def moving_average_crossover_strategy(security,
window=5,
opp_direction=False,
out_sample_start=None,
return_directions=False):
df = select_security(security)
mva = df["Settle"].rolling(center=False, window=window).mean()
direction = []
position = 0
mva_below = False
mva_above = False
for i, value in enumerate(mva):
if value > df["Settle"][i] and mva_below:
position = 1
mva_below = False
elif value < df["Settle"][i] and mva_above:
position = -1
mva_above = False
elif value < df["Settle"][i]:
mva_below = True
elif value > df["Settle"][i]:
mva_above = True
direction.append(position)
direction = np.array(direction)
if return_directions:
return direction
if opp_direction:
direction = -direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)
def dmi_strategy(security,
window=14,
out_sample_start=None,
use_divergence=False,
return_directions=False):
"""
directional movement index
"""
df = select_security(security)
pnl = df['Settle'] - df['Settle'].shift(1)
upward = np.where((df['Settle'] - df['High'].shift(1)) > 0, 1, 0)
downward = np.where((df['Settle'] - df['Low'].shift(1)) < 0, 1, 0)
upward_sum = pd.Series(upward).rolling(window=window).sum()
downward_sum = pd.Series(downward).rolling(window=window).sum()
direction = None
if use_divergence:
divergence = upward_sum - downward_sum
direction = np.where(divergence > divergence.shift(1), 1, -1)
else:
direction = np.where(upward_sum >= downward_sum, 1, -1)
if return_directions:
return direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)
def bollinger_bands_strategy(security,
window=14,
out_sample_start=None,
upper=70,
lower=30,
return_directions=False):
"""
default RSI(14) is used
"""
df = select_security(security)
pnl = df['Settle'] - df['Settle'].shift(1)
profit = pnl.clip(lower=0, upper=None)
loss = pnl.clip(lower=None, upper=0)
avg_profit = profit.rolling(window=window).mean()
avg_loss = loss.rolling(window=window).mean()
rsi = 100 - 100 / (1 + avg_profit / abs(avg_loss))
direction = []
position = 0
for i in range(df.shape[0]):
if rsi[i] > upper:
position = 1
if rsi[i] < lower:
position = -1
direction.append(position)
direction = np.array(direction)
if return_directions:
return direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)
def trend_delay_strategy(security,
delay=1,
out_sample_start=None,
return_directions=False):
df = select_security(security)
direction = np.sign(df['Settle'] - df['Settle'].shift(delay))
direction = direction.fillna(0)
if return_directions:
return direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)
def macd_strategy(security,
long=26,
mid=12,
short=9,
use_divergence=False,
out_sample_start=None,
return_directions=False):
"""
default MACD(26, 12, 9) is used
"""
df = select_security(security)
ema_long = df['Settle'].ewm(span=long, adjust=False).mean()
ema_mid = df['Settle'].ewm(span=mid, adjust=False).mean()
macd = ema_mid - ema_long
macd_ema_short = macd.ewm(span=short, adjust=False).mean()
divergence = macd - macd_ema_short
direction = []
position = 0
if use_divergence:
div_prev = 0
for i in range(df.shape[0]):
if divergence[i] > div_prev:
position = 1
else:
position = -1
direction.append(position)
div_prev = divergence[i]
else:
for i in range(df.shape[0]):
if macd[i] > macd_ema_short[i]:
position = 1
else:
position = -1
direction.append(position)
direction = np.array(direction)
if return_directions:
return direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)
def slow_stochastic_strategy(security,
window=5,
lower=0.2,
upper=0.8,
out_sample_start=None,
return_directions=False):
df = select_security(security)
ss = slow_stochastic(close=df['Close'],
high=df['High'],
low=df['Low'],
window=window)
direction = []
position = 0
ss_below = False
ss_above = False
for value in ss:
if value > lower and ss_below:
position = 1
ss_below = False
elif value < upper and ss_above:
position = -1
ss_above = False
elif value < lower:
ss_below = True
elif value > upper:
ss_above = True
direction.append(position)
direction = np.array(direction)
if return_directions:
return direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)
def master_strategy(security, out_sample_start=None, return_directions=False):
df = select_security(security)
direction = None
if security == 'iron':
dir1 = slow_stochastic_strategy(security=security,
window=2,
lower=0.4,
upper=0.6,
out_sample_start=out_sample_start,
return_directions=True)
dir2 = moving_average_crossover_strategy(
security=security,
window=20,
opp_direction=True,
out_sample_start=out_sample_start,
return_directions=True)
dir3 = macd_strategy(security=security,
long=25,
mid=12,
short=2,
use_divergence=True,
out_sample_start=out_sample_start,
return_directions=True)
directions = np.stack((dir1, dir2, dir3), axis=1)
direction = stats.mode(directions, axis=1)[0]
direction = direction.reshape((direction.shape[0]))
direction = np.array(direction)
elif security == 'rebar':
dir1 = slow_stochastic_strategy(security=security,
window=2,
lower=0.3,
upper=0.7,
out_sample_start=out_sample_start,
return_directions=True)
dir2 = moving_average_crossover_strategy(
security=security,
window=5,
opp_direction=True,
out_sample_start=out_sample_start,
return_directions=True)
dir3 = macd_strategy(security=security,
long=26,
mid=12,
short=2,
use_divergence=True,
out_sample_start=out_sample_start,
return_directions=True)
directions = np.stack((dir1, dir2, dir3), axis=1)
direction = stats.mode(directions, axis=1)[0]
direction = direction.reshape((direction.shape[0]))
direction = np.array(direction)
elif security == 'hot coil':
dir1 = slow_stochastic_strategy(security=security,
window=7,
lower=0.4,
upper=0.6,
out_sample_start=out_sample_start,
return_directions=True)
dir2 = moving_average_crossover_strategy(
security=security,
window=2,
opp_direction=True,
out_sample_start=out_sample_start,
return_directions=True)
dir3 = macd_strategy(security=security,
long=26,
mid=12,
short=2,
use_divergence=True,
out_sample_start=out_sample_start,
return_directions=True)
directions = np.stack((dir1, dir2, dir3), axis=1)
direction = stats.mode(directions, axis=1)[0]
direction = direction.reshape((direction.shape[0]))
direction = np.array(direction)
if return_directions:
return direction
if out_sample_start is None:
show_pnl(direction, df["Settle"])
else:
os_data = df["Settle"][out_sample_start:]
os_direction = direction[-len(os_data):]
show_pnl(os_direction, os_data)