def create_bollinger_bands(prices):
bollUPPER, bollMIDDLE, bollLOWER = BBANDS(prices,
timeperiod=20,
nbdevup=2.,
nbdevdn=2.,
matype=0)
upper_bolling_percent = np.reshape(bollUPPER / prices, (len(prices), 1))
lower_bolling_percent = np.reshape(bollLOWER / prices, (len(prices), 1))
bollinger_bands = np.hstack((upper_bolling_percent, lower_bolling_percent))
for time_period in range(10, 31):
for up in range(18, 23):
for down in range(18, 23):
upper, middle, lower = BBANDS(prices,
timeperiod=time_period,
nbdevup=float(up) / 10,
nbdevdn=float(down) / 10,
matype=0)
bollinger_bands = np.hstack(
(bollinger_bands,
np.reshape(upper / prices, (len(prices), 1)),
np.reshape(lower / prices, (len(prices), 1))))
return bollinger_bands
def minwidth_BBANDS(p, ma_ts, sigma, mid_width):
"""
Custom technical indicator:
Bollinger bands, with a user-defined minimum width.
"""
from talib import BBANDS
import numpy as np
# Obtain indicators
(t_b, m_o, b_b) = BBANDS(p.values,
timeperiod=ma_ts,
nbdevup=sigma,
nbdevdn=sigma)
# Impose a minimum width on the bollinger bands - can't use nan values as clips
t_m = np.isfinite(
t_b * m_o) # top band mask: neither top band nor moving av can be nan
b_m = np.isfinite(
b_b * m_o) # bot band mask: neither bot band nor moving av can be nan
# Clip bands
t_b[t_m] = np.clip(t_b[t_m], m_o[t_m] + mid_width, np.inf)
b_b[b_m] = np.clip(b_b[b_m], -np.inf, m_o[b_m] - mid_width)
return t_b, m_o, b_b
def analyze(self,
historical_data,
signal=['close'],
hot_thresh=None,
cold_thresh=None,
period_count=20,
std_dev=2):
"""Check when close price cross the Upper/Lower bands.
Args:
historical_data (list): A matrix of historical OHCLV data.
period_count (int, optional): Defaults to 20. The number of data points to consider for the BB bands indicator.
signal (list, optional): Defaults to close. Unused in this indicator
std_dev (int, optional): number of std dev to use. Common values are 2 or 1
Returns:
pandas.DataFrame: A dataframe containing the indicator and hot/cold values.
"""
dataframe = self.convert_to_dataframe(historical_data)
#Required to avoid getting same values for low, middle, up
dataframe['close_10k'] = dataframe['close'] * 10000
up_band, mid_band, low_band = BBANDS(dataframe['close_10k'],
timeperiod=period_count,
nbdevup=std_dev,
nbdevdn=std_dev,
matype=0)
bollinger = pandas.concat([dataframe, up_band, mid_band, low_band],
axis=1)
bollinger.rename(columns={
0: 'up_band',
1: 'mid_band',
2: 'low_band'
},
inplace=True)
old_up, old_low = bollinger.iloc[-2]['up_band'], bollinger.iloc[-2][
'low_band']
cur_up, cur_low = bollinger.iloc[-1]['up_band'], bollinger.iloc[-1][
'low_band']
old_close = bollinger.iloc[-2]['close_10k']
cur_close = bollinger.iloc[-1]['close_10k']
bollinger['is_hot'] = False
bollinger['is_cold'] = False
bollinger['is_hot'].iloc[
-1] = old_low < old_close and cur_low > cur_close
bollinger['is_cold'].iloc[
-1] = old_up > old_close and cur_up < cur_close
bollinger['up_band'] = bollinger['up_band'] / 10000
bollinger['mid_band'] = bollinger['mid_band'] / 10000
bollinger['low_band'] = bollinger['low_band'] / 10000
return bollinger
def WADDAH_ATTAR_EXPLOSION(close, high, low, sensitive = 150, fast_period=20, slow_period = 40, channel_period = 20, channel_mult = 2, dead_zone=30):
from talib import MACD
from talib import BBANDS
from talib import ATR
from talib import WMA
macd, macdsignal, macdhist = MACD(close, fastperiod=fast_period, slowperiod=slow_period, signalperiod=9)
upperband, middleband, lowerband = BBANDS(close, timeperiod=channel_period, nbdevup=channel_mult, nbdevdn=channel_mult, matype=0)
ind_trend1 = np.full(len(close), np.nan)
ind_itrend1 = np.full(len(close), np.nan)
ind_explo1 = np.full(len(close), np.nan)
tr = WMA(ATR(high, low, close, 20),3)
ind_dead = tr*dead_zone / 10
for i in range(0,len(close)):
if(i<2):
continue
trend1 = (macd[i] - macd[i-1]) * sensitive;
trend2 = (macd[i-1] - macd[i-2]) * sensitive;
explo1 = (upperband[i] - lowerband[i])
#explo2 = (upperband[i-1] - lowerband[i-1])
if(trend1>=0):
ind_trend1[i]=trend1
ind_itrend1[i]=0
if(trend1<0):
ind_trend1[i]=0
ind_itrend1[i]=(trend1*-1)
ind_explo1[i] = explo1
#print(str(i)+"\t "+str(close[i])+"\t "+str(close[i])+"\t "+str(ind_trend1[i])+"\t"+str(ind_itrend1[i]))
return ind_trend1, ind_itrend1, ind_explo1, ind_dead
def refit(self, ohlcv, params=None, ba=None, **kwargs):
"""
refit the strategy using given parameters and return signals
:param ohlcv: a DataFrame object with OHLCV columns ordered by date, ascending
:param params: optimal parameters
:ba: bid ask spread
:return: signal
"""
required_params = ['trailing_window', 'ema_s', 'ema_l', 'bb']
if not all(param in params for param in required_params):
raise KeyError("incorrect parameters")
lookback = params['trailing_window']
close = ohlcv['close']
ema_s = EMA(close[-lookback:], timeperiod=params['ema_s'])
ema_l = EMA(close[-lookback:], timeperiod=params['ema_l'])
bb = BBANDS(close[-lookback:], timeperiod=params['bb'])
if ba in None:
buy_signal = (ema_s > ema_l) and (close[-1] >
(bb[1])) and (close[-1] > ema_s)
else:
bid = ba['bid']
ask = ba['ask']
mid = (bid[-1] + ask[-1]) / 2
buy_signal = (ema_s > ema_l) and (mid > (bb[1])) and (mid > ema_s)
return {'buy': buy_signal, 'sell': None}
def bbp(price):
up, mid, low = BBANDS(close,
timeperiod=200,
nbdevup=2,
nbdevdn=2,
matype=0)
bbp = (price['AdjClose'] - low) / (up - low)
return bbp
def add_technicals(price, close):
price["RSI"] = RSI(close, timeperiod=14)
price["BBP"] = bbp(price)
price["BB_up"], price["BB_mid"], price["BB_low"] = BBANDS(close,
timeperiod=20,
nbdevup=2,
nbdevdn=2,
matype=0)
return price
def process_data(self, data):
r = None
try:
r = BBANDS(np.array(data['close']), timeperiod=self.period, nbdevup=self.nbdevup, nbdevdn=self.nbdevdn, matype=self.matype)
except Exception as e:
print("Exception in BollingerBands", e)
logger.exception(e)
r = {self.k: [0,0]}
self.setd(0 if np.isnan(r[self.k][-1]) else r[self.k][-1])
return self
def bbp():
up, mid, low = BBANDS(close,
timeperiod=20,
nbdevup=2.5,
nbdevdn=2.5,
matype=8)
# up, mid, low = BBANDS(close, timeperiod=20, nbdevup=2, nbdevdn=2, matype=0)
# bbp = (price['Close'] - low) / (up - low)
bbp = (close - low) / (up - low)
return bbp, up, mid, low
async def calc_pullback(self) -> None:
for i, row in self.portfolio.iterrows():
# print(row.symbol, len(self.data_bars[row.symbol]))
bband = BBANDS(
self.data_bars[row.symbol].close,
timeperiod=7,
nbdevdn=1,
nbdevup=1,
matype=MA_Type.EMA,
)
lower_band = bband[2]
upper_band = bband[0]
volatility = upper_band[-1] - lower_band[-1]
position = 0
profit = 0.0
count = 0
for i in range(-self.pullback_days, -1, 1):
if (not position and self.data_bars[row.symbol].close[i - 1] <
lower_band[i - 1]
and self.data_bars[row.symbol].close[i] >
self.data_bars[row.symbol].close[i - 1] and
self.data_bars[row.symbol].close[i] > lower_band[i]):
position = (self.portfolio_size //
self.data_bars[row.symbol].open[i])
profit -= position * self.data_bars[row.symbol].open[i]
count += 1
elif (position and
self.data_bars[row.symbol].open[i] > upper_band[i - 1]):
profit += position * self.data_bars[row.symbol].open[i]
position = 0
self.portfolio.loc[self.portfolio.symbol == row.symbol,
"count"] = count
self.portfolio.loc[self.portfolio.symbol == row.symbol,
"profit"] = profit
self.portfolio.loc[self.portfolio.symbol == row.symbol,
"volatility"] = volatility
qty = int(count * self.risk_factor / volatility)
self.portfolio.loc[self.portfolio.symbol == row.symbol,
"qty"] = qty
self.portfolio.loc[self.portfolio.symbol == row.symbol,
"est"] = (qty *
self.data_bars[row.symbol].close[-1])
self.portfolio.loc[
self.portfolio.symbol == row.symbol, "opt"] = bool(
self.data_bars[row.symbol].close[-1] < lower_band[-1])
self.portfolio = self.portfolio.loc[(self.portfolio.profit > 0)
& (self.portfolio.opt == True)
& (self.portfolio.volatility > 2.0)
& (self.portfolio.score > 35.0)]
def init(self):
super().init()
up_band, mid, down_band = BBANDS(self.data.Close,
timeperiod=self.ma_len,
nbdevup=self.band_width,
nbdevdn=self.band_width,
matype=0)
self.up_band = self.I(lambda x: up_band, 'up_band')
self.mid = self.I(lambda x: mid, 'mid')
self.down_band = self.I(lambda x: down_band, 'down_band')
def get_bbands_score(self, product_df):
upper_df, middle_df, lower_df = BBANDS(product_df.close, 20, 2, 2)
close = product_df.close.iloc[-1]
upper = upper_df.iloc[-1]
middle = middle_df.iloc[-1]
lower = lower_df.iloc[-1]
#print(f"Upper: {upper}, Middle: {middle}, Lower: {lower}")
if close < lower:
return 1
elif close > upper:
return -1
else:
return 0
def analyze(self,
historical_data,
signal=['bbp'],
hot_thresh=0,
cold_thresh=0.8,
period_count=20,
std_dev=2):
"""Check when close price cross the Upper/Lower bands.
Args:
historical_data (list): A matrix of historical OHCLV data.
period_count (int, optional): Defaults to 20. The number of data points to consider for the BB bands indicator.
signal (list, optional): Defaults bbp value.
hot_thresh (float, optional): Defaults to 0. The threshold at which this might be
good to purchase.
cold_thresh (float, optional): Defaults to 0.8. The threshold at which this might be
good to sell.
std_dev (int, optional): number of std dev to use. Common values are 2 or 1
Returns:
pandas.DataFrame: A dataframe containing the indicator and hot/cold values.
"""
dataframe = self.convert_to_dataframe(historical_data)
mfi = abstract.MFI(dataframe, period_count=14)
# Required to avoid getting same values for low, middle, up
dataframe['close_10k'] = dataframe['close'] * 10000
up_band, mid_band, low_band = BBANDS(dataframe['close_10k'],
timeperiod=period_count,
nbdevup=std_dev,
nbdevdn=std_dev,
matype=0)
bbp = (dataframe['close_10k'] - low_band) / (up_band - low_band)
bollinger = pandas.concat([dataframe, bbp, mfi], axis=1)
bollinger.rename(columns={0: 'bbp', 1: 'mfi'}, inplace=True)
bollinger['is_hot'] = False
bollinger['is_cold'] = False
bollinger['is_hot'].iloc[
-1] = bollinger['bbp'].iloc[-2] <= hot_thresh and bollinger[
'bbp'].iloc[-2] < bollinger['bbp'].iloc[-1]
bollinger['is_cold'].iloc[
-1] = bollinger['bbp'].iloc[-1] >= cold_thresh
return bollinger
def get_indicators(df, indicators=[]):
if "bbands" in indicators:
df["upperband"], df["middleband"], df["lowerband"] = BBANDS(
df["close"], timeperiod=20, nbdevup=2, nbdevdn=2, matype=0)
if "macd" in indicators:
df["macd"], df["macdsignal"], df["macdhist"] = MACD(df["close"],
fastperiod=12,
slowperiod=26,
signalperiod=9)
if "price_logreturn" in indicators:
df["price_logreturn"] = log_return(df["close"].to_numpy())
if "volume_logreturn" in indicators:
df["volume_logreturn"] = log_return(df["volume"].to_numpy())
return df
def bbww(price):
multiplier = 1000000
# Bollinger Band Width = (Upper Band - Lower Band) / Simple Moving Average for the same period.
up, mid, low = BBANDS(price['Close'].values * multiplier,
timeperiod=5,
nbdevup=2,
nbdevdn=2,
matype=talib.MA_Type.T3)
# up, mid, low = up * 10000000, mid * 10000000, low * 10000000
sma = SMA(price['Close'].values, timeperiod=14) * multiplier
bbw = (up - low) / (sma)
return bbw
def get_technical_analysis(stock_data: pd.DataFrame,
stock_name: str) -> pd.DataFrame:
data = stock_data.query(f'stock_name=="{stock_name}"')
close_prices = data['Adj Close'].values
up, mid, low = BBANDS(close_prices,
timeperiod=20,
nbdevup=2,
nbdevdn=2,
matype=0)
rsi = RSI(close_prices, timeperiod=14)
bbp = (data['Adj Close'] - low) / (up - low)
data['lower_bound'] = low
data['upper_bound'] = up
data['RSI'] = rsi
data['BBP'] = bbp
return data
async def handle_sell_side(
self,
symbols_position: Dict[str, float],
data_loader: DataLoader,
now: datetime,
trade_fee_precentage: float,
) -> Dict[str, Dict]:
actions = {}
for symbol, position in symbols_position.items():
if position == 0:
continue
current_price = data_loader[symbol].close[now]
resampled_close = self.calc_close(symbol, data_loader, now)
bband = BBANDS(
resampled_close,
timeperiod=7,
nbdevdn=1,
nbdevup=1,
matype=MA_Type.EMA,
)
yesterday_upper_band = bband[0][-2]
print(
f"\ncurrent_price > yesterday_upper_band : {current_price > yesterday_upper_band}({current_price} < {yesterday_upper_band})"
)
if current_price > yesterday_upper_band:
sell_indicators[symbol] = {
"upper_band": bband[0][-2:].tolist(),
"lower_band": bband[2][-2:].tolist(),
}
tlog(
f"[{self.name}][{now}] Submitting sell for {position} shares of {symbol} at market"
)
tlog(f"indicators:{sell_indicators[symbol]}")
actions[symbol] = {
"side": "sell",
"qty": str(position),
"type": "limit",
"limit_price": str(current_price),
}
return actions
def bbp(price):
# bolinger bands %
multiplier = 1000000
up, mid, low = BBANDS(price['Close'].values * multiplier,
timeperiod=5,
nbdevup=2,
nbdevdn=2,
matype=talib.MA_Type.T3)
# multiply to avoid very small numbers leading to inf
# up, mid, low = up * mutiplier, mid * mutiplier, low * mutiplier
bbp = (((price['Close'] * multiplier) - low) / (up - low))
if isNaN(bbp.values[-1]):
print("ok")
return bbp
def handle_data(context, data):
required_params = ['trailing_window', 'ema_s', 'ema_l', 'bb']
if not all(param in context.params
for param in required_params):
raise KeyError("incorrect parameter list")
trailing_window = data.history(
context.asset, 'close', context.params['trailing_window'],
'1d')
if trailing_window.isnull().values.any():
return
ema_s = EMA(trailing_window.values,
timeperiod=context.params['ema_s'])
ema_l = EMA(trailing_window.values,
timeperiod=context.params['ema_l'])
bb = BBANDS(trailing_window.values,
timeperiod=context.params['bb'])
buy = False
sell = False
buy_signal = (ema_s[-1] > ema_l[-1]) and (
trailing_window.values[-1] >
(bb[1][-1])) and (trailing_window.values[-1] > ema_s[-1])
if buy_signal and not context.invested:
order(context.asset, 100)
context.invested = True
buy = True
elif not buy_signal and context.invested:
order(context.asset, -100)
context.invested = False
sell = True
record(BTC=data.current(context.asset, "price"),
ema_s=ema_s[-1],
ema_l=ema_l[-1],
bb=bb[1][-1],
buy=buy,
sell=sell)
def create_bollinger_bands(historical_data):
prices = historical_data['Adj Close'].as_matrix()
bollinger_bands = dict()
for time_period in range(10, 31):
for up in range(18, 23):
for down in range(18, 23):
upper, middle, lower = BBANDS(prices,
timeperiod=time_period,
nbdevup=float(up) / 10,
nbdevdn=float(down) / 10,
matype=0)
bollinger_bands['bbands_upper_{}_{}_{}'.format(
time_period, up, down)] = upper / prices
bollinger_bands['bbands_lower_{}_{}_{}'.format(
time_period, up, down)] = lower / prices
macds_dataframe = pd.DataFrame(bollinger_bands)
return historical_data.join(macds_dataframe)
def _calc_indicator(self, OHLCV_input):
"""
Calculates the Bollinger bands technical indicator using a wrapper for the TA-lib
Args:
:param OHLCV_input: the dataframe with the Open, High, Low, Close and Volume values
:type OHLCV_input: pandas DataFrame
Returns:
DataFrame with scaled features with size (n_observations, n_features).
"""
try:
close = OHLCV_input['close'].values[:, 0]
except IndexError:
close = OHLCV_input['close'].values
upperband, middleband, lowerband = BBANDS(close, self.__timeperiod, self.__nbdevup, self.__nbdevdn,
self.__matype)
upperband = DataFrame(upperband)
middleband = DataFrame(middleband)
lowerband = DataFrame(lowerband)
return concat([upperband, middleband, lowerband], axis=1, ignore_index=True)
def main():
file_content = file_io_service.load_historical_data('PETR4.SA')
file_content = file_content.dropna()
close = file_content['Adj Close'].values
date = np.array(
list(
map(lambda x: date_helper.parse_date_to_datetime(x),
file_content['Date'])))
rsi = RSI(close, timeperiod=14)
macd, macdsignal, macdhist = MACD(close,
fastperiod=12,
slowperiod=26,
signalperiod=9)
up, mid, low = BBANDS(close, timeperiod=20, nbdevup=2, nbdevdn=2, matype=0)
fig, (ax0, ax1, ax2) = plt.subplots(3, 1, sharex=True, figsize=(12, 8))
ax0.plot(date, close, label='Close')
ax0.set_xlabel('Date')
ax0.set_ylabel('Close')
ax0.grid()
ax1.plot(date, rsi, label='Close', linewidth=.5)
ax1.set_xlabel('Date')
ax1.set_ylabel('Close')
ax1.grid()
ax2.plot(date, up, label='BB_up', color='BLUE', linewidth=.5)
ax2.plot(date, close, label='AdjClose', color='BLACK', linewidth=.5)
ax2.plot(date, low, label='BB_low', color='RED', linewidth=.5)
ax2.fill_between(date, y1=low, y2=up, color='#adccff', alpha='0.3')
ax2.set_xlabel('Date')
ax2.set_ylabel('Bollinger Bands')
ax2.grid()
fig.tight_layout()
plt.show()
def compute_bb(close):
high, mid, low = BBANDS(close, timeperiod=20)
return pd.DataFrame({'bb_high': high, 'bb_low': low}, index=close.index)
def bbands200(self, df, pattern, patterntype):
df = df.dropna()
#1. Calculate ATR for potential trade
atr = atrcalc.ATRcalc(df)
##b. get BBands
bband = df.head(21)
bband = bband.iloc[1:]
bband = bband.iloc[::-1]
bbandInput = bband['close'].values
upperband, middleband, lowerband = BBANDS(bbandInput,
timeperiod=20,
nbdevup=2,
nbdevdn=2,
matype=0)
##c. 200EMA
emaInput = df.head(200)
emaInput = emaInput.iloc[::-1]
EMAclose = emaInput['close'].values
ema = EMA(EMAclose, timeperiod=200)
##d. current price action
priceaction = df.head(1)
pricehigh = priceaction['high'].values[0]
pricelow = priceaction['low'].values[0]
priceclose = priceaction['close'].values[0]
if pricehigh > upperband[-1]: breakBand = -1
elif pricelow < lowerband[-1]: breakBand = 1
else: breakBand = 0
if pricelow > ema[-1]: marketEMA = 1
elif pricehigh < ema[-1]: marketEMA = -1
else: marketEMA = 0
##OUTPUT
if marketEMA == 1 and pattern > 0 and patterntype == -1 and breakBand == 1:
position = 1
elif marketEMA == -1 and pattern < 0 and patterntype == -1 and breakBand == -1:
position = -1
else:
position = 0
amount = 50
if position == 1:
stoploss = priceclose - 1.05 * atr
takeprofit = priceclose + 1.45 * atr
# amount = priceclose / (priceclose - stoploss)
if (stoploss - priceclose) * amount < 0.1:
position = 0
amount = 0
stoploss = 0
takeprofit = 0
elif position == -1:
stoploss = priceclose + 1.05 * atr
takeprofit = priceclose - 1.45 * atr
# amount = priceclose / (stoploss - priceclose)
if (stoploss - priceclose) * amount < 0.1:
position = 0
amount = 0
stoploss = 0
takeprofit = 0
else:
stoploss = 0
takeprofit = 0
amount = 0
if position * pattern > 0:
confidence = abs(pattern)
elif position * pattern < 0:
confidence = abs(1 / pattern)
elif position == 0:
confidence = 0
else:
confidence = 1
# ##FOR TEST
# position = 1
return [position, amount, priceclose, stoploss, takeprofit, confidence]
def bbp(data):
price = data.dropna()
close = price['Close'].values
up, mid, low = BBANDS(close, timeperiod=20, nbdevup=2, nbdevdn=2, matype=0)
bbp = (price['Close'] - low) / (up - low)
return bbp
import pandas as pd
pd.core.common.is_list_like = pd.api.types.is_list_like
import pandas_datareader.data as web
import matplotlib.pyplot as plt
from talib import RSI, BBANDS
import datetime as dt
import numpy as np
start = '2015-04-22'
end = '2017-04-22'
symbol = 'MCD'
max_holding = 100
price = web.DataReader(name=symbol, data_source='quandl', start=start, end=end)
print(price.head())
price = price.iloc[::-1]
print(price.head())
price = price.dropna()
close = price['AdjClose'].values
up, mid, low = BBANDS(close, timeperiod=20, nbdevup=2, nbdevdn=2, matype=0)
rsi = RSI(close, timeperiod=14)
print("RSI (first 10 elements)\n", rsi[14:24])
def bbands(close,
length=None,
std=None,
mamode=None,
ddof=0,
offset=None,
**kwargs):
"""Indicator: Bollinger Bands (BBANDS)"""
# Validate arguments
length = int(length) if length and length > 0 else 5
std = float(std) if std and std > 0 else 2.0
mamode = mamode if isinstance(mamode, str) else "sma"
ddof = int(ddof) if ddof >= 0 and ddof < length else 1
close = verify_series(close, length)
offset = get_offset(offset)
if close is None: return
# Calculate Result
if Imports["talib"]:
from talib import BBANDS
upper, mid, lower = BBANDS(close, length)
else:
standard_deviation = stdev(close=close, length=length, ddof=ddof)
deviations = std * standard_deviation
# deviations = std * standard_deviation.loc[standard_deviation.first_valid_index():,]
mid = ma(mamode, close, length=length, **kwargs)
lower = mid - deviations
upper = mid + deviations
ulr = non_zero_range(upper, lower)
bandwidth = 100 * ulr / mid
percent = non_zero_range(close, lower) / ulr
# Offset
if offset != 0:
lower = lower.shift(offset)
mid = mid.shift(offset)
upper = upper.shift(offset)
bandwidth = bandwidth.shift(offset)
percent = bandwidth.shift(offset)
# Handle fills
if "fillna" in kwargs:
lower.fillna(kwargs["fillna"], inplace=True)
mid.fillna(kwargs["fillna"], inplace=True)
upper.fillna(kwargs["fillna"], inplace=True)
bandwidth.fillna(kwargs["fillna"], inplace=True)
percent.fillna(kwargs["fillna"], inplace=True)
if "fill_method" in kwargs:
lower.fillna(method=kwargs["fill_method"], inplace=True)
mid.fillna(method=kwargs["fill_method"], inplace=True)
upper.fillna(method=kwargs["fill_method"], inplace=True)
bandwidth.fillna(method=kwargs["fill_method"], inplace=True)
percent.fillna(method=kwargs["fill_method"], inplace=True)
# Name and Categorize it
lower.name = f"BBL_{length}_{std}"
mid.name = f"BBM_{length}_{std}"
upper.name = f"BBU_{length}_{std}"
bandwidth.name = f"BBB_{length}_{std}"
percent.name = f"BBP_{length}_{std}"
upper.category = lower.category = "volatility"
mid.category = bandwidth.category = upper.category
# Prepare DataFrame to return
data = {
lower.name: lower,
mid.name: mid,
upper.name: upper,
bandwidth.name: bandwidth,
percent.name: percent
}
bbandsdf = DataFrame(data)
bbandsdf.name = f"BBANDS_{length}_{std}"
bbandsdf.category = mid.category
return bbandsdf
def runModel(factors, paras):
import pandas as pd
from talib import MACD, SMA, LINEARREG_SLOPE, RSI, BBANDS, KAMA
import numpy as np
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers import Dense, Activation, LSTM
from keras.optimizers import SGD, RMSprop, Adagrad, Adam, Adamax, Nadam
from keras.callbacks import EarlyStopping
close = data['adjclose']
date = list(data['date'])
ret = close.diff() / close.shift(1)
#2: rise, 1:stay, 0:drop
ret_dir = 1 * (ret > ret.quantile(0.66)) + 1 * (ret > ret.quantile(0.33))
#ret_dir=ret>ret.quantile(0.6)
MA_diff1 = SMA(close, timeperiod=5) - close
MA_diff2 = SMA(close, timeperiod=5) - SMA(close, timeperiod=15)
macd, macdsignal, macdhist = MACD(close,
fastperiod=12,
slowperiod=26,
signalperiod=9)
SLOPE5 = LINEARREG_SLOPE(close, timeperiod=5)
SLOPE15 = LINEARREG_SLOPE(close, timeperiod=15)
SLOPE_diff = LINEARREG_SLOPE(close, timeperiod=5) - LINEARREG_SLOPE(
close, timeperiod=10)
SKEW = ret.rolling(window=10, center=False).skew()
RSI20 = RSI(close, timeperiod=20)
RSI_SIG = 1 * (RSI20 < 40) - (RSI20 > 60)
upper, middle, lower = BBANDS(close, timeperiod=15)
BOLL_SIG = 1 * (close < lower) - (close > upper)
MACD_RATIO = macd / SMA(macd, timeperiod=5)
Kalman = KAMA(close, timeperiod=5)
STD = close.rolling(window=10, center=False).std()
KALMAN_SIG = 1 * (close < Kalman - 0.1 * STD) - (close >
Kalman + 0.1 * STD)
train_ratio = paras[0]
trans_fee = paras[1]
stop_loss = paras[2]
take_profit = paras[3]
originalDataSet = pd.DataFrame({
'dir': ret_dir.shift(-1),
0: MA_diff1,
1: MA_diff2,
2: macd,
3: SLOPE5,
4: SLOPE15,
5: SLOPE_diff,
6: SKEW,
7: RSI20,
8: RSI_SIG,
9: BOLL_SIG,
10: KALMAN_SIG,
11: MACD_RATIO
})
originalDataSet = originalDataSet.dropna(axis=0)
del_num = []
for i in range(len(factors)):
if factors[i] == 0:
del_num.append(i)
originalDataSet = originalDataSet.drop(labels=del_num, axis=1)
def normalize(train, flag=0):
if flag == 0:
train_norm = train.apply(lambda x: (x - np.mean(x)) /
(np.max(x) - np.min(x)))
else:
y = train.iloc[:, 0]
train_norm = train.apply(lambda x: (x - np.mean(x)) /
(np.max(x) - np.min(x)))
train_norm.iloc[:, 0] = y
return train_norm
win_size = 5
def windows(da, window_size):
start = 0
while start < len(da):
yield start, start + window_size
start += 1
# start += (window_size // 2)
def extract_segments(da, window_size=30):
segments = np.empty((0, (window_size) * (da.shape[1] - 1)))
labels = np.empty((0))
for (start, end) in windows(da, window_size):
if end >= len(originalDataSet):
break
if (len(da.iloc[start:end]) == (window_size)):
signal = np.array(da.iloc[start:end, 1:]).reshape(1, -1)[0]
segments = np.vstack([segments, signal])
labels = np.append(labels, da.iloc[end, 0])
return segments, labels
segments, labels = extract_segments(originalDataSet, win_size)
num_train = int(train_ratio * labels.shape[0])
num_test = labels.shape[0] - num_train
dummy_labels = np.asarray(pd.get_dummies(labels), dtype=np.int8)
reshaped_segments = segments.reshape(
[labels.shape[0], win_size, originalDataSet.shape[1] - 1])
train_x = reshaped_segments[:num_train]
train_y = dummy_labels[:num_train]
test_x = reshaped_segments[num_train:]
test_y = dummy_labels[num_train:]
batch_size = 32
model = Sequential()
model.add(
LSTM(units=50,
return_sequences=True,
input_shape=[train_x.shape[1], train_x.shape[2]],
dropout=0.2))
model.add(Activation('relu'))
model.add(
LSTM(units=50,
return_sequences=True,
input_shape=[train_x.shape[1], train_x.shape[2]],
dropout=0.2))
model.add(Activation('relu'))
model.add(LSTM(units=24, dropout=0.2, recurrent_dropout=0.2))
model.add(Activation('relu'))
model.add(Dense(3, activation='softmax'))
sgd = SGD(lr=0.002, decay=1e-6, momentum=0.9, nesterov=True)
rmsprop = RMSprop(lr=0.0002, rho=0.9, epsilon=1e-06)
adagrad = Adagrad(lr=0.001, epsilon=1e-06)
adam = Adam(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
adamax = Adamax(lr=0.002, beta_1=0.9, beta_2=0.999, epsilon=1e-08)
nadam = Nadam(lr=0.0005,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
schedule_decay=0.004)
model.compile(loss='categorical_crossentropy',
optimizer=sgd,
metrics=['accuracy'])
model.fit(train_x,
train_y,
epochs=50,
batch_size=batch_size,
validation_data=(test_x, test_y),
verbose=2,
shuffle=False,
callbacks=[EarlyStopping(patience=40)])
initial_captial = 10000000
pred_y = model.predict(test_x)
norm_y = np.diff(np.log(pred_y), axis=0)
norm_y = np.array(norm_y, dtype=float)
price = list(close[len(close) - 1 - len(pred_y):len(close) - 1])
signal = [0 for i in range(len(pred_y))]
buySell = [0 for i in range(len(pred_y))]
buySell_size = [0 for i in range(len(pred_y))]
position = [0 for i in range(len(pred_y))]
Notional = [0 for i in range(len(pred_y))]
CumNotional = [0 for i in range(len(pred_y))]
PnL = [0 for i in range(len(pred_y))]
cost = 1
station = 0
for i in range(1, len(pred_y)):
temp = list(norm_y[i - 1])
signal[i] = temp.index(max(temp)) - 1
if i == 0:
buySell[i] = signal[i]
buySell_size[i] = (initial_captial +
PnL[-1]) // price[i] * buySell[i]
position[i] = buySell_size[i]
Notional[i] = -buySell_size[i] * price[i]
CumNotional[i] = Notional[i]
PnL[i] = CumNotional[i] + position[i] * price[i]
else:
station = np.sign(position[i - 1])
if signal[i] != station:
if station == 1:
if signal[i] == 0:
buySell_size[i] = -position[i - 1]
if signal[i] == -1:
buySell_size[i] = -position[i - 1] - (
initial_captial + PnL[-1]) // price[i]
elif station == 0:
if signal[i] == 1:
buySell_size[i] = (initial_captial +
PnL[-1]) // price[i]
if signal[i] == -1:
buySell_size[i] = -(initial_captial +
PnL[-1]) // price[i]
else:
if signal[i] == 0:
buySell_size[i] = -position[i - 1]
if signal[i] == 1:
buySell_size[i] = -position[i - 1] + (
initial_captial + PnL[-1]) // price[i]
#stop loss and take profit
if i != 0 and buySell_size[i] == 0 and position[i - 1] != 0:
if position[i - 1] > 0:
if price[i] / cost > 1 + take_profit or price[
i] / cost < 1 - stop_loss:
buySell_size[i] = -position[i - 1]
else:
if price[i] / cost < 1 - take_profit or price[
i] / cost > 1 + stop_loss:
buySell_size[i] = -position[i - 1]
position[i] = position[i - 1] + buySell_size[i]
Notional[i] = -buySell_size[i] * price[i]
CumNotional[i] = CumNotional[i - 1] + Notional[i]
PnL[i] = CumNotional[i] + position[i] * price[i]
if buySell_size[i] != 0 and position[i] != 0:
cost = price[i]
oneMinPnL = np.diff(PnL)
oneMinPnL[0] = 0
from math import sqrt
Sharpe = oneMinPnL.mean() / oneMinPnL.std() * sqrt(252)
pnl = list(PnL)
price = [round(x, 2) for x in price]
pnl = [round(x, 2) for x in pnl]
date = date[len(close) - 1 - len(pred_y):len(close) - 1]
srp = round(Sharpe, 2)
totalrtn = round(PnL[-1] / initial_captial, 2)
return date, pnl, price, srp, totalrtn
def PlotStockChart(shareCode, showChart=False):
RSI_T = 10
MACD_FAST = 10
MACD_SLOW = 21
MACD_SIGNAL = 5
BBAND_PERIOD = 21
data = pd.read_json('D:\\Personal\\share\\data\\' + shareCode + '.txt')
data['t'] = pd.to_datetime(data['t'], unit='s')
data['date'] = data['t'].dt.date
data.set_index(['date'], inplace=True, drop=False)
df = data
# First instance
trace1 = {
'x': df.index,
'open': df.o,
'close': df.c,
'high': df.h,
'low': df.l,
'type': 'candlestick',
'name': shareCode,
'showlegend': False
}
rsi = RSI(df.c, timeperiod=10)
macd, macdSignal, macdHist = MACD(df.c, MACD_FAST, MACD_SLOW, MACD_SIGNAL)
upper, middle, lower = BBANDS(df.c,
timeperiod=BBAND_PERIOD,
nbdevup=2,
nbdevdn=2)
# Second instance - avg_30
trace2 = {
'x': df.index,
'y': rsi,
'type': 'scatter',
'mode': 'lines',
'line': {
'width': 1,
'color': 'blue'
},
'name': 'RSI'
}
rsiLower = {
'x': df.index,
'y': 30 * (rsi / rsi),
'type': 'scatter',
'line': {
'width': 1,
'color': 'red'
},
'name': 'RSI Lower'
}
rsiUpper = {
'x': df.index,
'y': 70 * (rsi / rsi),
'type': 'scatter',
'line': {
'width': 1,
'color': 'red'
},
'name': 'RSI Upper'
}
# MACD
trace4 = {
'x': df.index,
'y': macd,
'type': 'scatter',
'mode': 'lines',
'line': {
'width': 1,
'color': 'blue'
},
'name': 'macd'
}
trace5 = {'x': df.index, 'y': macdHist, 'type': 'bar', 'name': 'macdHist'}
trace6 = {
'x': df.index,
'y': macdSignal,
'type': 'scatter',
'mode': 'lines',
'line': {
'width': 1,
'color': 'red'
},
'name': 'macdSignal'
}
bbUpper = {
'x': df.index,
'y': upper,
'type': 'scatter',
'mode': 'lines',
'line': {
'width': 1,
'color': 'blue'
},
'name': 'BB Upper'
}
bbLower = {
'x': df.index,
'y': lower,
'type': 'scatter',
'mode': 'lines',
'line': {
'width': 1,
'color': 'blue'
},
'name': 'BB Lower'
}
bbMiddle = {
'x': df.index,
'y': middle,
'type': 'scatter',
'mode': 'lines',
'line': {
'width': 1,
'color': 'orange'
},
'name': 'BB Middle'
}
# Aggregate all instances and define 'data' variable
xaxisType = {'type': "category", 'categoryorder': 'category ascending'}
layout = dict(title=shareCode,
xaxis=xaxisType,
xaxis3=xaxisType,
xaxis4=xaxisType)
fig = make_subplots(rows=4,
cols=1,
shared_xaxes=True,
vertical_spacing=0.02)
fig.append_trace(trace1, row=1, col=1)
fig.append_trace(bbLower, row=1, col=1)
fig.append_trace(bbMiddle, row=1, col=1)
fig.append_trace(bbUpper, row=1, col=1)
fig.append_trace(trace2, row=4, col=1)
fig.append_trace(rsiLower, row=4, col=1)
fig.append_trace(rsiUpper, row=4, col=1)
fig.append_trace(trace4, row=3, col=1)
fig.append_trace(trace5, row=3, col=1)
fig.append_trace(trace6, row=3, col=1)
fig.update_layout(layout)
def zoom(layout, xrange):
in_view = df.loc[fig.layout.xaxis.range[0]:fig.layout.xaxis.range[1]]
fig.layout.yaxis.range = [
in_view.High.min() - 10,
in_view.High.max() + 10
]
#fig.layout.on_change(zoom, 'xaxis.range')
fig.layout.on_change(
lambda obj, xrange, yrange: print("%s-%s" % (xrange, yrange)),
('xaxis', 'range'), ('yaxis', 'range'))
if showChart:
fig.show()
else:
fig.write_html('D:\\Personal\\share\\results\\charts\\' + shareCode +
'.html',
include_plotlyjs='cdn')
return
#PlotStockChart('JKH.N0000', True)
def calculate_signals(self):
for symbol in self.symbol_list:
print('Searching for {1} signals ...'.format(symbol))
last_price = self.data_handler.current_price(symbol)[0]
h1_highs = self.data_handler.get_latest_bars_values(
symbol, 'high', timeframe='H1', N=self.bars_window)
h1_lows = self.data_handler.get_latest_bars_values(
symbol, 'low', timeframe='H1', N=self.bars_window)
h1_closes = self.data_handler.get_latest_bars_values(
symbol, 'close', timeframe='H1', N=self.bars_window)
h1_hlc3 = (h1_highs + h1_lows + h1_closes) / 3
h1_ema_50 = EMA(h1_hlc3, timeperiod=50)
h1_ema_100 = EMA(h1_hlc3, timeperiod=100)
h1_ema_200 = EMA(h1_hlc3, timeperiod=200)
m15_highs = self.data_handler.get_latest_bars_values(
symbol, 'high', timeframe='m15', N=self.bars_window)
m15_lows = self.data_handler.get_latest_bars_values(
symbol, 'low', timeframe='m15', N=self.bars_window)
m15_closes = self.data_handler.get_latest_bars_values(
symbol, 'close', timeframe='m15', N=self.bars_window)
m15_hlc3 = (m15_highs + m15_lows + m15_closes) / 3
m15_ema_50 = EMA(m15_hlc3, timeperiod=50)
m15_ema_100 = EMA(m15_hlc3, timeperiod=100)
m15_ema_200 = EMA(m15_hlc3, timeperiod=200)
m15_price_bb_up, m15_price_bb_mid, m15_price_bb_low = BBANDS(
m15_hlc3, timeperiod=20, nbdevup=2, nbdevdn=2)
m15_price_bb_low_dist = (m15_price_bb_mid + m15_price_bb_low) / 2
m15_price_bb_up_dist = (m15_price_bb_up + m15_price_bb_mid) / 2
m15_rsi = RSI(m15_closes, timeperiod=14)
m15_rsi_bb_up, m15_rsi_bb_mid, m15_rsi_bb_low = BBANDS(
m15_rsi, timeperiod=20, nbdevup=2, nbdevdn=2)
m15_stochrsi = STOCHRSI(pd.Series(m15_closes), timeperiod=14)
m15_atr = ATR(m15_highs, m15_lows, m15_closes, timeperiod=14)
datetime = self.data_handler.get_latest_bar_datetime(symbol)[0]
direction = None
status = None
position = 'Not yet'
print(datetime)
if last_price > h1_ema_50[-1] and last_price > h1_ema_100[
-1] and last_price > h1_ema_200[-1]:
print('Uprend detected')
if last_price > m15_ema_50[-1] and last_price > m15_ema_100[
-1] and last_price > m15_ema_200[-1]:
print('Uprend confirmed')
direction = 'LONG'
if m15_rsi[-1] < m15_rsi_bb_low[-1]:
print('Reversal Detected')
status = 'Reversal detected'
if last_price < m15_price_bb_low_dist[-1]:
print('Reversal confirmed')
status = 'Reversal confirmed'
if m15_stochrsi[-1] < 25:
print('Go LONG')
position = 'Go ahead'
elif last_price < h1_ema_50[-1] and last_price < h1_ema_100[
-1] and last_price < h1_ema_200[-1]:
print('Downrend detected')
if last_price < m15_ema_50[-1] and last_price < m15_ema_100[
-1] and last_price < m15_ema_200[-1]:
print('Downrend confirmed')
direction = 'SHORT'
if m15_rsi[-1] > m15_rsi_bb_up[-1]:
print('Reversal Detected')
status = 'Reversal detected'
if last_price > m15_price_bb_up_dist[-1]:
print('Reversal confirmed')
status = 'Reversal confirmed'
if m15_stochrsi[-1] > 75:
print('Go SHORT')
position = 'Go ahead'
else:
pass
print(symbol, direction, status, position, datetime)
if status is not None:
self.telegram.send_text({
'symbol':
symbol,
'direction':
direction,
'status':
status,
'position':
position,
'atr':
np.around(m15_atr[-1], decimals=5),
'datetime':
datetime
})
