAMA(close)]
cols += ['KAMA']
# MESAの適応型移動平均(MAMA: MESA Adaptive Moving Average)
for arr in talib.MAMA(close):
    output = np.c_[output, arr]
cols += ['MAMA', 'FAMA']
# トレンドライン(Hilbert Transform - Instantaneous Trendline)
output = np.c_[output, talib.HT_TRENDLINE(close)]
cols += ['HT_TRENDLINE']
# ボリンジャー・バンド(Bollinger Bands)
for arr in talib.BBANDS(close):
    output = np.c_[output, arr]
cols += ['BBANDS_upperband', 'BBANDS_middleband', 'BBANDS_lowerband']
# MidPoint over period
output = np.c_[output, talib.MIDPOINT(close)]
cols += ['MIDPOINT']

# 変化率(ROC: Rate of change Percentage)
output = np.c_[output, talib.ROCP(close)]
cols += ['ROC']
# モメンタム(Momentum)
output = np.c_[output, talib.MOM(close)]
cols += ['MOM']
# RSI: Relative Strength Index
output = np.c_[output, talib.RSI(close)]
cols += ['RSI']
# MACD: Moving Average Convergence/Divergence
for arr in talib.MACD(close):
    output = np.c_[output, arr]
cols += ['MACD', 'MACD_signal', 'MACD_hist']
# APO: Absolute Price Oscillator
output = np.c_[output, talib.APO(close)]
cols += ['APO']
# PPO: Percentage Price Oscillator
output = np.c_[output, talib.PPO(close)]
cols += ['PPO']
# CMO: Chande Momentum Oscillator
output = np.c_[output, talib.CMO(close)]
cols += ['CMO']

# ヒルベルト変換 - Dominant Cycle Period
output = np.c_[output, talib.HT_DCPERIOD(close)]
cols += ['HT_DCPERIOD']
# ヒルベルト変換 - Dominant Cycle Phase
output = np.c_[output, talib.HT_DCPHASE(close)]
cols += ['HT_DCPHASE']
# ヒルベルト変換 - Phasor Components
for arr in talib.HT_PHASOR(close):
    output = np.c_[output, arr]
cols += ['HT_PHASOR_inphase', 'HT_PHASOR_quadrature']
# ヒルベルト変換 - SineWave
for arr in talib.HT_SINE(close):
    output = np.c_[output, arr]
cols += ['HT_SINE_sine', 'HT_SINE_leadsine']    
# ヒルベルト変換 - Trend vs Cycle Mode
output = np.c_[output, talib.HT_TRENDMODE(close)]
cols += ['HT_TRENDMODE']
output.shape

# 60日単純移動平均
output = np.c_[output, talib.SMA(close, timeperiod=60)]
cols += ['SMA60']
# 15日ボリンジャー・バンド
for arr in talib.BBANDS(close, timeperiod=15, nbdevup=2, nbdevdn=2, matype=0):
    output = np.c_[output, arr]
cols += ['BBANDS15_upperband', 'BBANDS15_middleband', 'BBANDS15_lowerband']
# 21日RSI
output = np.c_[output, talib.RSI(close, timeperiod=21)]
cols += ['RSI21']

引数のtimeperiodを指定して，いろんな期間を計算できます．

DataFrame形式にして，日付と対応させます．

data = pd.DataFrame(output, index=df.index, columns=cols)
data.tail()

	Original	SMA	WMA	EMA	DEMA	TEMA	TMA	KAMA	MAMA	FAMA	...	HT_PHASOR_inphase	HT_PHASOR_quadrature	HT_SINE_sine	HT_SINE_leadsine	HT_TRENDMODE	SMA60	BBANDS15_upperband	BBANDS15_middleband	BBANDS15_lowerband	RSI21
Date
2018-01-26	23631.88	23336.888333	23577.879333	23396.084808	23950.015680	23881.542216	23358.915125	23401.776887	23774.262642	23579.472022	...	21.325892	-43.817842	-0.589713	-0.988059	1.0	22920.246667	24056.610766	23800.404000	23544.197234	58.287358
2018-01-29	23629.34	23362.327333	23596.747183	23411.133530	23943.404763	23819.197807	23416.499625	23410.021533	23701.801321	23610.054347	...	9.146663	81.389471	-0.713710	-0.999956	1.0	22947.261500	24061.838563	23794.724667	23527.610770	58.232275
2018-01-30	23291.97	23380.124000	23592.208000	23403.445560	23894.184742	23717.474237	23468.224208	23407.458663	23681.309755	23611.835732	...	139.081959	117.191239	-0.837169	-0.978716	1.0	22968.599833	24121.395394	23757.523333	23393.651273	51.451184
2018-01-31	23098.29	23393.585333	23574.025161	23383.758105	23824.419397	23575.717839	23512.524500	23402.617512	23652.158767	23612.843808	...	165.813706	-406.110394	-0.942737	-0.902462	0.0	22986.711167	24200.998868	23711.529333	23222.059798	48.076476
2018-02-01	23486.11	23424.681667	23579.994495	23390.361453	23808.770310	23470.100518	23551.332833	23405.839309	23569.134384	23601.916452	...	-122.751421	-700.414139	-0.978461	-0.837845	0.0	23004.478333	24198.805592	23696.574667	23194.343742	54.369135

5 rows × 36 columns

とりあえず，CSVに書き出します．

data.to_csv('NIKKEI_ta.csv')

TA-Libは予測モデルの特徴量の作成にも便利です．