def __init__(self, param_dict):
super(ValParamMovingAverage, self).__init__(param_dict)
self.type = 'moving_average'
if 'window_size' in param_dict.keys():
self.window_size = param_dict['window_size']
else:
error('window_size needs to be given when method = moving_average')
if 'tolerance_up' in param_dict.keys():
self.tolerance_up = param_dict['tolerance_up']
else:
self.tolerance_up = 100
if 'tolerance_down' in param_dict.keys():
self.tolerance_down = param_dict['tolerance_down']
else:
self.tolerance_down = 1
if 'calculation_period' in param_dict.keys():
self.calculation_period = param_dict['calculation_period']
else:
self.calculation_period = 2
if 'damping_factor' in param_dict.keys():
self.damping_factor = param_dict['damping_factor']
else:
self.damping_factor = 1
def get_close_moving_average(self, window_size, time=None, damping_factor=None):
if time is None:
time_end = self.data.index[-1]
time_start = self.data.index[0]
else:
time_end = time
time_start = time - timedelta(days=window_size)
df = self.get_price_close()
if damping_factor is None:
ma_series = df.truncate(before=time_start, after=time_end)
ma_series = ma_series.rolling(window_size).mean()
ma_series = ma_series.dropna()
else:
series = df.truncate(before=time_start, after=time_end)
series_size = len(series)
if series_size < window_size:
error('The input series is shorter than the window size.')
ma_series = pd.Series()
for i in range(series_size - window_size):
avg = 0.0
for j in range(window_size):
scalar = damping_factor ** j
avg = avg + scalar * series.iloc[window_size + i - j - 1]
avg = avg / window_size
ma_series.at[window_size + i - 1] = avg
return ma_series
def signal_average_envelope(env, asset, time=None, val_param=None):
if val_param is None:
val_param = {
'window_size': [5],
'tolerance_up': 0.025,
'tolerance_down': 0.025,
'damping_factor': 1.0
}
ma_param = ValParamMovingAverage(val_param)
if len(ma_param.window_size) != 1:
error('The window size list needs to be of size 1')
window_size = ma_param.window_size[0]
section = env.get_section(asset)
if time is None:
time = section.data.index[-1]
if section.data.size < window_size + 2:
error('Cannot calculate the moving average. The series is too short.')
factor = ma_param.damping_factor
price = section.get_price_close(time)
ma = section.get_close_moving_average(window_size, time, factor)
if price > (1 + ma_param.tolerance_up) * ma.iloc[-1]:
return -1
elif price < (1 - ma_param.tolerance_down) * ma.iloc[-1]:
return 1
else:
return 0
def plot_close_moving_average(self, window_sizes, damping_factor=None):
if len(window_sizes) == 0:
error('List of window sizes needs to be provided.')
for size in window_sizes:
series = self.get_close_moving_average(size, None, damping_factor)
plt.plot(series.index, series.values)
plt.show()
def value(self, asset_prices):
total = self.cash + self.wife_pocket
for asset in self.holdings.keys():
if asset not in asset_prices.keys():
error('The price of ' + asset +
'is not given in the input dict.')
total = total + self.holdings[asset] * asset_prices[asset]
return total
def get_price_close(self, time=None):
if 'price_close' in self.data.columns:
price_close = 'price_close'
elif self.target + '_price_close' in self.data.columns:
price_close = self.target + '_price_close'
else:
error('Price close is not in the env section data.')
if time is None:
return self.data[price_close]
series = self.data[price_close]
series_trunc = series.truncate(after=time)
return series_trunc.iloc[-1]
def sim_price_close(self, num_regimes=1, time_start=None, time_end=None):
time_spot = self.data.index[-1]
sim_price_close = pd.Series()
sim_price_close.index.name = 'time_close'
sim_price_close.rename('price_close_sim')
sim_price_close[time_spot] = self.get_price_close(time_spot)
if num_regimes <= 0:
error('num_regimes needs to be positive integer.')
if time_start is None:
time_start = self.data.index[0]
if time_end is None:
time_end = self.data.index[-1]
time_start = self._round_time(time_start)
time_end = self._round_time(time_end)
inter_length = (time_end - time_start) / num_regimes
for i in range(num_regimes):
hist_start = time_start + i * inter_length
hist_end = hist_start + inter_length
hist_start = self._round_time(hist_start)
hist_end = self._round_time(hist_end)
statistics = self.stat(hist_start, hist_end)
avg = statistics['period_log_return']['mean']
stdev = statistics['period_log_return']['std']
count = int(statistics['period_log_return']['count'])
# TODO: use ARIMA for sampling
# Here we use p(t_i+1) = p(t_i) * exp( (avg+stdev*N(0,1)) * (t_i+1-t_i) )
norm_vec = np.random.normal(avg, stdev, count - 1)
for j in range(count - 1):
# TODO: only handle hourly or daily data.
if self.data_info['period_id'] == '1HRS':
sim_time = sim_price_close.index[-1] + timedelta(hours=1)
else:
sim_time = sim_price_close.index[-1] + timedelta(days=1)
sim_time = self._round_time(sim_time)
sim_price_close[sim_time] = sim_price_close.values[
-1] * np.exp(norm_vec[j])
return sim_price_close
def signal_double_dip(env, asset, time=None, val_param=None):
if val_param is None:
val_param = {'window_size': [3, 20]}
if len(val_param['window_size']) != 2:
error(
'For double dip, two window sizes need to be provided in the rule.'
)
dip_param = ValParamMovingAverage(val_param)
window_ma = dip_param.window_size[0]
window_dip = dip_param.window_size[1]
section = env.get_section(asset)
series = section.get_price_close()
if time is None:
time = series.index[-1]
if series.index[0] + timedelta(days=window_ma + window_dip) > time:
error(
'The given time is early than the series start date plus window.')
series_trunc = series.rolling(window=window_ma,
center=False).mean().dropna()
end_date = time
start_date = time - timedelta(days=window_dip)
series_trunc = series_trunc.truncate(before=start_date, after=end_date)
y = np.array(list(series_trunc.values))
x = np.arange(len(y))
# t = np.array(list(series_trunc.index))
z = np.polyfit(x, y, 4)
a = 4 * z.item(0)
b = 3 * z.item(1)
c = 2 * z.item(2)
d = z.item(3)
ind_roots = -27 * a**2 * d**2 + 18 * a * b * c * d - 4 * a * c**3 - 4 * b**3 * d + b**2 * c**2
if z.item(0) > 0:
if ind_roots > 0:
return 1
else:
if ind_roots > 0:
return -1
return 0
def signal_consecutive_moves(env, asset, time=None, val_param=None):
if val_param is None:
val_param = {
'method': 'moving_average',
'window_size': [10],
'tolerance_up': 0.03,
'tolerance_down': 0.03
}
if val_param['method'] == 'moving_average':
ma_param = ValParamMovingAverage(val_param)
section = env.get_section(asset)
window_size = ma_param.window_size[0]
if time is None:
time = section.data.index[-1]
if section.data.size < window_size + 2:
error(
'Cannot calculate the moving average. The series is too short.'
)
factor = ma_param.damping_factor
is_bull = True
is_bear = True
average = []
for i in range(ma_param.calculation_period):
time_end = time - timedelta(days=i)
average.append(
section.get_close_moving_average(window_size, time_end,
factor).iloc[-1])
for i in range(len(average) - 1):
is_bull = is_bull and (
average[i] > (1 - ma_param.tolerance_up) * average[i + 1])
is_bear = is_bear and (
average[i] < (1 - ma_param.tolerance_down) * average[i + 1])
if is_bear:
return -1
elif is_bull:
return 1
else:
return 0
else:
error('Only moving average is implemented for bear market indicator.')
def __init__(self, section_list):
"""
create a collection of market environment based on a list of envr sections
:param section_list:
"""
model_dict = {}
if len(section_list) > 0 and 'period_id' in section_list[0].data_info:
freq = section_list[0].data_info['period_id']
for i in range(len(section_list)):
model_dict[section_list[i].target] = section_list[i]
if 'period_id' in section_list[i].data_info:
if section_list[0].data_info['period_id'] != freq:
error(
'All the sections in the list should have the data with same freq.'
)
self.model_dict = model_dict
def signal_long_short_crossing(env, asset, time=None, val_param=None):
if val_param is None:
val_param = {
'window_size': [5, 30],
'tolerance_up': 0.0,
'tolerance_down': 0.0
}
if val_param['window_size'][0] > val_param['window_size'][1]:
window_short = val_param['window_size'][1]
window_long = val_param['window_size'][0]
else:
window_short = val_param['window_size'][0]
window_long = val_param['window_size'][1]
ma_param = ValParamMovingAverage(val_param)
section = env.get_section(asset)
if time is None:
time = section.data.index[-1]
if section.data.size < window_long + 2:
error('Cannot calculate the moving average. The series is too short.')
factor = ma_param.damping_factor
series_short = section.get_close_moving_average(window_short, time, factor)
series_long = section.get_close_moving_average(window_long, time, factor)
if series_short.iloc[-1] > (1 +
ma_param.tolerance_up) * series_long.iloc[-1]:
return -1
elif series_short.iloc[-1] < (
1 - ma_param.tolerance_down) * series_long.iloc[-1]:
return 1
else:
return 0
