vectorbt - 분할매수 + MA Gap ZScore
1Pager
Hypothesis
1.현재 가격과 이동평균선 사이의 Gap이 존재한다.
- 그 Gap에 대한 데이터를 샘플링 한다.
- Gap데이터를 모아서 평균 및 표준정규분포 곡선을 그린다.
- 그 정규분포곡선에서 하위 1%일때가 매수하기 좋은 시점이다.
- 의미 : 지금까지의 Period 중 가장 저가에 매수하는 것
주의
- 상승장 only
- 충분히 학습기간이 필요하기에 10일 이상 매매를 금지한다. (초기 데이터의 이상치 제거 필요)
- 미래 데이터를 미리 가져와서 평균치를 내면 안되므로 expanding 함수를 이용한다.
2.분할 매수를 한다.
- 시그널 이후 진입시 size값을 이용해서 진입한다.
3.가중 분할 매수를 한다.
- MA12 가중치 2
- MA24 가중치 8
4.분할매수 Level1적용한다.
- 최소 10번에 나누어서 매수
- 각 시행마다 가중치 매수를 진행한다.
Long Position
- MA12 ZScore 하위 1%
- MA12 ZScore 하위 1%
Short Position
long only
- stoploss 30%
Result - BTC/USDT, 1h
import numpy as np
import pandas as pd
import ccxt
import ta
import vectorbt as vbt
import quantstats as qs
import warnings
import time
from datetime import datetime, timedelta
import matplotlib.pyplot as plt
from scipy.stats import norm
# 데이터 호출 함수
def getHourlyOHLCV(days=100, ticker='BTC/USDT'):
startSince = datetime.now() - timedelta(days)
dfOHLCV = pd.DataFrame(columns=['Time', 'Open', 'High', 'Low', 'Close', 'Volume'])
since = int(startSince.timestamp() * 1000)
prevSince = None
while True:
data = ccxt.upbit().fetch_ohlcv(ticker, '1h', since)
df = pd.DataFrame(data, columns=['Time', 'Open', 'High', 'Low', 'Close', 'Volume'])
df['Time'] = pd.to_datetime(df['Time'], unit='ms') # ms(long) > datetime
dfOHLCV = pd.concat([dfOHLCV, df], ignore_index=True)
since = int(dfOHLCV['Time'].iloc[-1].timestamp() * 1000) + 1 # +1ms for no dedup
if since != None and prevSince == since:
break
prevSince = since
time.sleep(0.1)
dfOHLCV = dfOHLCV.drop_duplicates(subset=['Time']) # dedup
dfOHLCV = dfOHLCV.set_index('Time').resample('1H').ffill() # backfill prev data
return dfOHLCV
# 이상값 제거
def remove_outliers(series):
rev_range = 3 # 제거 범위 조절 변수
level_1q, level_3q = series.quantile(0.25), series.quantile(0.75)
IQR = level_3q - level_1q
return series[(series <= level_3q + (rev_range * IQR)) & (series >= level_1q - (rev_range * IQR))]
# ---- layer:data
dfOHLCV = getHourlyOHLCV(days=100,ticker="LINK/KRW")
# ---- layer:indicator
df = dfOHLCV.copy()
df['RowNum'] = df.reset_index().index
df['MA4'] = ta.trend.SMAIndicator(df['Close'],4).sma_indicator()
df['MA12'] = ta.trend.SMAIndicator(df['Close'], 12).sma_indicator()
df['MA24'] = ta.trend.SMAIndicator(df['Close'], 24).sma_indicator()
df['MA72'] = ta.trend.SMAIndicator(df['Close'], 72).sma_indicator()
# zscore
df['MA12Gap'] = df['Close'] - df['MA12']
df['MA12ZScore'] = (df['MA12Gap'] - df['MA12Gap'].expanding().mean()) / df['MA12Gap'].expanding().std()
df['MA12ZScore'] = remove_outliers(df['MA12ZScore'])
# ---- layer:signal
# long only
df['SignalMA12ZScore'] = np.select( [ df['MA12ZScore'] < df['MA12ZScore'].quantile(0.01), df['MA12ZScore'] > df['MA12ZScore'].quantile(0.99), ],[1,0],0 )
# ---- layer:position
cond_go = [ (df['SignalMA12ZScore'] == 1) & (df['RowNum'] > 12*10), df['SignalMA12ZScore'] == -1 ]
df['go'] = np.select(cond_go, [1,-1], 0)
df['go'] = df['go'].mask(df['go'] == df['go'].shift(1), 0) # 최초값유지
df['size'] = (df[['go']].expanding().sum()*10).abs() # 분할매수 사이즈
# ---- layer:anomalo
plt.figure(figsize=(25,5))
plt.grid(True)
plt.hist([df['MA12ZScore']],bins=300,label=['a'])
plt.legend()
plt.show()
benchmarkTotlaReturns = ((df['Close'].iloc[-1] - df['Close'].iloc[0]) / df['Close'].iloc[0])*100
lastSize = df['size'][-1]
countGoTrue = df['go'].value_counts()[1]
countGoHold = df['go'].value_counts()[0]
countSignalMA12ZScoreTrue = df['SignalMA12ZScore'].value_counts()[1]
countSignalMA12ZScoreFalse = df['SignalMA12ZScore'].value_counts()[0]
print(f"countGoTrue {countGoTrue}")
print(f"countGoHold {countGoHold}")
print(f"countSignalMA12ZScoreTrue {countSignalMA12ZScoreTrue}")
print(f"countSignalMA12ZScoreFalse {countSignalMA12ZScoreFalse}")
print(f"benchmarkTotlaReturns {benchmarkTotlaReturns}")
print(f"lastSize {lastSize}")
# layer:trading
pf = vbt.Portfolio.from_signals(
df['Close'],
entries = (df['go']==1), # long 포지션 진입
exits = (df['go']==-1), # long 포지션 탈출
sl_stop = 0.3, # stop loss, 10% 손해 > 손절
sl_trail = True, # 트레일링 스탑
fees = 0.0006, # 거래수수료 ( bybit ) 주문시 post-only
# allow_partial = True,
size_type = 1, # Value 1 | Percent 2 | TargetValue 4
size = df['size'],
log=True,
freq='1H',
accumulate = True # 분할 매수 가능하게, 기본값=False > 매수 후 Hold
# post-only True -> 무조건 지정한 가격에만 체결 / 수수료 낮게 지불 가능 / 실수해서 손해보는 경우가 없다. (시장가 체결 되는 케이스)
# tp_stop = 0.09, # take profit, 9% 이익 > 익절
# short_entries = (df['go']==-1), # short 포지션 진입
# short_exits = (df['go']==1) # short 포지션 탈출
)
print(pf.stats())
# layer:visualize
pf.plot().show()
# layer:report
pf.qs.plot_snapshot()
qs.plots.snapshot(pf.benchmark_returns())
# layer:notice
buyNotice = (df.tail(4)['go'] == 1).any()
stats
Start 2023-11-25 08:00:00
End 2024-03-04 07:00:00
Period 100 days 00:00:00
Start Value 100.0
End Value 168.937901
Total Return [%] 68.937901
Benchmark Return [%] 68.358269
Max Gross Exposure [%] 100.0
Total Fees Paid 0.322645
Max Drawdown [%] 19.575762
Max Drawdown Duration 30 days 12:00:00
Total Trades 3
Total Closed Trades 2
Total Open Trades 1
Open Trade PnL 64.883777
Win Rate [%] 50.0
Best Trade [%] 14.855785
Worst Trade [%] -9.403154
Avg Winning Trade [%] 14.855785
Avg Losing Trade [%] -9.403154
Avg Winning Trade Duration 47 days 14:00:00
Avg Losing Trade Duration 10 days 12:00:00
Profit Factor 1.375634
Expectancy 2.027062
Sharpe Ratio 2.920345
Calmar Ratio 29.524251
Omega Ratio 1.09474
Sortino Ratio 4.242156
Result - LINK/KRW, 1h - TC2
매도 : -10% drawdown 손실
stats
benchmark total return 43.658413340070744
Start 2023-11-25 08:00:00
End 2024-03-04 07:00:00
Period 100 days 00:00:00
Start Value 100.0
End Value 134.324584
Total Return [%] 34.324584
Benchmark Return [%] 43.658413
Max Gross Exposure [%] 100.0
Total Fees Paid 0.888398
Max Drawdown [%] 24.620326
Max Drawdown Duration 54 days 06:00:00
Total Trades 8
Total Closed Trades 7
Total Open Trades 1
Open Trade PnL 10.461915
Win Rate [%] 42.857143
Best Trade [%] 25.874756
Worst Trade [%] -9.663215
Avg Winning Trade [%] 17.630381
Avg Losing Trade [%] -6.444443
Avg Winning Trade Duration 13 days 10:20:00
Avg Losing Trade Duration 7 days 15:15:00
Profit Factor 1.949065
Expectancy 3.408953
Sharpe Ratio 1.879882
Calmar Ratio 7.863738
Omega Ratio 1.064586
Sortino Ratio 2.685104
dtype: object
Result - LINK/KRW, 1h - TC3
매도 : -20% drawdown 손실
stats
Start 2023-11-25 08:00:00
End 2024-03-04 07:00:00
Period 100 days 00:00:00
Start Value 100.0
End Value 129.473206
Total Return [%] 29.473206
Benchmark Return [%] 43.759474
Max Gross Exposure [%] 100.0
Total Fees Paid 0.265968
Max Drawdown [%] 30.675143
Max Drawdown Duration 55 days 07:00:00
Total Trades 3
Total Closed Trades 2
Total Open Trades 1
Open Trade PnL 48.190837
Win Rate [%] 0.0
Best Trade [%] -9.617927
Worst Trade [%] -10.079824
Avg Winning Trade [%] NaN
Avg Losing Trade [%] -9.848875
Avg Winning Trade Duration NaT
Avg Losing Trade Duration 16 days 20:00:00
Profit Factor 0.0
Expectancy -9.358816
Sharpe Ratio 1.638765
Calmar Ratio 5.108964
Omega Ratio 1.053357
Sortino Ratio 2.355005
Result - LINK/KRW, 1h - TC3
매도 : -10% drawdown 손실
stats
Result - LINK/KRW, 1h - TC4
매도 : -30% drawdown 손실 매집 : df['MA12ZScore'] < df['MA12ZScore'].quantile(0.01)
stats
Start 2023-11-25 09:00:00
End 2024-03-04 08:00:00
Period 100 days 00:00:00
Start Value 100.0
End Value 135.95403
Total Return [%] 35.95403
Benchmark Return [%] 43.574297
Max Gross Exposure [%] 100.0
Total Fees Paid 0.059964
Max Drawdown [%] 25.721672
Max Drawdown Duration 35 days 13:00:00
Total Trades 1
Total Closed Trades 0
Total Open Trades 1
Open Trade PnL 35.95403
Win Rate [%] NaN
Best Trade [%] NaN
Worst Trade [%] NaN
Avg Winning Trade [%] NaN
Avg Losing Trade [%] NaN
Avg Winning Trade Duration NaT
Avg Losing Trade Duration NaT
Profit Factor NaN
Expectancy NaN
Sharpe Ratio 1.880543
Calmar Ratio 8.04062
Omega Ratio 1.060568
Sortino Ratio 2.71913
dtype: object
Result - LINK/KRW, 1h - TC5
매도 : -30% drawdown 손실
매집 : df['MA12ZScore'] < df['MA12ZScore'].quantile(0.01) , 2
매집 : df['MA24ZScore'] < df['MA24ZScore'].quantile(0.01) , 8
# ---- layer:indicator
df = dfOHLCV.copy()
df['RowNum'] = df.reset_index().index
df['MA4'] = ta.trend.SMAIndicator(df['Close'],4).sma_indicator()
df['MA12'] = ta.trend.SMAIndicator(df['Close'], 12).sma_indicator()
df['MA24'] = ta.trend.SMAIndicator(df['Close'], 24).sma_indicator()
df['MA72'] = ta.trend.SMAIndicator(df['Close'], 72).sma_indicator()
# zscore
df['MA12Gap'] = df['Close'] - df['MA12']
df['MA12ZScore'] = (df['MA12Gap'] - df['MA12Gap'].expanding().mean()) / df['MA12Gap'].expanding().std()
df['MA12ZScore'] = remove_outliers(df['MA12ZScore'])
df['MA24Gap'] = df['Close'] - df['MA24']
df['MA24ZScore'] = (df['MA24Gap'] - df['MA24Gap'].expanding().mean()) / df['MA24Gap'].expanding().std()
df['MA24ZScore'] = remove_outliers(df['MA24ZScore'])
# ---- layer:signal
# long only
df['SignalMA12ZScore'] = np.select( [ df['MA12ZScore'] < df['MA12ZScore'].quantile(0.01), df['MA12ZScore'] > df['MA12ZScore'].quantile(0.99), ],[1,0],0 )
df['SignalMA24ZScore'] = np.select( [ df['MA24ZScore'] < df['MA24ZScore'].quantile(0.01), df['MA24ZScore'] > df['MA24ZScore'].quantile(0.99), ],[1,0],0 )
# ---- layer:position
cond_go = [ (df['RowNum'] > 12*10) & (df['SignalMA12ZScore'] == 1) , df['SignalMA12ZScore'] == -1 ]
df['go1'] = np.select(cond_go, [1,-1], 0)
df['go1'] = df['go1'].mask(df['go1'] == df['go1'].shift(1), 0) # 최초값유지
cond_go = [ (df['RowNum'] > 12*10) & (df['SignalMA24ZScore'] == 1) , df['SignalMA12ZScore'] == -1 ]
df['go2'] = np.select(cond_go, [1,-1], 0)
df['go2'] = df['go2'].mask(df['go2'] == df['go2'].shift(1), 0) # 최초값유지
result_df = pd.concat([(df[['go1']].expanding().sum()*2).abs(), (df[['go2']].expanding().sum()*8).abs()], axis=1 )
df['go'] = df['go1'] & df['go2']
df['size'] = result_df.sum(axis=1)
# df['size'] = (df[['go1']].expanding().sum()*1).abs() + (df[['go2']].expanding().sum()*5).abs() # 분할매수 사이즈
# ---- layer:anomalo
plt.figure(figsize=(25,5))
plt.grid(True)
plt.hist([df['MA12ZScore']],bins=300,label=['a'])
plt.legend()
plt.show()
benchmarkTotlaReturns = ((df['Close'].iloc[-1] - df['Close'].iloc[0]) / df['Close'].iloc[0])*100
lastSize = df['size'][-1]
countGoTrue = df['go'].value_counts()[1]
countGoHold = df['go'].value_counts()[0]
countSignalMA12ZScoreTrue = df['SignalMA12ZScore'].value_counts()[1]
countSignalMA12ZScoreFalse = df['SignalMA12ZScore'].value_counts()[0]
print(f"countGoTrue {countGoTrue}")
print(f"countGoHold {countGoHold}")
print(f"countSignalMA12ZScoreTrue {countSignalMA12ZScoreTrue}")
print(f"countSignalMA12ZScoreFalse {countSignalMA12ZScoreFalse}")
print(f"benchmarkTotlaReturns {benchmarkTotlaReturns}")
print(f"lastSize {lastSize}")
# layer:trading
pf = vbt.Portfolio.from_signals(
df['Close'],
entries = (df['go']>=1), # long 포지션 진입
exits = (df['go']<=-1), # long 포지션 탈출
sl_stop = 0.3, # stop loss, 10% 손해 > 손절
sl_trail = True, # 트레일링 스탑
fees = 0.0006, # 거래수수료 ( bybit ) 주문시 post-only
# allow_partial = True,
size_type = 1, # Value 1 | Percent 2 | TargetValue 4
size = df['size'],
log=True,
freq='1H',
accumulate = True # 분할 매수 가능하게, 기본값=False > 매수 후 Hold
# post-only True -> 무조건 지정한 가격에만 체결 / 수수료 낮게 지불 가능 / 실수해서 손해보는 경우가 없다. (시장가 체결 되는 케이스)
# tp_stop = 0.09, # take profit, 9% 이익 > 익절
# short_entries = (df['go']==-1), # short 포지션 진입
# short_exits = (df['go']==1) # short 포지션 탈출
)
print(pf.stats())
# layer:visualize
pf.plot().show()
# layer:report
pf.qs.plot_snapshot()
qs.plots.snapshot(pf.benchmark_returns())
# layer:notice
buyNotice = (df.tail(4)['go'] == 1).any()
stats
lastSize 108.0
Start 2023-11-25 09:00:00
End 2024-03-04 08:00:00
Period 100 days 00:00:00
Start Value 100.0
End Value 144.534144
Total Return [%] 44.534144
Benchmark Return [%] 43.574297
Max Gross Exposure [%] 100.0
Total Fees Paid 0.059964
Max Drawdown [%] 17.492328
Max Drawdown Duration 17 days 01:00:00
Total Trades 1
Total Closed Trades 0
Total Open Trades 1
Open Trade PnL 44.534144
Win Rate [%] NaN
Best Trade [%] NaN
Worst Trade [%] NaN
Avg Winning Trade [%] NaN
Avg Losing Trade [%] NaN
Avg Winning Trade Duration NaT
Avg Losing Trade Duration NaT
Profit Factor NaN
Expectancy NaN
Sharpe Ratio 2.563104
Calmar Ratio 16.213507
Omega Ratio 1.091704
Sortino Ratio 3.855461