def test_release_collateral(self):
call = Option('call', TSLA, 500, datetime(2020, 12, 28))
port = Portfolio(50000, {TSLA: 100})
te = TradingEngine([port])
# sell a contract to start, putting up 100 shares as collateral
te.sell_contract(port, call, 10)
self.assertEqual(-1, port.securities[call])
self.assertEqual(100, port.collateral[TSLA])
# buy back the contract, eliminating the obligation
te.buy_contract(port, call, 10)
self.assertEqual(0, port.securities[call])
self.assertEqual(0, port.collateral[TSLA])
# go long instead, purchasing a call
te.buy_contract(port, call, 10)
self.assertEqual(1, port.securities[call])
self.assertEqual(0, port.collateral[TSLA])
# close that long call
te.sell_contract(port, call, 10)
self.assertEqual(0, port.securities[call])
self.assertEqual(0, port.collateral[TSLA])
# sell an option to go short again
te.sell_contract(port, call, 10)
self.assertEqual(-1, port.securities[call])
self.assertEqual(100, port.collateral[TSLA])
def test_buy_contract(self):
call = Option('call', TSLA, 500, datetime(2020, 12, 28))
port = Portfolio(1000, {})
te = TradingEngine([port])
te.buy_contract(port, call, 10)
self.assertEqual(0, port.cash)
self.assertEqual(1, port.securities[call])
def test_exercise_put(self):
put = Option('put', TSLA, 500, datetime(2020, 12, 28))
port = Portfolio(2000, {TSLA: 100})
te = TradingEngine([port])
te.buy_contract(port, put, 10)
self.assertEqual(2000 - 1000, port.cash)
self.assertEqual(1, port.contracts()[put])
te.eval({TSLA: 400}, datetime(2020, 12, 28))
self.assertEqual(2000 - 1000 + 50000, port.cash)
self.assertEqual(0, port.securities.get(TSLA, 0))
self.assertEqual(0, port.securities[put])
def test_exercise_call(self):
call = Option('call', TSLA, 500, datetime(2020, 12, 28))
port = Portfolio(51000, {})
te = TradingEngine([port])
te.buy_contract(port, call, 10)
self.assertEqual(51000 - 1000, port.cash)
self.assertEqual(1, port.contracts()[call])
te.eval({TSLA: 600}, datetime(2020, 12, 28))
self.assertEqual(0, port.cash)
self.assertEqual(100, port.securities[TSLA])
self.assertEqual(0, port.securities[call])
class Simulator:
def __init__(self, security, portfolio, path, training, reporter=None):
self.security = security
self.portfolio = portfolio
self.path = path
self.training = training
self.reporter = reporter
self.scales = pd.read_csv(path / 'scales.csv', index_col=0)
self.engine = TradingEngine([portfolio], reporter)
self.split_handler = StockSplitHandler(path / 'splits.csv', security)
self.importer = CsvImporter()
def simulate(self, net, start=None, end=None):
"""
runs a simulation with provided network
net: neural network
start: datetime to start simulation
end: datetime to end the simulation
"""
close = None
for row in self._days_in_range(start, end):
params = self._map_row(row)
date = params[0]
close = params[1]
params = params[1::] # shave off date
try:
# process assignments, expirations
self.engine.eval({self.security: self._denormalize(close)},
date)
if not np.isnan(params).any():
# Check for stock split and adjust portfolio accordingly
self.split_handler.check_and_invoke(self.portfolio, date)
# Activate! 🤖
buy, sell, hold, delta, theta = net.activate(params)
# Buy
if buy > sell and buy > hold:
self._buy(date)
# Sell
elif sell > buy and sell > hold:
self._sell(date, close, delta, theta)
except Exception as e:
print(f"Failed on {self.security}:{date}")
raise e
return self._calculate_fitness(close, end)
def _most_recent_chain(self, date):
while True:
path = self.path / 'chains' / f"{small_date(date)}.csv"
if path.exists():
return self.importer.parse_chain(date, self.security, path)
else:
date -= timedelta(days=1)
def _calculate_fitness(self, close, end):
cash = 0
for contract, amt in self.portfolio.contracts().items():
if amt != 0:
chain = self._most_recent_chain(end)
# option prices are not normalized
cash += chain.get_price(contract) * amt * 100
denorm_close = self._denormalize(close)
for stock, amt in self.portfolio.stocks().items():
cash += amt * denorm_close
cash += self.portfolio.cash
# compare against a buy-and-hold strategy
return cash - (100 * denorm_close)
def _days_in_range(self, start, end):
mask = (self.training['date'] > start) & (self.training['date'] <= end)
for idx, row in self.training.loc[mask].iterrows():
yield row
def _map_row(self, row):
# TODO add held option delta/theta, maybe vega?
date = row['date']
close = row['close']
macd = row['macd']
macd_signal = row['macd_signal']
macd_diff = row['macd_diff']
bb_bbm = row['bb_bbm']
bb_bbh = row['bb_bbh']
bb_bbl = row['bb_bbl']
rsi = row['rsi']
return (date, close, macd, macd_signal, macd_diff, bb_bbm, bb_bbh,
bb_bbl, rsi)
def _buy(self, date):
# only covered calls are supported right now so this means close the position
for contract, amt in self.portfolio.contracts().items():
if amt < 0:
# right now there should only be one short contract at a time
chain = self._most_recent_chain(date)
new_price = chain.get_price(contract)
try:
self.engine.buy_contract(self.portfolio, contract,
new_price)
if self.reporter:
self.reporter.record(date, 'buy', contract, new_price)
except Exception as e:
print(e)
def _sell(self, date, close, delta, theta):
if not self.portfolio.contracts():
chain = self._most_recent_chain(date)
# sell a new contract
contract = chain.search(self._denormalize(close),
delta=delta,
theta=theta)
try:
self.engine.sell_contract(self.portfolio, contract,
contract.price)
if self.reporter:
self.reporter.record(date, 'sell', contract,
contract.price)
except Exception as e:
print(e)
def _denormalize(self, x):
mn = self.scales.loc['close', 'min']
mx = self.scales.loc['close', 'max']
return un_min_max(x, mn, mx)