def get_broker_realized_pnl(self, current_value, **kwargs):
"""
Args:
current_value: current portfolio value
Returns:
normalized realized profit/loss for last closed trade (is zero if no pos. closures within last env. step)
"""
if self.trade_just_closed:
pnl = decayed_result(
self.trade_result,
current_value,
self.env.broker.startingcash,
self.p.drawdown_call,
self.p.target_call,
gamma=1
)
# self.log.warning('get_broker_realized_pnl: got result: {} --> pnl: {}'.format(self.trade_result, pnl))
# Reset flag:
# self.trade_just_closed = False
# print('broker_realized_pnl: step {}, just closed.'.format(self.iteration))
else:
pnl = 0.0
return pnl
def update_sliding_stat(self):
"""
Updates all sliding statistics deques with latest-step values:
- normalized broker value
- normalized broker cash
- normalized exposure (position size)
- position duration in steps, normalized wrt. max possible episode steps
- exp. scaled episode duration in steps, normalized wrt. max possible episode steps
- normalized decayed realized profit/loss for last closed trade
(or zero if no trades been closed within last step);
- normalized profit/loss for current opened trade (unrealized p/l);
- normalized best possible up to present point unrealized result for current opened trade;
- normalized worst possible up to present point unrealized result for current opened trade;
- one hot encoding for actions received;
- rewards received (based on self.reward variable values);
"""
stat = self.sliding_stat
current_value = self.env.broker.get_value()
stat['broker_value'].append(
norm_value(
current_value,
self.env.broker.startingcash,
self.p.drawdown_call,
self.p.target_call,
))
stat['broker_cash'].append(
norm_value(
self.env.broker.get_cash(),
self.env.broker.startingcash,
99.0,
self.p.target_call,
))
stat['exposure'].append(
self.position.size /
(self.env.broker.startingcash * self.env.broker.get_leverage() +
1e-2))
stat['leverage'].append(
self.env.broker.get_leverage()) # TODO: Do we need this?
if self.trade_just_closed:
stat['realized_pnl'].append(
decayed_result(self.trade_result,
current_value,
self.env.broker.startingcash,
self.p.drawdown_call,
self.p.target_call,
gamma=1))
# Reset flag:
self.trade_just_closed = False
# print('POS_OBS: step {}, just closed.'.format(self.iteration))
else:
stat['realized_pnl'].append(0.0)
if self.position.size == 0:
self.current_pos_duration = 0
self.current_pos_min_value = current_value
self.current_pos_max_value = current_value
# print('ZERO_POSITION\n')
else:
self.current_pos_duration += 1
if self.current_pos_max_value < current_value:
self.current_pos_max_value = current_value
elif self.current_pos_min_value > current_value:
self.current_pos_min_value = current_value
stat['pos_duration'].append(
self.current_pos_duration /
(self.data.numrecords - self.inner_embedding))
stat['episode_step'].append(
exp_scale(self.iteration /
(self.data.numrecords - self.inner_embedding),
gamma=3))
stat['max_unrealized_pnl'].append(
(self.current_pos_max_value - self.realized_broker_value) *
self.broker_value_normalizer)
stat['min_unrealized_pnl'].append(
(self.current_pos_min_value - self.realized_broker_value) *
self.broker_value_normalizer)
stat['unrealized_pnl'].append(
(current_value - self.realized_broker_value) *
self.broker_value_normalizer)
stat['action'].append(self.action_norm(self.last_action))
stat['reward'].append(self.reward)