머신러닝을 이용한 트레이딩: (6) 매매 규칙

Trading Rules

trading
Author

Cheonghyo Cho

Trading rules: 매수 진입만 허용

# lib
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns;sns.set()
plt.style.use('tableau-colorblind10')

from sklearn.preprocessing import StandardScaler, MinMaxScaler, normalize

# homemade
from features import trnd_scan, tautil
from labeling import labeling
from triple_barrier import get_barrier, make_rt
from backtest import print_round_trip_stats, round_trip
import warnings
warnings.filterwarnings(action='ignore')
market_df = pd.read_csv('C:data/market_samsung.csv')
market_df = market_df.rename(columns={market_df.columns[0]:'Date'})
market_df.index = pd.to_datetime(market_df.Date)
market_df.drop(columns='Date',inplace=True)
market_df.dropna(inplace=True)

close = market_df.close['2010':'2020']

매매 규칙

진입 규칙

Momentum Prediction

signals = pd.read_csv('C:data/momentum_signals.csv')
signals.index = pd.to_datetime(signals['Date'])
signals.drop(columns='Date',inplace=True)
signals = signals['signals'].loc['2010':'2020']
scaler = normalize
scaler2 = MinMaxScaler()
signals = pd.Series(scaler2.fit_transform(normalize(signals.values.reshape(-1,1),axis=0)).reshape((-1,)), 
                           index=signals.index).rename('signals')
plt.hist(signals,bins=50)[2]
plt.title('Distribution of momentum signals')
plt.show()

thresholds = [0, 0.3]
enter_ml_list=[]
for h in thresholds:
    enter_ml_list.append(signals.loc[signals>h].index)
for i in range(len(thresholds)):
    plt.figure(figsize=(10,3))
    plt.plot(close, alpha=0.5)
    plt.title('Enter points ML Prediction (Option {})'.format(i+1))
    plt.plot(close.loc[enter_ml_list[i]],marker='^',linewidth=0,alpha=0.3)
    plt.legend(['price','Long signals from momentum classifier'])
    plt.show()

Tech. Analysis Long/short decision

open = market_df.open['2010':'2020']
rsi = tautil.RSIIndicator(open,14).rsi().dropna()
long = (rsi>=50) & (rsi<70)
enter_ta = rsi.loc[long].index
plt.figure(figsize=(10,3))
plt.plot(close, alpha=0.5)
plt.title('Enter points TA')
plt.plot(close.loc[enter_ta],marker='^',linewidth=0,alpha=0.3)
plt.legend(['price','Long signals from rsi'])
plt.show()

enter_list = [enter_ta]
enter_list.append((enter_ml_list[1]& enter_ta).sort_values().drop_duplicates())
for i in range(len(thresholds)):
    plt.figure(figsize=(10,3))
    plt.plot(close, alpha=0.5)
    plt.title('Enter points (Option {})'.format(i+1))
    plt.plot(close.loc[enter_list[i]],marker='^',linewidth=0,alpha=0.3)
    plt.legend(['price','Enter points'])
    plt.show()

plt.figure(figsize=(10,3))
plt.plot(close, alpha=0.5)
plt.title('Position Enter points'.format(i+1))
plt.plot(close.loc[enter_list[1]],marker='^',linewidth=0,alpha=0.3)
plt.legend(['price','Enter points'])
plt.show()

청산 규칙

# no Rule (benchmark)
pt_sl_bm = [1000,1000]
max_holding_bm = [1, 0]
no_exit_rule = [pt_sl_bm,max_holding_bm]
#dynamic target rule
max_holding = [60, 0]
close_ = market_df.close['2009':'2020']
changes = close_.pct_change(1).to_frame()
for i in range(2,max_holding[0]+1):
    changes = changes.join(close_.pct_change(i).rename('close {}'.format(i)))
dynamic_target = changes.abs().dropna().mean(axis=1)['2010':]
barrier_exit_list=[]
barrier_exit_list.append(get_barrier(close, enter_list[1], [1,1], max_holding, target = dynamic_target))  #dynamic  

rts_exit_list=[]
for i in range(len(barrier_exit_list)):
    rts_exit_list.append(make_rt(close,barrier_exit_list[i].dropna()))
plt.figure(figsize=(10,2))
plt.title('Dynamic exit rule returns')
plt.plot(barrier_exit_list[0].ret)
plt.legend(['Returns of dynamic exit target rate'])
<matplotlib.legend.Legend at 0x1fe01fa7148>

결과

벤치마크

  • 매번 매매시
barrier_bm = get_barrier(close, close.index, no_exit_rule[0], no_exit_rule[1]) #no rule
rts_bm = make_rt(close,barrier_bm.dropna())
round_trip.get_df_ann_sr(rts_bm,'Benchmark',years=11)
Benchmark
avg_n_bets_per_year 243.272727
win_ratio 0.518835
annualized_sharpe_ratio 0.537080 
result_df = pd.concat([round_trip.get_df_ann_sr(rts_bm,'No Rule')], axis=1)
for i in range(len(rts_exit_list)):
    result_df = result_df.join(round_trip.get_df_ann_sr(rts_exit_list[i],'Enter & Exit Rule'))
    
result_df
No Rule Enter & Exit Rule
avg_n_bets_per_year 243.272727 105.000000
win_ratio 0.518835 0.590988
annualized_sharpe_ratio 0.537080 1.800316

최적 청산 규칙 파라미터

#dynamic target rule
# different maximum holding
close_ = market_df.close['2009':'2020']
rolling = np.arange(20,260,10)
mhs = [20,60,120,260]
win_ratios = pd.DataFrame()

for mh in mhs:
    max_holding = [mh, 0]
    dynamic_targets = []
    for j in rolling:
        for i in range(2,j+1):
            changes = close_.pct_change(1).to_frame()
            changes = changes.join(close_.pct_change(i).rename('close {}'.format(i)))
        dynamic_target = changes.abs().dropna().mean(axis=1)['2010':]
        dynamic_targets.append(dynamic_target)

    barrier_exit_list_rolling=[]
    for i in range(len(dynamic_targets)):
        barrier_exit_list_rolling.append(get_barrier(close, enter_list[1], [1,1], max_holding, target = dynamic_targets[i]))  #dynamic  

    rts_exit_list=[]
    for i in range(len(barrier_exit_list_rolling)):
        rts_exit_list.append(make_rt(close,barrier_exit_list_rolling[i].dropna()))

    result_df = pd.concat([round_trip.get_df_ann_sr(rts_bm,'No Rule')], axis=1)
    for i in range(len(rts_exit_list)):
        result_df = result_df.join(round_trip.get_df_ann_sr(rts_exit_list[i],'{}'.format(rolling[i])))

    win_ratios['Max. holding {} days'.format(mh)] = result_df.T.win_ratio
win_ratios
Max. holding 20 days Max. holding 60 days Max. holding 120 days Max. holding 260 days
No Rule 0.518835 0.518835 0.518835 0.518835
20 0.541054 0.576857 0.572169 0.575130
30 0.564014 0.573898 0.570069 0.570441
40 0.532872 0.559689 0.553155 0.550562
50 0.555363 0.588591 0.589455 0.587727
60 0.564991 0.605195 0.605195 0.603463
70 0.562392 0.605195 0.600866 0.595671
80 0.557192 0.600000 0.584416 0.586147
90 0.555459 0.599134 0.598787 0.601732
100 0.563258 0.587522 0.600520 0.599653
110 0.555459 0.575022 0.595486 0.604510
120 0.561525 0.585281 0.611785 0.620451
130 0.560659 0.593560 0.608014 0.620209
140 0.559792 0.601739 0.607485 0.623151
150 0.555844 0.579496 0.578261 0.597391
160 0.545927 0.574413 0.581010 0.600174
170 0.558925 0.598432 0.598082 0.625981
180 0.551127 0.586297 0.594266 0.615451
190 0.548918 0.594805 0.601386 0.629116
200 0.548918 0.591342 0.606586 0.623050
210 0.535065 0.596886 0.612987 0.627706
220 0.555844 0.614187 0.628571 0.643290
230 0.552768 0.595506 0.605195 0.624567
240 0.561525 0.618387 0.629887 0.647569
250 0.559792 0.609714 0.625543 0.644097 
plt.figure(figsize=(15,6))
plt.title("Exit rules")
plt.plot(win_ratios)
plt.legend(win_ratios)
plt.ylabel('win ratio')
plt.xlabel('Rolling days for calculating target rate')
plt.show()

매매 결과

barrier = get_barrier(close, enter_list[1], [1,1], max_holding, target = dynamic_target)
rts = make_rt(close,barrier.dropna())
round_trip.get_df_ann_sr(rts,'Chose for 2nd model')
Chose for 2nd model
avg_n_bets_per_year 104.818182
win_ratio 0.644097
annualized_sharpe_ratio 3.122372 
print_round_trip_stats(rts)
Summary stats All trades Long trades
Total number of round_trips 1153.00 1153.00
Percent profitable 0.64 0.64
Winning round_trips 742.00 742.00
Losing round_trips 410.00 410.00
Even round_trips 1.00 1.00
PnL stats All trades Long trades
Total profit $1562060.00 $1562060.00
Gross profit $2900830.00 $2900830.00
Gross loss $-1338770.00 $-1338770.00
Profit factor $2.17 $2.17
Avg. trade net profit $1354.78 $1354.78
Avg. winning trade $3909.47 $3909.47
Avg. losing trade $-3265.29 $-3265.29
Ratio Avg. Win:Avg. Loss $1.20 $1.20
Largest winning trade $15720.00 $15720.00
Largest losing trade $-19350.00 $-19350.00
Duration stats All trades Long trades
Avg duration 98 days 20:41:25.342584562 98 days 20:41:25.342584562
Median duration 51 days 00:00:00 51 days 00:00:00
Longest duration 264 days 00:00:00 264 days 00:00:00
Shortest duration 1 days 00:00:00 1 days 00:00:00
Return stats All trades Long trades
Avg returns all round_trips 0.22% 0.22%
Avg returns winning 1.38% 1.38%
Avg returns losing -1.22% -1.22%
Median returns all round_trips 0.18% 0.18%
Median returns winning 1.06% 1.06%
Median returns losing -0.99% -0.99%
Largest winning trade 8.69% 8.69%
Largest losing trade -9.44% -9.44%
Symbol stats Asset
Avg returns all round_trips 0.22%
Avg returns winning 1.38%
Avg returns losing -1.22%
Median returns all round_trips 0.18%
Median returns winning 1.06%
Median returns losing -0.99%
Largest winning trade 8.69%
Largest losing trade -9.44% 
barrier.to_csv('C:data/barrier.csv')
barrier_bm.to_csv('C:data/barrier_bm.csv')